## wpiea2025218-source-pdf

## Source details

**Canonical URL:** [wpiea2025218-source-pdf](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025218-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025218-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025218-source-pdf.pdf.json)

---

### Introduction — research question, strategy, and implications
- Research question and scope:
  - Investigates how large earthquakes affect sovereign bond spreads in emerging markets and why effects vary across countries.
  - Monthly data for 96 emerging-market countries from January 2012 to November 2023.
  - Treats large earthquakes as quasi-random, exogenous shocks to test investor pricing of sovereign repayment capacity.
- Empirical strategy and identification:
  - Exploits quasi-random timing and geographic distinctness of earthquakes as natural experiments.
  - Employs heterogeneity-robust difference-in-differences estimators (Sun and Abraham 2021; de Chaisemartin and d’Haultfoeuille 2020) and event-study plots showing no pre-trends.
- Key findings (summary):
  - On average, earthquakes raise sovereign spreads.
  - Heterogeneous effects by institutional strength:
    - Low state capacity: spreads rise sharply and persist after an earthquake.
    - Robust institutions: same earthquakes have no effect or even reduce spreads.
  - Results robust to controls for macroeconomic fundamentals, global financial conditions, and alternative estimation techniques.
- Policy implications:
  - Investing in fiscal institutions reduces sovereign financing costs and improves shock absorption.
  - Sovereign risk models should incorporate institutional resilience as well as physical exposure.
  - Multilateral assistance and conditionality design should account for state capacity to withstand external shocks.

### Data — spreads, state capacity, earthquakes, and controls
- Sovereign Spread (panel construction and stats):
  - Panel: 96 emerging-market economies; monthly frequency January 2012 to November 2023.
  - Sovereign borrowing cost measure: Weighted-Average Sovereign Spread (WASS) from Bloomberg’s Back Office.
  - Inclusion criteria: Currency U.S. dollars or euros; law under foreign (UK or New York) or euro-area law; minimum size USD/EUR 250 million; maturity bullet ≥ 1 year remaining or amortizing ≥ 18 months.
  - Aggregation: par-value weighted, computed monthly; base curves capped at a fifteen-year maturity horizon.
  - Key statistics:
    - Average spread across the panel: 403 basis points.
    - Distributional tail: some crisis-episode observations exceed 14,000 basis points.
    - Dataset total data points: 11,799.
- State capacity and external debt:
  - State capacity metric: O’Reilly and Murphy (2022) index; annual coverage 1789 to 2018.
  - External debt: short-term external debt from World Development Indicators (WDI).
  - Use: SC̅ and DEBT̅ used to assess how institutional resilience and near-term refinancing needs moderate market reactions.
- Earthquake data and constructed indicators:
  - Source: National Earthquake Hazards Reduction Program (NEHRP).
  - Raw events: 1,667,000 events from January 2012 to November 2023; highest recorded Richter magnitude in raw data: 8.6.
  - Merged events with country locations (96 countries): 44,089 events.
  - Country-month indicators:
    - Binary: dummy indicator if an earthquake occurred in that month.
    - Intensity: average magnitude of earthquakes in that month.
  - Final temporal coverage aligned with spread data: January 2012 to November 2023.
- Additional controls and sample coverage:
  - Controls: GDP per capita (PPP, constant 2021 international dollars), Banking system crises dummy (Laeven and Valencia 2020), short-term external debt.
  - Sample after merging: complete country-month observations retained: 11,389 (approximately 97 percent of the theoretical maximum).
  - Missingness driven by gaps in short-term debt reporting among low-income countries.
  - Descriptive patterns:
    - Roughly 28 percent of country-months experience at least one seismic event.
    - Interquartile range for spreads: 64 to 410 basis points.
    - State capacity: mean and dispersion with majority clustering near global mean.
  - Robustness: results robust to excluding controls or imputing missing values with country-specific averages.

### Estimation strategy and identification details
- Benchmark DiD framework (monthly frequency):
  - Main equation: ∆Spread_{c,m} = α_i + τ Spread_{c,m−1} + β Earthquake_{c,m−s} + λ_m + δ_season + ε_{c,m}
  - Dependent variable: ∆spread_{c,m} = sovereign spread change for country c in month m.
  - Earthquake variables: binary indicator and continuous intensity (average Richter magnitude).
  - Lag parameter s captures delayed impact (s = 1 signifies one-month delay).
  - Fixed effects: country (α_i), month (λ_m ordered January 2012 =1 to November 2023 =140), seasonal (δ_season).
  - Estimator: high-dimensional fixed-effect estimator (Correia, 2017).
  - Coefficient of interest: β.
- Heterogeneity specifications:
  - Interaction with average state capacity (SC̅):
    - ∆Spread_{c,m} = α_i + τ Spread_{c,m−1} + β Earthquake_{c,m−s} + δ Earthquake_{c,m−s} × SC̅ + λ_m + δ_season + ε_{c,m}
  - Interaction with average short-term external debt (DEBT̅):
    - analogous specification with DEBT̅.
  - SC̅: averaged state capacity; DEBT̅: average short-term external debt over GDP.
- Unanticipated shock measurement:
  - EQ_{c,y,m} = γ EQ_{c,y,m−1} + λ_m + δ_season + α_i + ε_{c,y,m}
  - ε_{c,y,m} interpreted as unanticipated component; used in re-estimation.
- Heterogeneity-robust DiD and Local Projections DID (LP-DiD):
  - De Chaisemartin and d’Haultfoeuille approach: assess effects across six post-treatment periods and three placebo pre-treatment periods; focus on switchers.
  - LP-DiD (Dube et al. 2023): estimate β_h for h = 0,...,6; sample restriction to newly treated or clean control; use re-weighted (‘rw’) estimator.
  - Stata implementation reference: lpdid change_spread, time(cmonth) unit(ID) treat(treat) pre(3) post(6) rw.
- Inference and robustness:
  - Standard errors heteroskedasticity-robust clustered at country level.
  - Additional inference: clustering by region and Conley-type spatial autocorrelation corrections.
- Research question enabled:
  - Leverages exogeneity of earthquakes with institutional heterogeneity to test whether sovereign spreads react systematically and whether state capacity conditions that response.

### Main results — benchmark, dynamics, thresholds, and institutional heterogeneity
- Benchmark dynamic effects (timing and magnitude):
  - Coefficient on earthquake dummy two months after event (m−2): approximately 0.097 (9.7 basis points), significant at the 5 percent level.
  - Pre-event coefficients at m−1 and m−3: examples −0.022, −0.021 (small and statistically insignificant).
  - Intensity measure: each unit increase in average earthquake magnitude two months prior associated with approximately 0.024 (2.4 basis points), significant at the 5 percent level.
- Event-study (heterogeneity-robust DiD) point estimates (t+1 to t+6):
  - Effect t+1: 0.114 (SE 0.139; LB-CI −0.159; UB-CI 0.386; Observation 795.000; Switchers 33.000)
  - Effect t+2: 0.262 (SE 0.204; LB-CI −0.137; UB-CI 0.661; Observation 776.000; Switchers 33.000)
  - Effect t+3: 0.500 (SE 0.257; LB-CI −0.005; UB-CI 1.004; Observation 755.000; Switchers 32.000)
  - Effect t+4: 0.547 (SE 0.278; LB-CI 0.003; UB-CI 1.091; Observation 740.000; Switchers 32.000)
  - Effect t+5: 0.685 (SE 0.354; LB-CI −0.008; UB-CI 1.378; Observation 726.000; Switchers 32.000)
  - Effect t+6: 0.209 (SE 0.503; LB-CI −0.777; UB-CI 1.194; Observation 674.000; Switchers 31.000)
  - Placebo leads t−1 to t−3 jointly insignificant; p-value joint nullity = 0.384.
  - By months 3–5 after event, spreads rise by up to 0.685 (68.5 basis points) at month 5 with confidence intervals excluding zero in later months.
- Threshold heterogeneity by intensity percentiles:
  - Intensity effect at m−2 (continuous): 0.022 (SE 0.010), significant at **.
  - Threshold estimates (m−2) for percentiles:
    - 30th–80th percentiles: 0.089** (SE 0.044) at multiple thresholds.
    - 90th percentile: 0.109** (SE 0.048).
  - Interpretation: markets use intensity thresholds; top-decile events elicit larger spread responses.
- Role of State Capacity (interactions and split samples):
  - Baseline dummy m−2: 0.143** (SE 0.056) in Column (1).
  - Interaction: each standard deviation increase in state capacity reduces spread impact by −0.064 (−6.4 basis points) (SE 0.039).
  - Split-sample interactions (dummy m−2):
    - Dummy m−2 × bottom SC (p40): 0.177* (SE 0.090)
    - Dummy m−2 × top SC (p40): −0.214** (SE 0.091)
  - Intensity baseline m−2: 0.034*** (SE 0.013).
  - Intensity × SC̅: −0.015* (SE 0.009)
    - Intensity × bottom SC (p40): 0.041** (SE 0.021)
    - Intensity × top SC (p40): −0.049** (SE 0.022)
  - Split-sample summary (Annex III):
    - High state capacity (≥ p50): WA Spread_{m−1} = -0.036*** (0.004); (Shock) Dummy m−2 ~ 0.019 (0.031) (insignificant in many specs); Observations 6,145.
    - Low state capacity (< p50): WA Spread_{m−1} = -0.008*** (0.003); (Shock) Dummy m−2 = 0.180** (0.089) in some columns; Intensity_{m−2} = 0.041** (0.020); Observations 5,244.
  - Interpretation: in low-capacity countries earthquakes raise spreads; in high-capacity countries effects are muted or negative.
- Unanticipated shocks:
  - Using ε_{c,y,m} as unanticipated earthquake component, re-estimates confirm:
    - Unanticipated dummy m−2: 0.091** and 0.088** (SEs ~0.046–0.045).
    - Unanticipated dummy m−2 × State Capacity: −0.056 (SE 0.040).
    - Unanticipated dummy m−2 × bottom SC (p40): 0.165* (SE 0.092).
    - Unanticipated dummy m−2 × top SC (p40): −0.209** (SE 0.093).
    - Unanticipated intensity m−2: 0.033** (SE 0.013) to 0.039*** (SE 0.013).
  - Conclusion: isolating unanticipated events reinforces earthquake effects and the moderating role of state capacity.

### Robustness checks, triple interactions, and alternative specifications
- Controlling for banking system crises:
  - WA Spread lag coefficient across Tables 7–9: -0.010*** (0.002).
  - Banking system Crisis (m−3) coefficients:
    - Table 7 column (1): 2.791*** (0.411); similar estimates in Tables 8 and 9 (2.785*** (0.412); 2.824*** (0.412) etc.).
  - Interpretation: crises raise spreads by over 280 basis points, but controlling for them does not alter earthquake results or institutional interactions.
  - Sample sizes and fit (examples): Table 7 Observations 11,389; R-squared 0.015–0.081; months 137; countries 96.
- Triple interactions (Earthquake × Income Level × State Capacity / Debt):
  - Earthquake × Income × State Capacity (Table 10) selected coefficients:
    - Dummy_{m−2}: 1.512** (0.729) in column (1).
    - Dummy_{m−2} × IL̅ : -0.145* (0.077).
    - Dummy_{m−2} × SC̅ : -1.023* (0.579).
    - Dummy_{m−2} × SC̅ × IL̅ : 0.098* (0.058).
    - Intensity_{m−2} × SC̅ : -0.283** (0.134).
    - Intensity_{m−2} × SC̅ × IL̅ : 0.027** (0.013).
    - Sample: Observations 10,877; R-squared ~0.080; months 138; countries 89.
  - Earthquake × Income × DEBT̅ (Table 11) selected coefficients:
    - WA Spread_{m−1}: -0.013*** (0.003).
    - Dummy_{m−2} × DEBT̅ : -0.043*** (0.014) in column (3).
    - Intensity_{m−2} × DEBT̅ : -0.010*** (0.003) in column (3).
    - Sample: Observations 6,331; R-squared 0.112; months 138; countries 53.
  - Interpretation: state capacity effects are especially meaningful in countries below the 40th income percentile; higher external debt linked with reduced spread impacts, consistent with partial offset via international/concessional support.
- Alternative earthquake intensity measures (maximum magnitude):
  - Annex VII Panel A: Max Intensity m−2 estimates include 0.022** (0.010), 0.034*** (0.012), and 0.040*** (0.012) in various columns.
  - Max Intensity m−2 × SC̅ : -0.015* (0.009).
  - Panel B (unanticipated max intensity) yields similar coefficients and significance patterns.
  - Interpretation: replacing average monthly magnitude with maximum magnitude yields consistent baseline and conditioning results.
- Debt-to-GDP interactions (Annex VI):
  - WA Spread m−1: -0.013*** (0.003).
  - (Shock) Dummy m−2 × DEBT̅ : -0.042*** (0.014).
  - (Shock) Intensity m−2 × DEBT̅ : -0.010*** (0.003).
  - Bottom/top DEBT̅ interactions show sign reversals: bottom DEBT̅ (p40) positive; top DEBT̅ (p40) negative.
  - Observations: 6,278; R-squared 0.111–0.112; months 137; countries 53.
- Descriptive statistics (Annex II key entries):
  - Weighted Average Sovereign Spread (100 bps): Obs 11,695; Mean 4.034; SD 7.845; Min 0.003; Max 139.86.
  - Dummy Earthquake (NEHRP): Obs 11,799; Mean 0.281; SD 0.450; Min 0.000; Max 1.000.
  - Intensity Earthquake (monthly average Richter): Obs 11,799; Mean 1.172; SD 1.912; Min 0.000; Max 6.800.
  - Short-term External Debt (% GDP): Obs 6,537; Mean 8.172; SD 8.112; Min 0.000; Max 73.907.
  - State Capacity (z-score): Obs 9,135; Mean 0.933; SD 1.216; Min -2.152; Max 3.047.
  - GDP per capita (log, constant 2021 international US$, PPP): Obs 11,618; Mean 9.888; SD 0.906; Min 7.273; Max 11.835.
  - Banking Systematic Crisis (Dummy): Obs 11,799; Mean 0.0010; SD 0.032; Min 0.000; Max 1.000.
- Robustness conclusion:
  - Across specifications and samples, earthquakes affect sovereign spreads predominantly through institutional channels.
  - State capacity mitigates post-earthquake spread increases, particularly in low-income or low-capacity settings.
  - Higher external debt can be associated with reduced spread responses, consistent with partial offsetting via international/concessional support.
  - Repeated quantitative anchors: WA Spread lag coefficients around -0.010*** (0.002) and banking system crisis increases spreads by coefficients near 2.78–2.83 (*** p<0.01).

### Annex findings on identifying unanticipated shocks and split-sample estimates
- Annex IV (identifying unanticipated shocks):
  - Dummy m−1: 0.028*** (0.009)
  - Magnitude m−1: 0.031*** (0.009)
  - Observations: 11,703; R-squared: 0.615 (Dummy), 0.605 (Magnitude).
- Annex V (unanticipated shock — split SC sample) highlights:
  - PANEL A High SC (≥p50): WA Spread m−1 = -0.036*** (0.004); (Shock) Dummy m−2 ~ 0.019 (0.031); Observations 6,096.
  - PANEL B Low SC (<p50): WA Spread m−1 = -0.008*** (0.003); (Shock) Dummy m−2 = 0.167* (0.090); (Shock) Intensity m−2 = 0.039* (0.020); Observations 5,199.
- Annex VI (unanticipated shock × Debt-to-GDP):
  - WA Spread m−1: -0.013*** (0.003).
  - (Shock) Dummy m−2 × DEBT̅ : -0.042*** (0.014).
  - (Shock) Intensity m−2 × DEBT̅ : -0.010*** (0.003).
  - Observations: 6,278; R-squared: 0.111–0.112.
- Annex VII (maximum magnitude robustness):
  - Max Intensity m−2 baseline coefficients include 0.022** (0.010), 0.034*** (0.012), and 0.040*** (0.012) across columns.
  - Max Intensity m−2 × SC̅ : -0.015* (0.009).
  - Observations vary by column (e.g., 11,484; 11,389; 10,789), R-squared 0.076–0.081.

*Source: IMF Working Paper "Earthquakes and Emerging Market Sovereign Bond Spreads" — content unit wpiea2025218-source-pdf.*

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

### 1. Introduction

### Research question and scope
- Investigates how large earthquakes affect sovereign bond spreads in emerging markets and why effects vary across countries.
- Uses monthly data on sovereign bond spreads, seismic activity, and state capacity for 96 emerging-market countries from 2012 to 2023.
- Treats large earthquakes as quasi-random, exogenous shocks to test investor pricing of sovereign repayment capacity.

### Empirical strategy and identification
- Exploits the quasi-random timing and geographic distinctness of earthquakes as natural experiments.
- Employs heterogeneity-robust difference-in-differences estimators (Sun and Abraham 2021; de Chaisemartin and d’Haultfoeuille 2020).
- Uses event-study plots showing no pre-trends and divergent post-shock dynamics across groups.

### Key findings
- On average, earthquakes raise sovereign spreads.
- Heterogeneous effects by institutional strength:
  - In countries with low state capacity, spreads rise sharply and persist after an earthquake.
  - In countries with robust institutions, the same earthquakes have no effect or even reduce spreads.
- Results are robust to controls for macroeconomic fundamentals, global financial conditions, and alternative estimation techniques.
- Interpretation: investors condition the market response on institutional credibility; where tax systems are credible, bureaucracies competent, and relief swift, markets treat shocks as transitory and may anticipate reconstruction funding or multilateral support.

### Contributions to literature
- Adds causal, cross-country evidence on how sudden natural disasters affect sovereign finance in real time.
- Integrates rare, acute shocks into sovereign-risk determinants models and shows these shocks’ pricing is conditional on state capacity.
- Demonstrates that institutional credibility can invert, not just attenuate, market responses to disasters.

### Policy implications
- For countries: investing in fiscal institutions functions as risk management by reducing sovereign financing costs and improving shock absorption.
- For creditors: sovereign risk models should incorporate institutional resilience as well as physical exposure.
- For multilateral lenders: assistance and conditionality design should account for state capacity to withstand external shocks, not only outcomes.

*Source: IMF Working Paper — “Earthquakes and Emerging Market Sovereign Bond Spreads,” Introduction (pages 4–8).*

### 3. Data

### 3. Data

### A. Sovereign Spread
- Panel: 96 emerging-market economies, monthly frequency from January 2012 to November 2023.
- Sovereign borrowing cost measure: Weighted-Average Sovereign Spread (WASS) from Bloomberg’s Back Office.
- Inclusion criteria:
  - Currency: U.S. dollars or euros.
  - Law: issued under foreign (UK or New York) or euro-area law.
  - Minimum size: USD/EUR 250 million.
  - Maturity: bullet bonds ≥ 1 year remaining; amortizing ≥ 18 months.
- Spreads computed relative to U.S. Treasury or German Bund curve (currency-dependent); base curves smoothed using most liquid recent benchmarks and capped at a fifteen-year maturity horizon.
- Aggregation: par-value weighted, computed monthly, producing a balanced series for tradable debt instruments.
- Key statistics:
  - Average spread across the panel: 403 basis points.
  - Distributional tail: some crisis-episode observations exceed 14,000 basis points.
  - Dataset total data points: 11,799.

### B. The State Capacity and External Debt
- State capacity metric: O’Reilly and Murphy (2022) index.
  - Definition: government’s ability to collect revenue, maintain control over violence, deliver public goods, and enforce law.
  - Index assesses impartial and strict enforcement of laws, control over territory, efficiency of public administration, public fund allocation, and modernity/efficiency of revenue sources.
  - Annual data coverage: 1789 to 2018.
- External debt: short-term external debt data sourced from World Development Indicators (WDI).
- Use in analysis: state capacity and short-term external debt used to assess how institutional resilience and near-term refinancing needs moderate market reactions to earthquakes.

### C. Earthquake Disasters and Additional Controls
- Earthquake data:
  - Source: National Earthquake Hazards Reduction Program (NEHRP).
  - Raw events collected daily: 1,667,000 events from January 2012 to November 2023.
  - Highest recorded Richter magnitude in raw data: 8.6.
  - Events merged with country locations (96 countries): 44,089 events.
- Indicators constructed (country c, month m):
  - Binary: dummy indicator if an earthquake occurred in that month.
  - Intensity: average magnitude of earthquakes in that month.
- Final temporal coverage: January 2012 to November 2023 (aligned with spread data).
- Additional controls:
  - GDP per capita (PPP, constant 2021 international dollars) from World Development Indicators.
  - Banking system crises dummy as per Laeven and Valencia (2020): defined by significant banking sector distress plus substantial policy interventions.
  - Short-term debt especially relevant for countries with high near-term refinancing needs.
- Sample after merging data blocks:
  - Complete country-month observations retained: 11,389 (approximately 97 percent of the theoretical maximum).
  - Missingness primarily driven by gaps in short-term debt reporting among low-income countries.
  - Robustness: results robust to excluding these controls or imputing missing values with country-specific averages.
- Descriptive statistics and patterns:
  - Earthquake occurrence: roughly 28 percent of country-months experience at least one seismic event.
  - Most events are mild; upper tail includes several magnitude-6 or higher events.
  - Spreads: volatile and skewed; interquartile range from 64 to 410 basis points.
  - State capacity: dispersed, with majority clustering near the global mean and meaningful representation in both tails.

### D. Estimation Strategy
- Identification premise:
  - Earthquakes considered exogenous shocks (arrive without warning, unrelated to domestic policy or political cycles, vary in timing and magnitude across space).
  - Financial consequences depend on investors’ perceptions of fiscal capacity, institutional quality, and access to external resources.
- Benchmark panel difference-in-differences (DiD) framework (monthly frequency):
  - Equation (1):
    - ∆Spread_{c,m} = α_i + τ Spread_{c,m−1} + β Earthquake_{c,m−s} + λ_m + δ_season + ε_{c,m}
  - Variables and modeling choices:
    - Dependent variable: ∆spread_{c,m} = sovereign spread change for country c in month m (weighted average yield premium over U.S. Treasuries or German Bunds).
    - Earthquake variables: binary indicator and continuous intensity measure (average Richter magnitude in the month).
    - Lag parameter s captures delayed impact (s = 1 signifies one-month delay).
    - Fixed effects: country-fixed effects (α_i), month-fixed effects (λ_m), seasonal-fixed effects (δ_season). Month-fixed effects ordered from January 2012 (=1) to November 2023 (=140).
    - Estimator: high-dimensional fixed-effect estimator (Correia, 2017) to control multi-dimensional fixed effects.
  - Coefficient of interest: β (expected to be positive).
- Heterogeneity by state capacity and external debt:
  - Equation (2) adds interaction with average state capacity (SC̅):
    - ∆Spread_{c,m} = α_i + τ Spread_{c,m−1} + β Earthquake_{c,m−s} + δ Earthquake_{c,m−s} × SC̅ + λ_m + δ_season + ε_{c,m}
  - Equation (2.1) adds interaction with average short-term external debt (DEBT̅):
    - ∆Spread_{c,m} = α_i + τ Spread_{c,m−1} + β Earthquake_{c,m−s} + δ Earthquake_{c,m−s} × DEBT̅ + λ_m + δ_season + ε_{c,m}
  - SC̅: averaged value of state capacity; DEBT̅: average short-term external debt over GDP for the entire period.
- Unanticipated shock measurement:
  - Model to capture predictability and isolate unanticipated component:
    - EQ_{c,y,m} = γ EQ_{c,y,m−1} + λ_m + δ_season + α_i + ε_{c,y,m}
  - ε_{c,y,m} interpreted as the unanticipated (unpredicted) component of earthquake occurrence/intensity, used to denote earthquakes that occur without anticipation.
- Heterogeneity-robust DiD:
  - Methodology: De Chaisemartin and d’Haultfoeuille (2020, 2023) to handle heterogeneous treatment effects across time and space.
  - Implementation features:
    - Assess impact across six post-treatment periods (effects(6)).
    - Examine three pre-treatment periods for placebo tests (placebo(3)).
    - Focus on switchers in (treated cohorts) rather than switchers out.
    - Include state capacity as controls(Z) and linear time trends (trends_lin).
- Local Projections DID (LP-DiD):
  - Approach: Dube et al. (2023) LP-DiD to mitigate biases from negative weighting in conventional DiD and to flexibly define treated and control groups with staggered treatment adoption.
  - Estimation specification (for h = 0,...,6):
    - ∆_h Change in Spread_{c,m} = EQ_{c,m+h} − EQ_{c,m−1} = β_h LP−DiD ∆EQ_{c,m} + γ_m^h + ε_{c,m}^h
  - Sample restriction: newly treated (∆EQ_{c,m} = 1) or clean control (EQ_{c,p+h} = 0).
  - Weighting: use re-weighted LP-DiD regression approach (‘rw’) rather than equally-weighted average treatment effect on the treated.
  - Implementation note (Stata command reference provided in source): lpdid change_spread, time(cmonth) unit(ID) treat(treat) pre(3) post(6) rw.
- Inference and robustness:
  - Standard errors: heteroskedasticity-robust clustered at the country level.
  - Robustness checks: clustering by region and Conley-type spatial autocorrelation corrections (Cameron & Miller, 2015); inference remains valid under plausible correlation structures.
- Research question enabled by framework:
  - Leverage exogeneity of earthquakes with institutional heterogeneity to answer whether sovereign spreads react systematically to earthquakes and whether that response depends on state capacity to manage fallout.

*Source: IMF Working Paper — chapter "3. Data" and section "4. Estimation Strategy" (content unit wpiea2025218-source-pdf).*

### 5. Main Results

### 5. Main Results

### A. Benchmark Results
- Research question: how do sovereign spreads respond to earthquakes, and how does that response vary with state capacity?
- Estimation approach: high-dimensional fixed effects; dependent variable is the logarithmic change in sovereign bond spreads (Bloomberg Back Office); sample covers 96 countries from January 2012 through November 2023; earthquakes identified from NEHRP (over 44,089 observations) and coded by longitude/latitude; intensity measured by magnitude on the Richter scale; seasonal adjustments applied.
- Timing and magnitude of average effect:
  - Coefficient on earthquake dummy two months after the event (m−2): approximately 0.097 (9.7 basis points), significant at the 5 percent level.
  - Coefficients on earthquake dummies one month prior (m−1) and three months prior (m−3): small and statistically insignificant (examples: −0.022, −0.021).
  - Using intensity: each unit increase in average earthquake magnitude two months prior is associated with an increase of approximately 0.024 (2.4 basis points), significant at the 5 percent level.
- Dynamic/robust estimates using heterogeneity-robust DiD and Local Projections DiD:
  - Event-study (Table 2) treatment effects (point estimates):
    - Effect t+1: 0.114 (SE 0.139; LB-CI −0.159; UB-CI 0.386; Observation 795.000; Switchers 33.000)
    - Effect t+2: 0.262 (SE 0.204; LB-CI −0.137; UB-CI 0.661; Observation 776.000; Switchers 33.000)
    - Effect t+3: 0.500 (SE 0.257; LB-CI −0.005; UB-CI 1.004; Observation 755.000; Switchers 32.000)
    - Effect t+4: 0.547 (SE 0.278; LB-CI 0.003; UB-CI 1.091; Observation 740.000; Switchers 32.000)
    - Effect t+5: 0.685 (SE 0.354; LB-CI −0.008; UB-CI 1.378; Observation 726.000; Switchers 32.000)
    - Effect t+6: 0.209 (SE 0.503; LB-CI −0.777; UB-CI 1.194; Observation 674.000; Switchers 31.000)
  - Placebo leads (t−1 to t−3) are jointly insignificant; p-value of joint nullity = 0.384, supporting parallel trends and no anticipation.
  - By months 3–5 after the event, spreads rise by up to 0.685 (68.5 basis points) at month 5 with confidence intervals excluding zero in later months.
- Interpretation:
  - Markets react with a lag (about two months) as fiscal implications, relief spending, borrowing needs, or political instability materialize.
  - Consistent with sovereign risk theory where default risk increases via erosion of repayment capacity rather than the immediate shock.
  - Each standard deviation increase in state capacity attenuates the spread impact by roughly 24 basis points (see heterogeneity section).

### B. Heterogeneous effects (Thresholds)
- Method: threshold dummies for earthquake intensity at percentiles from the 30th to the 90th of the magnitude distribution; thresholds correspond to magnitudes (example: 30th percentile ≈ 1.0 on Richter; 90th percentile ≈ 4.2).
- Main findings:
  - The spread response two months after the shock is positive and significant across all tested thresholds, indicating discontinuous pricing.
  - Intensity effect at m−2 (continuous measure): 0.022 (SE 0.010), significant at **.
  - Threshold estimates (m−2):
    - 30th percentile threshold: 0.089** (SE 0.044)
    - 40th percentile threshold: 0.089** (SE 0.044)
    - 50th percentile threshold: 0.089** (SE 0.044)
    - 60th percentile threshold: 0.089** (SE 0.044)
    - 70th percentile threshold: 0.089** (SE 0.044)
    - 80th percentile threshold: 0.090** (SE 0.044)
    - 90th percentile threshold: 0.109** (SE 0.048)
  - For earthquakes in the top decile (90th percentile), the spread response is 0.109 (10.9 basis points), larger than for milder events.
- Interpretation:
  - Markets appear to use intensity thresholds; milder tremors often do not move markets, but beyond certain magnitudes events are interpreted as meaningful tests of sovereign solvency.

### C. The role of State Capacity
- Interaction strategy: interact earthquake occurrences and intensity with state capacity (SC) measures; ‘bottom SC’ = below 40th percentile; ‘top SC’ = above 60th percentile; state capacity measured per O’Reilly and Murphy (2022).
- Main findings (dummy earthquake at m−2):
  - Baseline dummy m−2: 0.143** (SE 0.056) in Column (1).
  - Interaction: each standard deviation increase in state capacity reduces the spread impact by −0.064 (−6.4 basis points) (SE 0.039).
  - Split-sample interactions:
    - Dummy m−2 × bottom SC (p40): 0.177* (SE 0.090)
    - Dummy m−2 × top SC (p40): −0.214** (SE 0.091)
  - Intensity (m−2) baseline: 0.034*** (SE 0.013) in Column (4).
  - Intensity interaction with SC:
    - Intensity × SC̅: −0.015* (SE 0.009)
    - Intensity × bottom SC (p40): 0.041** (SE 0.021)
    - Intensity × top SC (p40): −0.049** (SE 0.022)
- Split-sample/sign direction results:
  - In low-capacity countries, earthquakes raise spreads significantly (positive coefficients).
  - In high-capacity countries, the effect is reversed: spreads decline (negative interaction), implying investors may view disasters as signals of credible fiscal response or resilience.
- Interpretation:
  - State capacity functions as a filter, altering how markets interpret identical shocks; strong institutions can transform a disaster from a risk signal into a demonstration of fiscal credibility.

### D. Unanticipated Shocks
- Approach: isolate unanticipated shocks by removing predictable variation in earthquake timing (error term in equation (*)). Re-estimate main models using only the unanticipated component.
- Robustness of main findings:
  - Results remain robust when using unanticipated shocks only; unanticipated earthquakes lead to increases in bond spreads, materializing two months after the shock.
  - Institutional capacity moderates or reverses the spread increase stemming from unanticipated shocks.
- Key estimates (unanticipated shock models):
  - Unanticipated (Shock) dummy m−2: 0.091** (Columns 1–3) and 0.088** (Column 3), SEs ~0.046–0.045.
  - Unanticipated Shock dummy m−2 × State Capacity: −0.056 (SE 0.040) in Table 6 Column (1).
  - Unanticipated Shock dummy m−2 × bottom SC (p40): 0.165* (SE 0.092).
  - Unanticipated Shock dummy m−2 × top SC (p40): −0.209** (SE 0.093).
  - Unanticipated Shock intensity m−2: 0.033** (SE 0.013) to 0.039*** (SE 0.013) across specifications.
  - Intensity interactions mirror the baseline: negative with average state capacity (−0.014; SE 0.009) and sign reversals across bottom/top SC groups.
- Conclusion:
  - Removing anticipated components reinforces the estimated effect of earthquakes on bond spreads and confirms the moderating role of state capacity.

*Source: Authors.*

### 6. Robustness Checks

### 6. Robustness Checks

### A. Overview
- The study subjects baseline findings to a wide array of robustness checks.  
- Table 11 confirms a similar result for debt: countries with high external debt are more likely to benefit from international support post-disaster, partially offsetting market pessimism.

### B. Controlling for Crisis
- Motivation: global crises can significantly affect spreads; the study controls for banking system crises (defined by: (i) visible signs of financial distress in the banking sector, evidenced by significant bank runs, losses, or liquidations; and (ii) major policy interventions by authorities in response to these losses — data from Laeven and Valencia (2020)).
- Key regression regularities (Tables 7–9):
  - WA Spread lag (푊퐴 푆푝푟푒𝑎𝑑_{m−1}) coefficient: -0.010*** (0.002) across multiple specifications (Tables 7 and 8) and -0.010*** (0.002) in Table 9.
  - Earthquake (Dummy) and intensity interactions keep similar significance patterns when controlling for crises; earthquake effects remain present in months m−2 for many specifications.
  - Banking system Crisis (m−3) coefficient:
    - Table 7: 2.791*** (0.411) in column (1), with comparable estimates reported across columns: 2.790*** (0.410), 2.819*** (0.401).
    - Table 8: 2.785*** (0.412), 2.784*** (0.412), 2.815*** (0.403) across columns.
    - Table 9: 2.824*** (0.412), 2.824*** (0.401), 2.823*** (0.401) across columns.
  - Interpretation: crises significantly raise spreads (by over 280 basis points), but controlling for them does not alter the main results: earthquakes retain their spread impact and institutional interactions remain significant.
- Sample and fit details:
  - Table 7: Observations 11,389; R-squared values 0.015–0.081; Number of months 137; Number of countries 96.
  - Table 8: Observations 11,295; R-squared values 0.015–0.081; Number of months 136; Number of countries 96.
  - Table 9: Observations vary by column: 10,789 and 11,389; R-squared 0.081–0.084; Number of months 137; Number of countries 89–96.
- Notes on significance: Standard errors in parentheses. Significance: *** p<0.01, ** p<0.05, * p<0.1.

### C. Additional Robustness Checks: Triple Interactions (Earthquake × Income Level × State Capacity / Debt)
- Purpose: test whether moderating factors (state capacity, external debt) are driven by a country’s income level (IL).
- Summary of main findings:
  - Triple interaction (earthquake × income × state capacity) (Table 10):
    - State capacity can reduce the impact of earthquakes on spreads (columns [1] and [4] of Table 10).
    - Effects are especially meaningful in countries below the 40th income percentile (columns [2] and [5]).
    - Selected coefficients (Table 10):
      - Dummy_{m−2}: 1.512** (0.729) in column (1); 0.051 (0.089) in column (2); 0.144** (0.062) in column (3).
      - Dummy_{m−2} × IL̅ : -0.145* (0.077) reported in column (1) set of interactions.
      - Dummy_{m−2} × SC̅ : -1.023* (0.579) in column (1); -0.005 (0.052) in column (2); -0.090* (0.054) in column (3).
      - Dummy_{m−2} × SC̅ × IL̅ : 0.098* (0.058) in column (1) set.
      - Intensity_{m−2}: 0.378** (0.165) in column (2); 0.012 (0.020) in column (3); 0.034** (0.014) in column (6) cluster.
      - Intensity_{m−2} × SC̅ : -0.283** (0.134) in column (2); -0.000 (0.012) in column (3); -0.023* (0.012) in column (6).
      - Intensity_{m−2} × SC̅ × IL̅ : 0.027** (0.013).
    - Sample: Observations 10,877; R-squared ~0.080; Number of months 138; Number of countries 89.
  - Triple interaction (earthquake × income × external debt) (Table 11):
    - Higher external debt is associated with a reduced impact of earthquakes on spreads (columns [3] and [6] of Table 11), potentially due to foreign aid or concessional financing that attenuates sovereign bond spreads.
    - Income level does not significantly moderate this relationship.
    - Selected coefficients (Table 11):
      - WA Spread_{m−1}: -0.013*** (0.003) across columns.
      - Dummy_{m−2}: 0.632 (1.668) in column (1); 0.204 (0.221) in column (2); 0.363*** (0.116) in column (3).
      - Dummy_{m−2} × DEBT̅ : 0.161 (0.308) in column (1); -0.032 (0.029) in column (2); -0.043*** (0.014) in column (3).
      - Intensity_{m−2}: 0.144 (0.375) in column (1); 0.043 (0.049) in column (2); 0.082*** (0.026) in column (3).
      - Intensity_{m−2} × DEBT̅ : 0.041 (0.070) in column (1); -0.007 (0.007) in column (2); -0.010*** (0.003) in column (3).
    - Sample: Observations 6,331; R-squared 0.112; Number of months 138; Number of countries 53.
    - Note: Footnote elaborates that concessional funds can attenuate sovereign bond spreads by providing low-cost, stable financing and complement emergency assistance following natural disasters.
- Additional check: replacing average monthly earthquake magnitude with the maximum magnitude in each country-month; results (Appendix VII Panels A and B) are consistent with baseline findings and the conditioning role of state capacity holds.

### D. Split Sample by State Capacity (Annex III)
- Rationale: examine heterogeneous effects by high vs low state capacity (split at the 50th percentile).
- Panel A — High State Capacity (≥ p50):
  - WA Spread_{m−1}: -0.036*** (0.004) across multiple columns.
  - Dummy Earthquake_{m−2}: 0.018 (0.031) in column (2); other columns show small or insignificant dummy effects.
  - Intensity_{m−2}: 0.005 (0.007); Intensity_{m−3}: -0.012 to -0.014* (0.007) in some columns.
  - Observations 6,145; R-squared 0.017–0.076; Number of months 137; Number of countries 51.
- Panel B — Low State Capacity (< p50):
  - WA Spread_{m−1}: -0.008*** (0.003) across columns.
  - Dummy Earthquake_{m−2}: 0.180** (0.089) and 0.181** (0.089) in columns showing significant effects; other dummy coefficients vary.
  - Intensity_{m−2}: 0.041** (0.020) across columns; Intensity_{m−2} shows a pronounced positive effect in low state capacity sample.
  - Observations 5,244; R-squared 0.011–0.119; Number of months 137; Number of countries 45.
- Interpretation: Earthquake-related increases in sovereign spreads are concentrated in the low state capacity subsample; high state capacity countries show muted or negative spread responses.

### E. Descriptive Statistics (Annex II, key entries)
- Weighted Average Sovereign Spread (measured in 100 bps, Bloomberg Back Office): Obs 11,695; Mean 4.034; SD 7.845; Min 0.003; Max 139.86.
- Dummy Earthquake (NERHP): Obs 11,799; Mean 0.281; SD 0.450; Min 0.000; Max 1.000.
- Intensity Earthquake (NERHP, monthly average Richter): Obs 11,799; Mean 1.172; SD 1.912; Min 0.000; Max 6.800.
- Short-term External Debt (% GDP): WDI, Obs 6,537; Mean 8.172; SD 8.112; Min 0.000; Max 73.907.
- State Capacity (baseline z-score): Obs 9,135; Mean 0.933; SD 1.216; Min -2.152; Max 3.047.
- GDP per capita (log, constant 2021 international US$, PPP): Obs 11,618; Mean 9.888; SD 0.906; Min 7.273; Max 11.835.
- Banking Systematic Crisis (Dummy, Laeven and Valencia (2020)): Obs 11,799; Mean 0.0010; SD 0.032; Min 0.000; Max 1.000.
- Note: Data recorded monthly from January 2012 to November 2023.

### F. Concluding summary of robustness evidence
- Across specifications, controlling for banking system crises and conducting triple-interaction tests preserves the paper’s principal message:
  - Earthquakes affect sovereign spreads predominantly through institutional channels.
  - State capacity mitigates post-earthquake spread increases, particularly in low-income or low-capacity settings.
  - Higher external debt can be associated with reduced spread responses to earthquakes, consistent with partial offsetting via international/concessional support.
- Quantitative anchors repeatedly appear in the regressions:
  - WA Spread lag coefficients around -0.010*** (0.002) and in some specifications -0.013*** (0.003) (Table 11).
  - Banking system crisis increases spreads by coefficients near 2.78–2.83 (statistically significant at *** p<0.01).

*Source: 6. Robustness Checks — wpiea2025218-source-pdf*

### Annex IV. Identifying Unanticipated Earthquake

### Annex IV. Identifying Unanticipated Earthquake Shocks

### Key estimation (Identifying unanticipated shocks)
- Dependent variables: Dummy; Magnitude (Intensity)
- Columns (1) Dummy; (2) Magnitude (Intensity)
- Dummy m−1: 0.028***  (0.009)
- Magnitude m−1: 0.031***  (0.009)
- Country Fes: Yes (both)
- Seasonal Fes: Yes (both)
- Time Fes: Yes (both)
- Observations: 11,703 (both)
- R-squared: 0.615 (Dummy), 0.605 (Magnitude)
- Note: Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

*Source: Authors. Note: The occurrence of an unanticipated earthquake is quantified using the error term in equation (*). Standard errors are presented in parentheses.*

---

### Annex V. The Effect (Unanticipated) Earthquake on Sovereign Spread: Split The (SC)̅ Sample

### PANEL A. HIGH STATE CAPACITY (≥p50)
- Dependent variable: ∆Weighted Average Sovereign Spread
- WA Spread m−1: -0.036*** (0.004) across columns (1)–(6); in columns (3) and (6) R-squared rises to 0.076
- (Shock) Dummy m−1: -0.025 (0.031) (cols 1–3)
- (Shock) Dummy m−2: 0.019 (0.031) (cols 1–3)
- (Shock) Dummy m−3: -0.051* (0.031) (cols 1–3)
- (Shock) Intensity m−1: -0.007 (0.007) (cols 4–6)
- (Shock) Intensity m−2: 0.005 (0.007) (cols 4–6)
- (Shock) Intensity m−3: -0.012 to -0.013* (0.007) (cols 4–6)
- Country Fes: Yes (all)
- Seasonal Fes: No/Yes/Yes as reported by column
- Time Fes: No/No/Yes as reported by column
- Observations: 6,096 (all columns)
- R-squared: 0.018, 0.020, 0.076 (cols 1–3) and same pattern for cols 4–6
- Number of months: 136
- Number of countries: 51

### PANEL B. LOW STATE CAPACITY (<p50)
- Dependent variable: ∆Weighted Average Sovereign Spread
- WA Spread m−1: -0.008*** (0.003) across columns (1)–(6)
- (Shock) Dummy m−1: -0.009, -0.006, -0.036 (0.090), (0.090), (0.087) (cols 1–3)
- (Shock) Dummy m−2: 0.167* (0.090) (cols 1–3)
- (Shock) Dummy m−3: 0.021, 0.027, 0.049 (0.091), (0.090), (0.087) (cols 1–3)
- (Shock) Intensity m−1: -0.002, -0.001, -0.008 (0.020) (cols 4–6)
- (Shock) Intensity m−2: 0.039* (0.020) (cols 4–6)
- (Shock) Intensity m−3: 0.006, 0.007, 0.012 (0.021), (0.020), (0.020) (cols 4–6)
- Country Fes: Yes (all)
- Seasonal Fes and Time Fes vary by column as reported
- Observations: 5,199 (all)
- R-squared: 0.011, 0.016, 0.119 (cols 1–3) and same pattern for cols 4–6
- Number of months: 136
- Number of countries: 45

*Source note: The occurrence of an unanticipated earthquake is quantified using the error term in equation (*). Standard errors are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1.*

---

### Annex VI. The Effect of (Unanticipated) Earthquake on Sovereign Spread: Debt-To-GDP

### Main coefficients and interactions
- Dependent variable: ∆Weighted Average Sovereign Spread
- WA Spread m−1: -0.013*** (0.003) (columns 1–6)
- (Shock) Dummy m−2:
  - Column (1): 0.352*** (0.116)
  - Column (2): -0.087 (0.094)
  - Column (3): 0.174* (0.090)
- (Shock) Dummy m−2 × DEBT̅ : -0.042*** (0.014) (reported)
- (Shock) Dummy m−2 × bottom DEBT̅ (p40): 0.366** (0.145)
- (Shock) Dummy m−2 × Top DEBT̅ (p40): -0.290** (0.148)
- (Shock) Intensity m−2:
  - Column (4): 0.079*** (0.026)
  - Column (5): -0.020 (0.022)
  - Column (6): 0.039* (0.020)
- (Shock) Intensity m−2 × DEBT̅ : -0.010*** (0.003)
- (Shock) Intensity m−2 × bottom DEBT̅ (p40): 0.082** (0.033)
- (Shock) Intensity m−2 × Top DEBT̅ (p40): -0.065* (0.034)
- Country Fes: Yes (all)
- Seasonal Fes: Yes (all)
- Time Fes: Yes (all)
- Observations: 6,278 (all)
- R-squared: 0.111–0.112 across columns
- Number of months: 137
- Number of countries: 53

- Note: The occurrence of an unanticipated earthquake is quantified using the error term in equation (*). Significance levels: *** p<0.01, ** p<0.05, * p<0.1.

---

### Annex VII. The Effect of Earthquakes on Sovereign Spreads: Robustness Check with Maximum Magnitude

### PANEL A: MAXIMUM MAGNITUDE OF EARTHQUAKE
- Dependent variable: ∆Weighted Average Sovereign Spread
- WA Spread m−1: -0.010*** (0.002) (columns 1–6)
- Max Intensity m−2:
  - Column (1): 0.022** (0.010)
  - Column (2): 0.034*** (0.012)
  - Column (3): 0.006 (0.013)
  - Column (4): 0.039*** (0.012)
  - Column (5): 0.006 (0.013)
  - Column (6): 0.040*** (0.012)
- Max Intensity m−2 × SC̅ : -0.015* (0.009)
- Max Intensity m−2 × bottom SC̅ (p40): 0.038* (0.020) and 0.040** (0.020) in alternate columns
- Max Intensity m−2 × Top SC̅ (p40): -0.050** (0.021) and -0.051** (0.021) in alternate columns
- Banking system Crisis m−3: 2.824*** (0.401) and 2.823*** (0.401) in relevant columns
- Country Fes, Seasonal Fes, Time Fes: Yes (all)
- Observations: 11,484; 10,877; 11,484; 11,484; 11,389; 11,389 (by column)
- R-squared: 0.076–0.081 across columns
- Number of months: 96 or 89 (as reported)
- Number of countries: 137–138 (as reported)
- Note: Maximum intensity is defined as the highest ground shaking recorded from an earthquake within a given month in country c, rather than being calculated as the average value on the Richter scale.

### PANEL B: (UNANTICIPATED) MAXIMUM MAGNITUDE OF EARTHQUAKE
- Dependent variable: ∆Weighted Average Sovereign Spread
- WA Spread m−1: -0.010*** (0.002) (columns 1–6)
- (Shock) Max Intensity m−2:
  - Column (1): 0.022** (0.010)
  - Column (2): 0.034*** (0.012)
  - Column (3): 0.007 (0.013)
  - Column (4): 0.039*** (0.012)
  - Column (5): 0.007 (0.013)
  - Column (6): 0.040*** (0.012)
- (Shock) Max Intensity m−2 × SC̅ : -0.015* (0.009)
- (Shock) Max Intensity m−2 × bottom SC̅ (p40): 0.037* (0.020) and 0.038* (0.020) in alternate columns
- (Shock) Max Intensity m−2 × Top SC̅ (p40): -0.050** (0.021) (repeated)
- Banking system Crisis m−3: 2.823*** (0.401) and 2.822*** (0.401) in relevant columns
- Country Fes, Seasonal Fes, Time Fes: Yes (all)
- Observations: 11,389; 10,789; 11,389; 11,389; 11,389; 11,389 (by column)
- R-squared: 0.077–0.081 across columns
- Number of months: 96 or 89 (as reported)
- Number of countries: 137–138 (as reported)
- Note: The occurrence of an unanticipated earthquake is quantified using the error term in equation (*). Standard errors are presented in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

---

*Italic: Source: IMF Working Paper "Earthquakes and Emerging Market Sovereign Bond Spreads" (Working Paper No. WP/2025/218). Authors’ tabulated results and notes as presented in Annexes IV–VII.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025218-source-pdf.pdf_