## 1.1    Robust Correction Mechanisms

## Source details

**Canonical URL:** [1.1    Robust Correction Mechanisms](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025195-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025195-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025195-source-pdf.pdf.json)

---

### Rationale and theoretical foundations
- Robust correction mechanisms align with policy advice and economic theory advocating for more ambitious fiscal consolidation when fiscal risk is higher.
- Key theoretical results cited:
  - Caselli et al. (2022): debt above the safety margin should prompt a medium-term fiscal path back to the anchor, with pace based on risk (higher risk → faster adjustment).
  - Bohn (1998, 2008): a fiscal reaction function where the primary fiscal balance increases with the level of debt is sufficient for the intertemporal government budget constraint to hold.
  - Hatchondo et al. (2020): optimal fiscal plans feature more ambitious consolidation when sovereign risk is endogenous and higher.
  - Sublet (2023): optimal fiscal rules imply a gradual schedule of tighter adjustments with larger tail risks.
  - Analogy to monetary policy: fiscal rules with a fiscal anchor and robust correction mechanisms resemble Taylor rules (Leeper, 2010).

### Current adoption and example countries
- In the 2025 IMF Fiscal Rules Dataset, of more than 120 countries with fiscal rules, only six are identified as implementing pre-emptive robust correction mechanisms (Alonso et al., 2025): Armenia, Costa Rica, Cyprus, Czech Republic, Poland, and Slovakia.
- The source presents brief descriptions of fiscal rules in these countries.

### Example: Costa Rica
- Design features:
  - Limits to the growth of expenditures as a function of GDP growth over the previous four years, with lower limits to expenditure growth for higher debt levels.
  - If debt exceeds 60% of GDP, the limit applies to total expenditure.
  - The rule prescribes a limit to the indexation of pensions and public sector salaries.
- Operational implications:
  - The rule does not directly limit fiscal balances or anchor debt levels, but compliance with expenditure limits makes it unlikely the government would implement discretionary measures (e.g., lowering taxes) that violate expenditure limits.
  - Cyclical revenue declines could weaken fiscal balances while remaining compliant; additional measures to reduce debt could be expected to avoid stricter limits and political costs.

### The cost of fiscal risk
- Fiscal risk raises sovereign spreads and broader borrowing costs:
  - Arellano et al. (2023): a 100 bps increase in sovereign spreads leads to a 60 bps increase in firm borrowing cost, which lowers GDP; example: in the 2012 crisis, real GDP in Italy would have fallen 3.1% instead of 6.3% if spreads had not increased.
- Countercyclicality of spreads is linked to volatility and procyclical fiscal policy (Bianchi et al., 2023; Cuadra et al., 2010; Neumeyer and Perri, 2005).
- Empirical pattern: sovereign spread in 2019 is positively correlated with the spread increase in 2020 (COVID-19), implying economies with more sovereign risk may have weaker access to debt markets after shocks.
- Political economy explanations for deficit bias: political myopia (Aguiar et al., 2020; Azzimonti, 2011; Halac and Yared, 2014, 2018) and time inconsistency (Chari and Kehoe, 2007; Hatchondo et al., 2020).
- Quantitative models rely on deficit bias to match high sovereign spreads (Chatterjee and Eyigungor, 2012; Hatchondo and Martinez, 2009).

### Empirical literature on fiscal rules and sovereign spreads
- Empirical findings summarized:
  - Capraru et al. (2025): in the European Union, fiscal rules and compliance reduce sovereign spreads.
  - Iara and Wolff (2014): stronger national fiscal rules in euro-area member states reduce spreads.
  - Kalan et al. (2018): for countries under an Excessive Deficit Procedure, spreads are on average 50-150 basis points higher than for countries not under the procedure.
  - Feld et al. (2017): strong and credible balanced budget rules at sub-national level in Switzerland reduce risk premia.
  - Islamaj et al. (2024): presence of fiscal rules is statistically significantly associated with lower spreads during COVID-19.
- Synthetic-control evidence:
  - Lang et al. (2023): eligibility for official debt relief via the Debt Service Suspension Initiative was associated with larger declines in borrowing costs.
  - Ulloa-Suarez et al. (2025): activation of escape clauses in fiscal rules reduced debt levels and sovereign spreads.

### Data and synthetic control approach used in the study
- Data sources and sample construction:
  - Fiscal rules: 2025 update of the IMF Fiscal Rules Dataset (Alonso et al., 2025).
  - Sovereign spreads: IMF Sovereign Spread Monitor (SSM) where possible; weekly data from April 27, 2012, to November 3, 2023.
  - For Poland and Slovakia: CDS spreads used when SSM spreads are not available at rule introduction.
  - Panel robustness regressions: five macro indicators at annual frequency from the IMF World Economic Outlook — public debt over GDP, real GDP growth, reserves over GDP, fiscal balance, current account balance.
- Empirical observation (case trajectories):
  - Figure 3 observation: fiscal risk (sovereign spread) decreased in countries adopting FRRC and continued to decline years after introduction, consistent with markets learning about rule credibility and complementary government measures.
  - Costa Rica: spread >500 basis points at rule approval (December 2018); spread decreased about 150 basis points after approval, increased to >900 basis points with COVID-19, then declined to about 200 basis points by end of 2020; distribution percentile moved from 80th at approval to below 50th later.
- Synthetic control methodology details:
  - Treated unit: first unit; donor pool: potential controls.
  - SC estimator: ˆτ1,t = Y1,t − sum_{c=2}^{C+1} w_c Y_c,t, with sum of w_c = 1 and w_c ≥ 0.
  - Weights chosen to minimize ∥X1 − X0 W∥ = (sum_{h=1}^k v_h (X_h1 − w_2 X_h2 − ... − w_{C+1} X_h,C+1)^2)^{1/2}, ensuring preintervention match.
  - Average effect across G = 6 case studies: ̄τ = (1/G) sum_{g=1}^G (ˆτ_{g,T0+1},...,ˆτ_{g,T}).
  - Inference via placebo tests: apply SC to every potential control and compute p-values for individual and average effects (formulas (3) and (4) in the source).
- Model specification notes:
  - Baseline treatment date: approval by Congress of the fiscal rule as in Figure 1.
  - Donor pool excludes countries with spreads >4,000 basis points at any time during the window.
  - Baseline predictors include sovereign spreads on seven pre-approval dates (last weekly spread and previous six month-end weekly spreads).
  - Robustness including standard macro determinants (GDP growth, public debt, fiscal balance, current account balance, international reserves) does not change main results.

### Main synthetic-control results (summary)
- Pretreatment fit:
  - Spread gap (actual log spread − synthetic log spread) is close to zero before treatment, indicating no systematic pretreatment differences.
- Post-treatment effects (baseline):
  - Median spread gap stabilizes around -25% after one year.
  - For the average treated economy with a pretreatment spread of 300 bps, a -25% median gap corresponds to -75 bps.
  - Single-country patterns:
    - Slovak Republic: approval did not immediately stop an increase; gap started to decrease after rule entered into force in March 2012.
    - Costa Rica: gap increased with COVID-19.
    - Cyprus: spreads were already declining before approval (possible market anticipation).
- Normalization: spreads are log-transformed so log differences are comparable across countries.

### Inference and placebo testing
- Placebo testing procedure:
  - Re-assign treatment to each donor pool country, estimate fictitious synthetic control using remaining donors and the originally treated unit.
  - Include only placebos with as good a pretreatment fit as the country they serve as control for (following Cavallo et al., 2013).
  - Computational constraint: full enumeration would require more than 2,500,000,000 averages.
  - Practical approach: draw with replacement 10,000 placebo series for each event and average across 6 events for each draw, producing 10,000 placebo average effects ̄τPLt.
  - Placebo counts used: 27 (Armenia), 65 (Costa Rica), 65 (Cyprus), 44 (Czech Republic), 12 (Poland), 42 (Slovak Republic).
- Statistical inference:
  - Probability the estimated average effects were obtained by chance is below 1 percent for most of the post-treatment sample, especially after one year.
  - Results become temporarily insignificant for about 4 weeks after the fifth month due to high-frequency volatility, then revert below 5 and 1 percent thresholds.
  - Country-level p-values: except for the Slovak Republic, country-specific p-values provide evidence of significant estimated effects.

### Robustness checks (synthetic-control)
- Alternative specifications examined (Figure 7):
  - i) Matching including pretreatment macro characteristics (GDP growth, public debt, fiscal balance, current account balance, international reserves) in addition to sovereign spread.
  - ii) Excluding donor countries with a spread greater than 1500 basis points at any point.
  - iii) Including only countries with fiscal rules in the control group.
  - iv) Removing the country with the largest weight from the donor pool for each treated unit.
- Robustness results:
  - Alternatives i), ii), iii): median effect somewhat lower but still economically significant, around 20% after one year (compared with 25% baseline).
  - Alternative iv): effect smaller, around 10%, with greater uncertainty due to reduced fit of synthetic controls, especially for early adopters.

### Key quantitative summary from synthetic-control analysis
- Median spread gap after one year: -25%.
- Equivalent basis-point change for average treated country (pretreatment spread 300 bps): -75 bps.
- Placebo draws per event: 10,000.
- Full enumeration of placebos would require more than 2,500,000,000 averages.
- Temporary insignificance window in placebo p-values: about 4 weeks after the fifth month.

### Panel regression framework (overview) and selected results
- Panel setup:
  - Sample: 99 countries, period 2012-2023, annual estimation.
  - Baseline dynamic specification (equation (5)):
    - Dependent variable: ln(spread) (SSM sovereign spread logged).
    - Controls: GDP growth, debt-GDP ratio, primary balance-GDP ratio, reserves-GDP ratio, current account-GDP ratio, lagged ln(spread), VIX, U.S. Federal Fund Rate.
    - Dummies: FR (fiscal rule without robust correction mechanism) and FRRC (fiscal rule with robust correction mechanism; subset of FR).
  - Estimator: System GMM (SGMM) with Roodman procedure, forward orthogonal deviations transformation, Windmeijer finite-sample correction; lags 2–4 used for instruments; parsimonious specification also reported.
  - Outliers excluded: Belarus, Ecuador, Ethiopia, Lebanon, Pakistan, Russia, Sri Lanka, Tunisia, Ukraine and Zambia (spreads >4,000 bps).
- Sample information (Table 1):
  - Observations: 776
  - # Countries: 89
- Selected OLS findings (columns (1)-(3)):
  - Fiscal Rule Dummy: -0.159**; -0.166**; -0.111*  [standard errors: [0.069], [0.068], [0.056]]
  - Robust Correction Mechanism Dummy: -0.215**; -0.224**; -0.157**  [standard errors: [0.099], [0.100], [0.065]]
  - Growth Rate: -0.013***  [0.003]
  - Current Account-GDP: -0.012***  [0.003]
  - Lagged ln(spread): 0.674***; 0.670***; 0.665***  [standard errors: [0.049], [0.048], [0.049]]
  - R-squared: 0.948; 0.949; 0.949
- Selected System GMM findings (columns (4)-(11)):
  - Robust Correction Mechanism Dummy coefficients across GMM columns: -0.182***; -0.118**; -0.138***; -0.142***; -0.157***; -0.120**; -0.141***  [standard errors reported in table].
  - Fiscal Rule Dummy coefficients (selected): -0.069; -0.077; -0.103*  [standard errors: [0.048], [0.055], [0.056]]
  - Growth Rate and Current Account-GDP negative and often significant across specifications.
  - Lagged ln(spread): ranges from 0.884*** to 0.917*** across GMM columns.
  - # Instruments across GMM columns: 18, 17, 25, 24, 10, 9, 17, 16.
  - AR(1) p-value: 0.000 (in many columns).
  - AR(2) p-values reported include 0.195, 0.212, 0.460, 0.456, 0.145, 0.151, 0.480, 0.490.
  - Hansen p-values reported include 0.000, 0.000, 0.397, 0.453, 0.643, 0.670, 0.601, 0.603.
- Interpretation:
  - FRRC dummy is negative and statistically significant in both OLS and GMM.
  - FR dummy results are more nuanced and less consistently significant than FRRC.
  - Estimated effect of FRRC ranges from about 17% to 32% across specifications.
  - Parsimonious specification with time fixed effects (column 10): FRRCs associated with a total long-run drop of about 20% in the sovereign spread (exp(-0.103-0.120) – 1 ≈ -20%).
  - Marginal effect of the presence of an FRRC ranges from about 12% to 20% across specifications.

### Robust patterns across methods
- Convergence of evidence:
  - Synthetic-control: one-year median spread decline of about 25% (≈75 bps for average treated country with pretreatment spread of 300 bps), statistically significant by placebo testing.
  - Panel regressions: FRRC dummy consistently negative and significant; long-run declines in spreads consistent with synthetic-control magnitudes (two-digit percent range).
- Other determinants consistent with literature:
  - Lower growth rates, higher current account deficits, higher global uncertainty (VIX), and higher global interest rates (U.S. Federal Fund Rate) are associated with higher spreads.
  - Lagged debt and reserves not statistically significant in most specifications.

### Overall interpretation and policy-relevant findings
- Main policy-relevant conclusions:
  - Fiscal rules with robust correction mechanisms (FRRC) are associated with economically important, statistically significant, and robust reductions in sovereign spreads for more than a year after introduction.
  - Quantitative highlights:
    - One-year median spread decline (synthetic control): about 25% or 75 bps for the average treated country (pretreatment spread 300 bps).
    - Panel estimates imply long-run reductions in spreads in the range of roughly 17%–32% across specifications; parsimonious specification shows ≈20% long-run drop.
    - Marginal effect of a robust correction mechanism relative to other fiscal rules: about 12%–20% reduction in spreads across specifications.
- Caveats and complementarities:
  - FRRCs are not panaceas: other rule characteristics (coverage and exclusions, calibration of the fiscal anchor, medium-term fiscal frameworks, institutions for fiscal oversight) matter for performance.
  - After extraordinary shocks (e.g., COVID-19), discretion may be applied via escape clauses to deviate from FRRC-prescribed consolidation.
  - Sovereign spreads are volatile at high frequency; governments should not mechanically respond to high-frequency spread movements when guiding fiscal policy.

*Source: wpiea2025195-source-pdf — 1.1 Robust Correction Mechanisms; 3.2 Results*

### 1.1    Robust Correction Mechanisms

### 1.1    Robust Correction Mechanisms

### Rationale and theoretical foundations
- Robust correction mechanisms are consistent with policy advice and economic theory advocating for more ambitious fiscal consolidation when fiscal risk is higher.
- Caselli et al. (2022): “A country whose debt level exceeds the safety margin to its debt limit should commit to a medium-term fiscal path that brings it back to the anchor over time. The pace of adjustment set in the fiscal plans should be based on an assessment of risks: the higher the risk, the faster the adjustment...”.
- Bohn (1998, 2008): a fiscal reaction function such that the primary fiscal balance increases in the level of debt (associated with fiscal risk) is sufficient for the intertemporal government budget constraint to hold.
- Hatchondo et al. (2020): in a model with endogenous sovereign risk, the optimal fiscal plan (which fiscal rules should try to implement) features a more ambitious fiscal consolidation when fiscal risk is higher.
- Sublet (2023): the optimal fiscal rule features a gradual schedule of tighter adjustments when the tail risks are larger.
- Analogy to monetary policy: fiscal rules with a clear fiscal anchor, a robust correction mechanism, and a stronger fiscal response when risk is higher resemble Taylor rules in monetary policy (Leeper, 2010).

### Current adoption and example countries
- In the 2025 IMF Fiscal Rules Dataset, of more than 120 countries with fiscal rules, only six are identified as implementing pre-emptive robust correction mechanisms (Alonso et al., 2025): Armenia, Costa Rica, Cyprus, Czech Republic, Poland, and Slovakia.
- Figure 1 in the source presents a brief description of fiscal rules in these countries.

### Example: Costa Rica
- The Costa Rica correction mechanism establishes limits to the growth of expenditures as a function of GDP growth over the previous four years, with lower limits to expenditure growth for higher debt levels.
- If debt exceeds 60% of GDP, the limit applies to total expenditure.
- The rule also prescribes a limit to the indexation of pensions and public sector salaries.
- Note: While the rule does not directly limit fiscal balances or anchor debt levels, it is unlikely the government would implement discretionary measures that violate expenditure limits (e.g., lowering taxes) while complying with the rule. Cyclical reductions in tax revenue could weaken fiscal balances and increase debt while complying with the rule, but additional measures to reduce debt could be expected to avoid the stricter limits and associated political costs.

### The cost of fiscal risk
- Fiscal risk is very costly and affects sovereign spreads and broader borrowing costs.
- Arellano et al. (2023): a 100 bps increase in sovereign spreads leads to a 60 bps increase in firm borrowing cost, which in turn lowers GDP. Example: in the 2012 crisis, real GDP in Italy would have fallen 3.1% instead of 6.3% if sovereign spreads had not increased.
- Countercyclicality of sovereign spreads is linked to increased volatility and procyclical fiscal policy (Bianchi et al., 2023; Cuadra et al., 2010; Neumeyer and Perri, 2005).
- Empirical pattern: the sovereign spread in 2019 is positively correlated with the spread increase in 2020 (owing in part to COVID-19), indicating economies with more sovereign risk may have weaker access to debt markets after adverse shocks.
- Deficit bias explanations: political myopia (Aguiar et al., 2020; Azzimonti, 2011; Halac and Yared, 2014, 2018) or time inconsistency problems (Chari and Kehoe, 2007; Hatchondo et al., 2020) can explain why governments tolerate significant fiscal risk.
- Quantitative models: deficit bias is essential to account for high sovereign spreads in the data (Chatterjee and Eyigungor, 2012; Hatchondo and Martinez, 2009).

### Empirical literature on fiscal rules and sovereign spreads
- Capraru et al. (2025): in the European Union, fiscal rules and compliance reduce sovereign spreads.
- Iara and Wolff (2014): within the euro area, stronger national fiscal rules in member states reduce sovereign spreads.
- Kalan et al. (2018): for countries under an Excessive Deficit Procedure, spreads are on average 50-150 basis points higher than for countries not under the procedure.
- Feld et al. (2017): strong and credible balanced budget rules at the sub-national level in Switzerland reduce risk premia.
- Islamaj et al. (2024): presence of fiscal rules is statistically significantly associated with lower sovereign spreads during the COVID-19 crisis.
- Synthetic control studies:
  - Lang et al. (2023): countries eligible for official debt relief through the Debt Service Suspension Initiative experienced a larger decline in borrowing costs compared to similar ineligible countries.
  - Ulloa-Suarez et al. (2025): activation of escape clauses in fiscal rules reduced debt levels and sovereign spreads.

### Data and synthetic control approach used in the study
- Fiscal rules data: 2025 update of the IMF Fiscal Rules Dataset (Alonso et al., 2025).
- Sovereign spreads: IMF Sovereign Spread Monitor (SSM) where possible; weekly data from April 27, 2012, to November 3, 2023.
- For Poland and Slovakia, SSM spreads are not available when they introduce the rule; CDS spreads are used in those cases.
- Robustness and panel regressions use five key macroeconomic indicators at annual frequency from the IMF World Economic Outlook: public debt over GDP, real GDP growth, reserves over GDP, fiscal balance, and current account balance.
- Empirical observation: Figure 3 shows fiscal risk (sovereign spread) decreased in countries that adopted FRRC and continued to decline years after rule introduction — markets learn over time about rule credibility and governments may adopt complementary measures.
- Costa Rica case: spread >500 basis points at rule approval (December 2018); spread decreased about 150 basis points after approval, increased to >900 basis points with COVID-19, then declined to about 200 basis points by end of 2020; distribution percentile moved from 80th at approval to below 50th later.
- Synthetic control methodology:
  - Treated country is the first unit; donor pool is potential controls.
  - SC estimator: ˆτ1,t = Y1,t − sum_{c=2}^{C+1} w_c Y_c,t, with weights summing to one and nonnegative.
  - Weights chosen to minimize ∥X1 − X0 W∥ = (sum_{h=1}^k v_h (X_h1 − w_2 X_h2 − ... − w_{C+1} X_h,C+1)^2)^{1/2}, ensuring preintervention match.
  - Average effect across G = 6 case studies: ̄τ = (1/G) sum_{g=1}^G (ˆτ_{g,T0+1},...,ˆτ_{g,T}).
  - Inference via placebo tests: apply SC to every potential control and compute p-values for individual and average effects (formulas (3) and (4) in the source).
- Model specification details:
  - Baseline treatment is approval by Congress of the fiscal rule as in Figure 1.
  - Donor pool excludes countries with spreads >4,000 basis points at any point during the window.
  - Baseline predictors include sovereign spreads on seven specific pre-approval dates (last weekly spread and previous six month-end weekly spreads) to minimize cherry-picking.
  - Robustness check includes standard macroeconomic determinants (GDP growth, public debt, fiscal balance, current account balance, international reserves) in addition to lagged sovereign spreads; this does not change main results.

*Source: wpiea2025195-source-pdf - 1.1    Robust Correction Mechanisms*

### 3.2    Results

### 3.2    Results

### Main results from synthetic control analysis
- Methodology follows Lang et al. (2023) and Cavallo et al. (2013) to study high-frequency spread responses across multiple countries with different treatment times.
- Synthetic control weights for each treated country are reported in Table A1 (not reproduced here).
- Figure 4 (described) reports mean, median, and inter-quartile range of the 12-week moving average of the difference between actual log sovereign bond spreads and their synthetic controls (the spread gap) around the treatment date.
- Pretreatment behavior:
  - In the months before treatment (before the rule is approved by Congress), the spread gap is close to zero, indicating no systematic pretreatment differences between treated countries and their counterfactuals.
- Post-treatment behavior:
  - After treatment, the spread gap decreases; the median gap stabilizes around -25% after one year.
  - For the average economy in the treated sample (which had a spread of 300 bps prior to treatment), a -25% median gap translates into -75 basis points (bps).
  - Single-country plots (Figure 5) show a significant and persistent spread decline relative to the counterfactual in all six FRRC countries (Armenia, Costa Rica, Cyprus, Czech Republic, Poland, Slovak Republic), though declines are not always immediate:
    - Slovak Republic: approval did not immediately stop the increase in the spread gap; the gap started to decrease after the rule entered into force in March 2012.
    - Costa Rica: gap increased with COVID-19.
    - Cyprus: spreads were already declining before approval, suggesting some market anticipation.
- Normalization: spreads are log-transformed so log differences can be compared across countries.

### Inference and placebo testing
- Placebo testing approach:
  - Sequentially “re-assign” the treatment to each donor pool country and estimate a fictitious synthetic control using remaining donor countries and the originally treated unit.
  - Following Cavallo et al. (2013), include only placebos with as good a pretreatment fit as the country they serve as control for.
  - Computational constraint: computing all possible placebo averages would require more than 2,500,000,000 averages.
  - Practical approach used: draw with replacement 10,000 placebo series for each event and take the average across all 6 events for each draw, producing 10,000 placebo average effects ̄τPLt.
  - Placebo counts computed: 27 placebos for Armenia, 65 for Costa Rica, 65 for Cyprus, 44 for Czech Republic, 12 for Poland, and 42 for Slovak Republic.
- Significance of average estimated effect (Figure 6):
  - The probability the estimated effects were obtained by chance is below 1 percent for most of the post-treatment sample, especially after a year.
  - Results become temporarily insignificant for about 4 weeks after the fifth month due to high-frequency volatility, then revert below 5 and 1 percent thresholds.
  - Overall indicates strong evidence of a causal average effect of FRRC approval on spread gaps.
- Country-specific inference:
  - Table A2 reports country-specific significance levels at four post-treatment dates.
  - Except for the Slovak Republic, p-values provide evidence that country-specific estimated effects are significant.

### Robustness checks
- Figure 7 examines sensitivity of median spread gap under alternative assumptions:
  - i) Match countries including pretreatment macroeconomic characteristics (GDP growth, public debt, fiscal balance, current account balance, international reserves) in addition to sovereign spread (upper-left panel).
  - ii) Exclude donor countries with a spread greater than 1500 basis points at any point (upper-right panel).
  - iii) Only include countries with fiscal rules in the control group (lower-left panel).
  - iv) Remove the country with the largest weight from the donor pool for each treated unit (lower-right panel).
- Robustness findings:
  - Alternatives i), ii), and iii): spread differentials are somewhat lower but still economically significant, with a median effect of around 20% after a year (compared with 25% in baseline).
  - Alternative iv) (removing largest-weight donor): effect smaller, around 10%, but with greater uncertainty due to reduced fit of synthetic controls, especially for early adopters of FRRCs.

### Key quantitative summary from synthetic-control analysis
- Median spread gap after one year: -25%.
- Equivalent basis-point change for average treated country (spread of 300 bps pretreatment): -75 bps.
- Placebo draws per event: 10,000.
- Full enumeration of placebos would require more than 2,500,000,000 averages.
- Temporary insignificance window: about 4 weeks after the fifth month in placebo p-values.

### Panel regression framework (preview of Section 4)
- Panel: 99 countries, period 2012-2023, annual estimation.
- Baseline dynamic specification (equation (5)) includes:
  - Dependent variable: ln(spread) (SSM sovereign spread logged).
  - Controls: GDP growth, debt-GDP ratio, primary balance-GDP ratio, reserves-GDP ratio, current account-GDP ratio, lagged ln(spread), VIX, U.S. Federal Fund Rate.
  - Dummies: FR (fiscal rule without robust correction mechanism) and FRRC (fiscal rule with robust correction mechanism; subset of FR).
- Estimator: System GMM (SGMM) with Roodman procedure, forward orthogonal deviations (FOD) transformation, Windmeijer finite-sample correction; lags 2–4 used for instruments; a parsimonious specification also reported.
- Outliers excluded: Belarus, Ecuador, Ethiopia, Lebanon, Pakistan, Russia, Sri Lanka, Tunisia, Ukraine and Zambia (countries with spreads greater than 4,000 basis points at any time).

### Panel regression results (Table 1) — selected coefficients and statistics
- Sample:
  - Observations: 776
  - # Countries: 89
- OLS estimates (columns (1)-(3)):
  - Fiscal Rule Dummy: -0.159**; -0.166**; -0.111*  [standard errors: [0.069], [0.068], [0.056]]
  - Robust Correction Mechanism Dummy: -0.215**; -0.224**; -0.157**  [standard errors: [0.099], [0.100], [0.065]]
  - Growth Rate: -0.013*** (in each of columns (1)-(3))  [0.003]
  - Current Account-GDP: -0.012*** (in each of columns (1)-(3))  [0.003]
  - Lagged ln(spread): 0.674***; 0.670***; 0.665***  [standard errors: [0.049], [0.048], [0.049]]
  - R-squared reported: 0.948; 0.949; 0.949
- System GMM estimates (columns (4)-(11)):
  - Fiscal Rule Dummy (selected columns): -0.069; -0.077; -0.103*  [standard errors: [0.048], [0.055], [0.056]]
  - Robust Correction Mechanism Dummy (columns (4)-(11)): -0.182***; -0.118**; -0.138***; -0.142***; -0.157***; -0.120**; -0.141***  [standard errors: [0.068], [0.050], [0.049], [0.050], [0.052], [0.050], [0.052]]
  - Growth Rate (selected): -0.022***; -0.022***; -0.005; -0.004; -0.011***; -0.012***; -0.008**; -0.008**  [various standard errors reported in table]
  - Current Account-GDP (selected): -0.025***; -0.024***; -0.013***; -0.013***; -0.014***; -0.014***; -0.015***; -0.014***  [standard errors reported]
  - Lagged ln(spread): ranges from 0.884*** to 0.917*** across GMM columns [standard errors reported].
  - # Instruments (reported across GMM columns): 18, 17, 25, 24, 10, 9, 17, 16 (varies by column).
  - AR(1) p-value: 0.000 (in many columns).
  - AR(2) p-value: values reported include 0.195, 0.212, 0.460, 0.456, 0.145, 0.151, 0.480, 0.490.
  - Hansen p-value: reported values include 0.000, 0.000, 0.397, 0.453, 0.643, 0.670, 0.601, 0.603 (varies by column).
- Interpretation provided in text:
  - In both OLS and GMM, the FRRC dummy is negative and significant.
  - FR dummy results are more nuanced and less consistently significant than FRRC.
  - The effect of the presence of a fiscal rule with robust correction mechanism ranges from about 17% to 32% across specifications.
  - Parsimonious specification with time fixed effects (column 10): FRRCs are associated with a total long-run drop of about 20% in the sovereign spread (exp(-0.103-0.120) – 1 ≈ -20%).
  - The marginal effect of the presence of a robust correction mechanism ranges from about 12% to 20% across specifications.

### Robust patterns across methods
- Synthetic control analysis: baseline one-year median spread decline of about 25% (≈75 bps for average treated country with pretreatment spread of 300 bps), statistically significant by placebo testing.
- Panel regressions: FRRC dummy consistently negative and significant; estimated long-run declines in spread consistent with synthetic-control magnitudes (roughly in the two-digit percent range).
- Other determinants consistent with literature:
  - Lower growth rates, higher current account deficits, higher global uncertainty (VIX), and higher global interest rates (U.S. Federal Fund Rate) are associated with higher spreads.
  - Lagged debt and reserves not statistically significant in most specifications.

### Overall interpretation and policy-relevant findings
- Fiscal rules with robust correction mechanisms (FRRC) are associated with economically important, statistically significant, and robust reductions in sovereign spreads for more than a year after introduction.
- Quantitative summary emphasized:
  - One-year median spread decline (synthetic control): about 25% or 75 bps for the average treated country (pretreatment spread 300 bps).
  - Panel estimates imply long-run reductions in spreads in the range of roughly 17%–32% across specifications, with parsimonious specification showing ≈20% long-run drop.
  - Marginal effect of robust correction mechanism relative to other fiscal rules: about 12%–20% reduction in spreads across specifications.
- Caveats and complementary considerations:
  - FRRCs are not panaceas: other rule characteristics (coverage and exclusions, calibration of the fiscal anchor, medium-term fiscal frameworks, institutions for fiscal oversight) matter for overall performance.
  - After extraordinary shocks (e.g., COVID-19), discretion may be applied via escape clauses to deviate from FRRC-prescribed consolidation.
  - Sovereign spreads are volatile at high frequency; governments should not mechanically respond to high-frequency spread movements when guiding fiscal policy.

*Source: 3.2 Results, wpiea2025195-source-pdf*

### References

### References and Appendix (Excerpt)

### Major referenced topics and literature
- Methodologies and applications:
  - Synthetic control methods and comparative case studies: Abadie (2021); Abadie, Diamond, and Hainmueller (2010, 2015); Ferman, Pinto, and Possebom (2020).
  - Difference and system GMM, finite sample corrections: Roodman (2009); Windmeijer (2005).
  - Numerical methods and policy communication: Leeper (2010); End and Hong (2022).
- Fiscal rules, credibility, and compliance:
  - Fiscal rules at-a-glance surveys and databases: Lledó et al. (2017); Alonso et al. (2025); Caselli et al. (2022); Larch and Santacroce (2020); Larch, Malzubris, and Santacroce (2023).
  - Heterogeneous effects, Maastricht criterion, and empirical strategies: Caselli and Reynaud (2020); Caselli and Wingender (2021).
  - Determinants and compliance with fiscal rules: Reuter (2019); Ulloa-Suárez (2023); Ulloa-Suárez and Valencia (2022); Ulloa-Suarez, Valencia, and Guerra (2025).
  - Fiscal rules, stability culture, and sovereign risk premia: Heinemann, Osterloh, and Kalb (2014); Feld et al. (2017); Kalan, Popescu, and Reynaud (2018); Islamaj, Samano Penaloza, and Sommers (2024).
- Sovereign default, spreads, and borrowing costs:
  - Sovereign default risk, borrowing duration, and employment cost: Arellano, Bai, and Bocola (2023); Chatterjee and Eyigungor (2012); Hatchondo and Martinez (2009); Balke (2023); Lang, Mihalyi, and Presbitero (2023).
  - Fiscal stimulus, sovereign risk, and welfare costs: Bianchi, Ottonello, and Presno (2023); Aguiar, Amador, and Fourakis (2020); Roldán (2025); Cuadra, Sánchez, and Sapriza (2010).
- Fiscal governance and European Union context:
  - EU fiscal rules, compliance, and numerical policy: Blanchard and Zettelmeyer (2023); Iara and Wolff (2014); Larch and van der Wielen (2024); Capraru, Georgescu, and Sprincean (2025).
- Other relevant empirical topics:
  - Natural disasters and growth: Cavallo et al. (2013).
  - Business cycles in emerging economies and interest rates: Neumeyer and Perri (2005).
  - Policy instruments and optimal discretion: Halac and Yared (2014, 2018); Sublet (2023).

### Appendix — Charts and Tables (selected content)
- Country composition of the synthetic control (Table A1 — country and donor weights):
  - Armenia: El Salvador (14%), Ghana (14%), Qatar (14%), Ecuador (8%), Suriname (3%), Ethiopia (3%), Other countries (45%)
  - Costa Rica: Tunisia (84%), China (10%), Albania (6%)
  - Cyprus: Belarus (100%)
  - Czech Republic: Guatemala (36%), Slovak Republic (23%), Finland (15%), Luxembourg (13%), Canada (12%), Slovenia (1%)
  - Poland: Czech Republic (22%), Italy (20%), Spain (15%), Argentina (9%), South Africa (5%), Portugal (3%), Other countries (26%)
  - Slovak Republic: Estonia (77%), Ukraine (23%)

- Country-specific p-values after 1, 2, 12, and 52 months (Table A2):
  - Month columns correspond to p-values for: Armenia, Costa Rica, Cyprus, Czech Republic, Poland, Slovak Republic
  - Row entries (preserving formatting):
    - 1: 0.07 0.00 0.02 0.00 0.00 1.00
    - 2: 0.04 0.08 0.02 0.00 0.00 1.00
    - 12: 0.15 0.06 0.00 0.07 0.17 1.00
    - 52: 0.22 0.17 0.00 0.07 0.08 0.26

*Fiscal Rules, Robust Correction Mechanisms, and Sovereign Spreads — Working Paper No. WP/2025/195*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025195-source-pdf.pdf_