## wpiea2020274-print-pdf

## Source details

**Canonical URL:** [wpiea2020274-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020274-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020274-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020274-print-pdf.pdf.json)

---

### II. SYNTHETIC DIFFERENCE IN DIFFERENCES AND MODEL SPECIFICATION
- Presents SDD as a two-step data-driven procedure to build counterfactual outcomes for treated observations.
- Two-step overview:
  - Step 1: estimate synthetic control weights using the usual SCM algorithm.
  - Step 2: use those weights to recover individual treatment effects via a standard difference in differences estimation.
- SCM setup and objectives:
  - Countries i = 1,...,N over T periods; country 1 is the only treated country in period s < T.
  - Fiscal rule indicator: FR_i ∈ {0,1}.
  - Potential outcomes: Y_{i,t}(FR_i) = { Y_{i,t}(0) if FR_i = 0; Y_{i,t}(1) if FR_i = 1, ∀ t ≥ s }.
  - Treatment effect: τ_{1,t} = Y_{1,t}(1) − Y_{1,t}(0), ∀ t ≥ s; Y_{1,t}(0) not observable.
- SCM formalization:
  - Country-level weights W = (ω_2,...,ω_N) with ω_i ≥ 0 and ∑_{i=2}^N ω_i = 1.
  - Predictor-level weights diagonal matrix V (K×K), positive semidefinite with trace equal to one.
  - Nested minimizations:
    - Ŵ(V) = arg min_{W ∈ W} (X_1 − X_0 W)' V (X_1 − X_0 W)    (Eq. (1))
    - V̂ = arg min_{V ∈ V} (Y_1 − Y_0 Ŵ(V))' (Y_1 − Y_0 Ŵ(V))    (Eq. (2))
  - Simplified SCM using pre-treatment outcome lags:
    - Ŵ = arg min_{W ∈ W} (Y_1 − Y_0 W)' (Y_1 − Y_0 W)    (Eq. (3))
  - Counterfactual and estimated effect:
    - Ŷ_{1,t}(0) = ∑_{i=2}^N ω̂_i Y_{i,t}, ∀ t ≥ s.
    - τ̂_{1,t} = Y_{1,t} − ∑_{i=2}^N ω̂_i Y_{i,t}, ∀ t ≥ s.    (Eq. (4))
- Practical guidance and baseline choices:
  - Use all pre-treatment outcome lags as predictors to reduce specification search.
  - Baseline V chosen so pre-treatment outcomes are matched on lags of the government balance with no additional controls.
  - Pre-treatment calibration window: as long as data allow but not longer than 10 years.
  - Post-treatment window in baseline: 10 years after introduction of the fiscal rule.
  - Maximum sample window per treated country: at most 21 years in total.
  - Donor pool restricted to countries with complete budget balance data over same horizon and that did not introduce a fiscal rule within this horizon.
  - Baseline donor restriction: donor countries must be in the same or adjacent World Bank income class group.
- Additional preprocessing:
  - De-mean government balance data by subtracting pre-treatment country average to account for country-specific and time-invariant factors.
  - If selection into treatment correlated only with time-invariant common factors, the de-trended synthetic control estimator is unbiased.
- Advantages: transparency in constructing counterfactuals; chosen weights explicitly show each control unit's contribution.

### B. Weighted Difference in Differences
- Second step: use estimated SCM weights in conventional difference-in-differences via weighted least squares with year fixed effects.
- Estimation objective:
  - (μ̂, β̂, τ̂_{1,t}) = arg min_{μ,β,τ_{1,t}} ∑_{i=1}^N ∑_{j=−10}^{10} (Y_{i,t} − μ − β_t − FR_i τ_{1,t})^2 ω̂_i    (Eq. (5))
    - β_t are year fixed effects.
    - FR_i = 1 for treated country.
- Inference:
  - Under standard assumptions, SCM weight estimation uncertainty can be ignored; weights from first step used directly in Eq. (5).
  - Stacking observations across treatment groups allows recovery of moments of the treatment effects distribution via pooled regression.

### III. DATA
- Sample: 151 countries between 1980 and 2018.
- Data sources combined: IMF World Economic Outlook database, historic fiscal balance series from Mauro and others (2015), and IMF Government Finance Statistics.
- Treatment incidence: 53 countries in the sample have adopted a budget balance rule (BBR) with varying numerical targets.
- Appendix Table A.1 lists adopters, year of adoption, numerical target, and coverage (general or central government).
- Country-sample restrictions and exclusions:
  - Restrict to countries with at least one million inhabitants in 2018.
  - Excluded six countries: Cameroon, Chad, Estonia, Indonesia, Mali and Singapore due to data limitations in the early 1990s.
  - India excluded because it had a budget balance rule in place only for one year.

### Individual treatment effects and main empirical results
- Heterogeneous individual effects:
  - Introduction of a budget balance rule improved balances for high-deficit countries but reduced balances among those with smaller deficits or surpluses.
  - The rule moved individual observations toward the value of the target ("magnet effect").
- Average treatment effect (ATE) dynamics:
  - The size of the treatment effect increases gradually in the first six years after introduction before reaching a long-run effect of around 2 percent of GDP.
  - Averaging individual treatment effects across countries with corresponding 95 percent confidence interval (from Eq. (5)) indicates an economically and statistically significant reduction in the fiscal deficit for the average country.
- Distributional evidence:
  - Time paths of the 10th and the 90th percentiles show the actual budget deficit at the lower end is considerably smaller in magnitude than the counterfactual without fiscal rules.
  - Figures (1–3 referenced) support substantial deficit reductions for those at the high-deficit end.

### Quality of synthetic weights and robustness checks
- Pre-treatment fit assessment:
  - Use Adhikari and others (2018) index: ratio of RMSPE to benchmark RMSPE; index values ≥ 0, smaller values indicate better fit.
  - If the index exceeds 1, the counterfactual fails to describe the actual path reasonably well.
  - Distribution of indices are well below 1, suggesting sufficiently high quality of the fit.
  - Alternative formal F-test for pre-treatment differences aligns with the index assessment.
- Donor-weight transparency:
  - Treated units tend to be matched with countries at a similar level of income.
  - Positive weights assigned to more than 40 countries from the pool of donor countries on average, but can be as high as 60 in some cases.
- Robustness exercises:
  - Removing donor-pool income restrictions marginally changes the index distribution; conditioning on income classes generally non-binding for pre-treatment fit quality.
  - Using macroeconomic predictors as covariates confirms magnet effect and positive average treatment effect; constructing synthetic controls using covariates is challenging due to incomplete series.
  - Restricting pre- and post-treatment windows to 5 years yields a distribution of individual treatment effects very similar to baseline.

### Effectiveness heterogeneity by rule design and institutions
- Aggregate descriptive facts:
  - On average, countries in the sample have 3 rules in place.
  - Most frequent combination: budget balance rules (many at supra-national level) together with restrictions on public debt.
  - If a country has adopted a supra-national budget balance rule, it always includes a monitoring and an enforcement mechanism; for national-level BBRs this is true in only half of the cases.
  - Escape clauses are included only in a minority of rules.
- Regression evidence (summary of Tables 3 and 4):
  - Complexity and number of rules:
    - Dummy: 2 numerical rules — 4.17*** (0.81)
    - Dummy: 3 numerical rules — 1.54** (0.65)
    - Dummy: 4 numerical rules — 2.65*** (0.75)
    - Dummy: 5 numerical rules — 2.56** (1.19)
    - Dummy: 6 numerical rules — 1.21 (1.12)
    - Dummy: 7 numerical rules — 1.13 (1.83)
    - Dummy: Debt rule — 2.33*** (0.81)
    - Dummy: BBR at the supranational level — 2.32** (0.80)
    - Observations: 420 in each specification; Standard errors clustered at the country-level; *** p < 0.01, ** p < 0.05, * p < 0.1.
  - Monitoring, enforcement, escape clauses:
    - Dummy: Monitoring outside gov’t at the supranational level — 2.48*** (0.69)
    - Dummy: Monitoring outside gov’t at the national × supranational level — 2.46*** (0.58)
    - Dummy: Formal enforcement at the supranational level — 1.93*** (0.69)
    - Dummy: Formal enforcement at the national × supranational level — 1.95*** (0.47)
    - Dummy: Existence of escape clause — -2.29** (0.86)
    - Observations: 420 in each specification; Standard errors clustered at the country-level; *** p < 0.01, ** p < 0.05, * p < 0.1.
- Key interpretations:
  - Fewer rules positively correlate with larger positive effects; example: fiscal deficits in countries with two rules are lower by almost 4 percent of GDP (benchmark: having only 1 rule).
  - Supra-national monitoring reduces deficit by more than 2 percent of GDP vis-à-vis no monitoring.
  - Escape clauses reduce the effect from around 2percent of GDP without escape clause by 2 percentage points to a zero-effect.

### Key numerical results and statistics (preserved exactly)
- Most rules restrict deficits to not larger than 3 percent or 0 percent of GDP.
- Sample period: 1980 to 2018.
- Pre-/post-treatment dynamics: increases gradually in the first six years, long-run effect of around 2 percent of GDP.
- Exclusions: Cameroon, Chad, Estonia, Indonesia, Mali and Singapore; India excluded for one-year rule.
- Donor-weight facts: positive weights assigned to more than 40 countries on average, up to 60 in some cases.
- Average number of rules per country: 3 rules.
- Maximum number of rules observed in sample: 7.
- Fiscal deficits in countries with two rules: lower by almost 4 percent of GDP.
- Supra-national monitoring reduces deficit by more than 2 percent of GDP vis-à-vis no monitoring.
- Escape clause effect: reduces the effect from around 2percent of GDP by 2 percentage points to a zero-effect.
- Long-run average improvement reported in conclusion: around 3 percent of GDP in the long run after introduction.

### Policy-relevant implications and interpretations
- Fiscal rules have overall potential to improve fiscal performance: average long-run improvement around 3 percent of GDP.
- Heterogeneity matters:
  - Average treatment effects mask substantial country-level heterogeneity (magnet effect pulls deficits and surpluses toward the rule target).
  - High-deficit countries benefit most; countries with surpluses may see reductions in surpluses due to convergence toward the rule target.
- Rule design and institutional context are critical:
  - Fewer overall rules and the existence of a credible and independent monitoring process enhance success of budget balance rules.
  - Supra-national monitoring and enforcement are associated with stronger improvements.
  - Escape clauses undermine effectiveness and can nullify the expected deficit-reduction impact.
- Policy caution:
  - Fiscal rules can trigger unintended consequences (e.g., diminishing fiscal surpluses).
  - Interactions among multiple rules and rule complexity can create potential conflicts and non-compliance risks.

*Source: wpiea2020274-print-pdf - Section 7 concludes.*

### Section 7 concludes.

### Section 7 concludes.

### II. SYNTHETIC DIFFERENCE IN DIFFERENCES AND MODEL SPECIFICATION
- Introduces notation and reviews the SDD estimation: a two-step data-driven procedure to build counterfactual outcomes for treated observations (e.g., introduction of a fiscal rule).
- Use case: particularly well suited when units are aggregate (countries) and it is difficult to find a single suitable counterfactual.
- Two-step overview:
  - Step 1: estimate synthetic control weights using the usual SCM algorithm.
  - Step 2: use those weights to recover individual treatment effects via a standard difference in differences estimation.

### A. Estimating synthetic control weights
- Setup:
  - Countries i = 1,...,N over T periods; country 1 is the only treated country in period s < T.
  - Fiscal rule indicator: FR_i ∈ {0,1}.
  - Potential outcomes: Y_{i,t}(FR_i) = { Y_{i,t}(0) if FR_i = 0; Y_{i,t}(1) if FR_i = 1, ∀ t ≥ s }.
  - Treatment effect: τ_{1,t} = Y_{1,t}(1) − Y_{1,t}(0), ∀ t ≥ s; Y_{1,t}(0) not observable.
- SCM objective: build counterfactual Y_{1,t}(0) as weighted average of control units closest in pre-treatment outcomes.
- Two matching objectives:
  1) Match weighted average of K predictors in control group X_0 to predictors for treated unit X_1.
  2) Match time path of (weighted) average outcome among control units Y_0 to outcome for treated unit Y_1.
- Weight sets:
  - Country-level weights W = (ω_2,...,ω_N) with ω_i ≥ 0 and ∑_{i=2}^N ω_i = 1.
  - Predictor-level weights as diagonal matrix V (K×K), positive semidefinite with trace equal to one.
- Formal nested minimization problems:
  - Objective 1 (given V):
    - Ŵ(V) = arg min_{W ∈ W} (X_1 − X_0 W)' V (X_1 − X_0 W)    (Eq. (1))
  - Objective 2 (choose V to yield Ŵ(V) minimizing outcome distance):
    - V̂ = arg min_{V ∈ V} (Y_1 − Y_0 Ŵ(V))' (Y_1 − Y_0 Ŵ(V))    (Eq. (2))
- Practical guidance and simplification:
  - Risk of specification searching for V and K predictors; recommendation: use all pre-treatment outcome lags as predictors (citing Ferman, Pinto and Possebom, 2020).
  - Simplified SCM when using pre-treatment outcome lags:
    - Ŵ = arg min_{W ∈ W} (Y_1 − Y_0 W)' (Y_1 − Y_0 W)    (Eq. (3))
- Construction of post-treatment counterfactual:
  - Ŷ_{1,t}(0) = ∑_{i=2}^N ω̂_i Y_{i,t}, ∀ t ≥ s.
  - Estimated treatment effect at each time:
    - τ̂_{1,t} = Y_{1,t} − ∑_{i=2}^N ω̂_i Y_{i,t}, ∀ t ≥ s.    (Eq. (4))
- Matching on pre-intervention outcomes can control for heterogeneous responses to multiple unobserved factors when the number of pre-intervention periods is sufficiently large.
- Baseline choices and implementation details:
  - Baseline V chosen so pre-treatment outcomes are matched on lags of the government balance with no additional controls.
  - Robustness checks: alternative covariates in nested SCM.
  - Pre-treatment calibration window: as long as data allow but not longer than 10 years.
  - Post-treatment window in baseline: 10 years after introduction of the fiscal rule.
  - Maximum sample window per treated country: at most 21 years in total.
  - Pool of donor countries restricted to those with complete budget balance data over same horizon and that did not introduce a fiscal rule within this horizon.
  - Baseline donor restriction: donor countries must be in the same or adjacent World Bank income class group (examples: high-income matched to high-income or upper middle-income; low-income matched to low-income or lower middle-income).
- Additional preprocessing:
  - De-mean government balance data by subtracting pre-treatment country average to account for country-specific and time-invariant factors.
  - If selection into treatment correlated only with time-invariant common factors, the de-trended synthetic control estimator is unbiased.
- Advantages of SCM: transparency in constructing counterfactuals; chosen weights explicitly show each control unit's contribution.

### B. Weighted Difference in Differences
- Second step: use estimated SCM weights in a conventional difference-in-differences (DD) estimation via weighted least squares with year fixed effects.
- Estimation objective:
  - (μ̂, β̂, τ̂_{1,t}) = arg min_{μ,β,τ_{1,t}} ∑_{i=1}^N ∑_{j=−10}^{10} (Y_{i,t} − μ − β_t − FR_i τ_{1,t})^2 ω̂_i    (Eq. (5))
    - β_t are year fixed effects.
    - FR_i is indicator equal to one for treated country.
- Inference simplification:
  - Under standard assumptions, estimation uncertainty of SCM weights ω̂_i can be ignored, so weights from first step can be used directly in Eq. (5) without further adjustment.
  - Stacking observations across treatment groups allows recovery of moments of the treatment effects distribution via pooled regression.

### III. DATA
- Sample: 151 countries between 1980 and 2018.
- Data sources combined: IMF World Economic Outlook database, historic fiscal balance series from Mauro and others (2015), and IMF Government Finance Statistics.
- Treatment incidence: 53 countries in the sample have adopted a budget balance rule (BBR) with varying numerical targets.

*Source: wpiea2020274-print-pdf - Section 7 concludes.*

### Appendix Table A.1 lists the rule adopters, year of adoption, the numerical

### wpiea2020274-print-pdf - Appendix Table A.1 lists the rule adopters, year of adoption, the numerical

### Data and sample overview
- Appendix Table A.1 lists the rule adopters, year of adoption, the numerical target of the rule and its coverage (general or central government).
- Most countries have adopted fiscal rules restricting the deficit to levels not larger than 3 percent or 0 percent of GDP.
- Sample period: 1980 to 2018.
- Country-sample restrictions and exclusions:
  - Restrict set of countries to those that have at least one million inhabitants in 2018.
  - Excluded six countries: Cameroon, Chad, Estonia, Indonesia, Mali and Singapore due to data limitations in the early 1990s.
  - India excluded because it had a budget balance rule in place only for one year.

### Individual treatment effects (main empirical results)
- Heterogeneous individual effects:
  - Introduction of a budget balance rule improved balances for high-deficit countries but reduced balances among those with smaller deficits or surpluses.
  - The rule moved individual observations toward the value of the target ("magnet effect").
- Average treatment effect (ATE) dynamics:
  - The size of the treatment effect in the post-treatment period increases gradually in the first six years after the introduction of the rule before reaching a long-run effect of around 2 percent of GDP.
  - Averaging individual treatment effects across countries with corresponding 95 percent confidence interval (from Eq. (5)) indicates an economically and statistically significant reduction in the fiscal deficit for the average country.
- Distributional evidence:
  - Time paths of the 10th and the 90th percentiles show the actual budget deficit at the lower end of the distribution is considerably smaller in magnitude than the counterfactual without fiscal rules.
  - Figure evidence (Figures 1–3 referenced) supports substantial deficit reductions for those at the high-deficit end.

### Quality of synthetic weights and robustness checks
- Pre-treatment fit assessment:
  - Follow Adhikari and others (2018) index: ratio of RMSPE to benchmark RMSPE; index values greater or equal to 0, with smaller values indicating better fit.
  - If the index exceeds 1, the counterfactual fails to describe the actual path reasonably well.
  - Distribution of indices are well below 1, suggesting sufficiently high quality of the fit.
  - Alternative formal F-test for pre-treatment differences aligns with the index assessment (cannot reject hypothesis that all pre-treatment differences are jointly different from zero).
- Donor-weight transparency:
  - Treated units tend to be matched with countries at a similar level of income.
  - Positive weights are assigned to more than 40 countries from the pool of donor countries on average, but can be as high as 60 in some cases.
- Robustness exercises:
  - Removing donor-pool income restrictions marginally changes the index distribution; conditioning on income classes is generally non-binding for pre-treatment fit quality.
  - Using macroeconomic predictors (debt-to-GDP ratio, GDP per capita, inflation rate, commodity terms of trade index, and IMF program indicator) as covariates confirms magnet effect and positive average treatment effect; however, constructing synthetic controls using covariates is challenging due to incomplete series.
  - Restricting pre- and post-treatment windows to 5 years yields a distribution of individual treatment effects very similar to baseline (Figure 10 vs Figure 1).

### Effectiveness of fiscal rules: heterogeneity by rule design and institutions
- Aggregate descriptive facts:
  - On average, countries in the sample have 3 rules in place (national and supra-national).
  - The most frequent combination: budget balance rules (many at supra-national level) together with restrictions on public debt.
  - If a country has adopted a supra-national budget balance rule, it always includes a monitoring and an enforcement mechanism; for national-level BBRs this is true in only half of the cases.
  - Escape clauses are included only in a minority of rules.
- Regression evidence on rule quantity and types (Table 3 referenced):
  - Indicator variables capture total number of rules (up to observed maximum of 7 in sample).
  - Column 1: fewer rules positively correlate with a positive effect of the rule on budget balances; example: fiscal deficits in countries with two rules are lower by almost 4 percent of GDP (benchmark: having only 1 rule).
  - Column 2: existence of other rules generally does not impact the size of the treatment effect of budget balance rules, except debt rules which appear to have a positive effect.
  - Column 3: supra-national BBRs in addition to national rules have a positive impact on treatment effect size.
- Regression evidence on monitoring, enforcement, and escape clauses (Table 4 referenced):
  - Monitoring mechanisms:
    - Improvement of fiscal balance is particularly strong for countries that have monitoring mechanisms at the supra-national level compared to national monitoring or none.
    - Magnitude: those with supra-national monitoring would reduce their deficit by more than 2 percent of GDP vis-à-vis countries with no monitoring at all.
  - Enforcement mechanisms:
    - Countries with enforcement mechanisms at the supra-national level also improve their balances after introducing fiscal rules; effect smaller in magnitude than monitoring.
  - Escape clauses:
    - Existence of an escape clause has a detrimental effect on treatment effect size.
    - Allowing for an escape clause reduces the effect of the fiscal rule from around 2percent of GDP without escape clause by 2 percentage points to a zero-effect.

### Key numerical results and statistics (preserved exactly)
- Most rules restrict deficits to not larger than 3 percent or 0 percent of GDP.
- Sample period: 1980 to 2018.
- Pre-/post-treatment dynamics: increases gradually in the first six years, long-run effect of around 2 percent of GDP.
- Exclusions: six countries explicitly listed (Cameroon, Chad, Estonia, Indonesia, Mali and Singapore); India excluded for one-year rule.
- Donor-weight facts: positive weights assigned to more than 40 countries on average, up to 60 in some cases.
- Average number of rules per country: 3 rules.
- Maximum number of rules observed in sample: 7.
- Fiscal deficits in countries with two rules: lower by almost 4 percent of GDP.
- Supra-national monitoring reduces deficit by more than 2 percent of GDP vis-à-vis no monitoring.
- Escape clause effect: reduces the effect from around 2percent of GDP by 2 percentage points to a zero-effect.
- Long-run average improvement reported in conclusion: around 3 percent of GDP in the long run after introduction.

### Policy-relevant implications and interpretations
- Fiscal rules have overall potential to improve fiscal performance: average long-run improvement around 3 percent of GDP.
- Heterogeneity matters:
  - Average treatment effects mask substantial country-level heterogeneity (magnet effect pulls deficits and surpluses toward the rule target).
  - High-deficit countries benefit most; countries with surpluses may see reductions in surpluses due to convergence toward the rule target.
- Rule design and institutional context are critical determinants of effectiveness:
  - Fewer overall rules and the existence of a credible and independent monitoring process enhance success of budget balance rules.
  - Supra-national monitoring and enforcement are associated with stronger improvements.
  - Escape clauses undermine effectiveness and can nullify the expected deficit-reduction impact.
- Policy caution:
  - Fiscal rules can trigger unintended consequences (e.g., diminishing fiscal surpluses).
  - Interactions among multiple rules and rule complexity can create potential conflicts and non-compliance risks.

*Source: wpiea2020274-print-pdf*

### References

### References

### Key literature and methodological themes
- Synthetic control and related methods: Abadie, A., Diamond, A., & Hainmueller, J. (2010); Abadie, A., & Gardeazabal, J. (2003); Arkhangelsky, D., Athey, S., Hirshberg, D. A., Imbens, G. W., & Wager, S. (2019); Doudchenko, N., & Imbens, G. W. (2016); Kaul, A., Klößner, S., Pfeifer, G., & Schieler, M. (2015); Ferman, B., Pinto, C., & Possebom, V. (2020); Adhikari, B., Duval, R., Hu, B., & Loungani, P. (2018).
- Evaluation and program-treatment literature: Abbring, J. H. & Heckman, J. J. (2007); Heckman, J. J., Smith, J., & Clements, N. (1997); Primo, D. M. (2007); Poterba, J. M. (1994, 1997).
- Fiscal rules, design, and effects: Badinger, H., & Reuter, W. H. (2017); Caselli, F. & Reynaud, J. (2020); Caselli, F. and P. Wingender (2018); Debrun, X., Moulin, L., Turrini, A., Ayuso-i-Casals, J., & Kumar, M. S. (2008); Heinemann, F., M. D. Moessinger, and M. Yeter (2018); International Monetary Fund (2018); Lledó, V., Yoon, S., Fang, X., Mbaye, S., & Kim, Y. (2017); Mauro, P., Romeu, R., Binder, A., & Zaman, A. (2015); Reuter, W. H. (2015).
- Data sources cited: IMF World Economic Outlook Database (April 2019), IMF International Financial Statistics (IFS), World Bank World Development Indicators Database, Monitoring of Fund Arrangements (MONA Database), and Gruss and Kabhaj, 2019; IMF Fiscal Rules Database, 1985-2015; IMF Fiscal Rules Dataset, 2016.

### Summary statistics (Table 1: Main economic variables for Countries with and without Fiscal Rule)
- Government Balance (percent of GDP)
  - Rulers: Obs. 808; Mean -2.17; St. Dev. 4.00; Min -15.14; Max 18.55
  - Non-Rulers: Obs. 3,817; Mean -2.60; St. Dev. 4.66; Min -19.8; Max 19.39
- Population Size (in million inhabitants)
  - Rulers: Obs. 815; Mean 24.72; St. Dev. 35.36; Min 0.67; Max 200.96
  - Non-Rulers: Obs. 4,516; Mean 34.67; St. Dev. 118.75; Min 0.24; Max 1395.38
- GDP per capita (in 1,000 USD, PPP adj.)
  - Rulers: Obs. 815; Mean 24.67; St. Dev. 15.32; Min 0.57; Max 70.03
  - Non-Rulers: Obs. 4,451; Mean 13.41; St. Dev. 18.06; Min 0.28; Max 172.99
- Government Debt (percent of GDP)
  - Rulers: Obs. 811; Mean 61.01; St. Dev. 36.83; Min 0.05; Max 233.72
  - Non-Rulers: Obs. 4,167; Mean 60.79; St. Dev. 60.02; Min 0; Max 2092.92
- Old Age Dependency Ratio (percent of working age population)
  - Rulers: Obs. 809; Mean 18.73; St. Dev. 9.36; Min 4.07; Max 35.59
  - Non-Rulers: Obs. 4,943; Mean 9.8; St. Dev. 5.87; Min 0.80; Max 46.17
- Inflation Rate (annual percentage change)
  - Rulers: Obs. 815; Mean 2.62; St. Dev. 2.84; Min -3.98; Max 25.87
  - Non-Rulers: Obs. 4,422; Mean 260.78; St. Dev. 13991; Min -72.73; Max 929790
- Trade openness (sum of imp. & exp., percent of GDP)
  - Rulers: Obs. 812; Mean 91.79; St. Dev. 58.05; Min 20.72; Max 442.62
  - Non-Rulers: Obs. 4,149; Mean 72.77; St. Dev. 37.72; Min 0.02; Max 376.22
- Short-Term Interest Rate (percent)
  - Rulers: Obs. 691; Mean 3.68; St. Dev. 3.11; Min -0.69; Max 19.92
  - Non-Rulers: Obs. 2,641; Mean 34.2; St. Dev. 827.97; Min -0.78; Max 41280
- Federal State (time-independent dummy)
  - Rulers: Obs. 815; Mean 0.19; St. Dev. 0.40; Min 0; Max 1
  - Non-Rulers: Obs. 5,035; Mean 0.15; St. Dev. 0.35; Min 0; Max 1
- Participation in IMF Program (time-dependent dummy)
  - Rulers: Obs. 815; Mean 0.26; St. Dev. 0.44; Min 0; Max 1
  - Non-Rulers: Obs. 5,035; Mean 0.34; St. Dev. 0.47; Min 0; Max 1
- Commodity Terms of Trade Index
  - Rulers: Obs. 802; Mean 100.06; St. Dev. 6.28; Min 48.56; Max 111.81
  - Non-Rulers: Obs. 4,455; Mean 98.03; St. Dev. 12.60; Min 31.92; Max 130.96

- Data sources: IMF World Economic Outlook Database (April 2019), IMF International Financial Statistics (IFS), World Bank World Development Indicators Database, Monitoring of Fund Arrangements (MONA Database), and Gruss and Kabhaj, 2019.

### Fiscal rule design statistics (Table 2: Rule Designs)
- Number of national rules: Obs. 420; Mean. 0.89; St. Dev. 1.03
- Number of supra-national rules: Obs. 420; Mean. 2.09; St. Dev. 1.14
- Number of rules, national and supra-national: Obs. 420; Mean. 2.97; St. Dev. 1.30
- Dummy, expenditure rule: Obs. 420; Mean. 0.32; St. Dev. 0.47
- Dummy, revenue rule in place: Obs. 420; Mean. 0.21; St. Dev. 0.41
- Dummy, debt rule in place: Obs. 420; Mean. 0.87; St. Dev. 0.33
- Dummy, BBR at the national level only: Obs. 420; Mean. 0.18; St. Dev. 0.38
- Dummy, BBR at the supra-national level only: Obs. 420; Mean. 0.64; St. Dev. 0.48
- Dummy, BBR at both national and supra-national level: Obs. 420; Mean. 0.18; St. Dev. 0.39
- If BBR at supra-national level exists: Dummy, monitoring outside gov’t: Obs. 346; Mean. 1.00; St. Dev. 0.00
- If BBR at national level exists: Dummy, monitoring outside gov’t: Obs. 166; Mean. 0.42; St. Dev. 0.50
- Either BBR at national or supra-national: Dummy, monitoring outside gov’t: Obs. 420; Mean. 0.90; St. Dev. 0.30
- If BBR at supra-national level exists: Dummy, enforcement mech.: Obs. 345; Mean. 1.00; St. Dev. 0.00
- If BBR at national level exists: Dummy, enforcement mech.: Obs. 149; Mean. 0.40; St. Dev. 0.49
- Either BBR at national or supra-national: Dummy, enforcement mech.: Obs. 420; Mean. 0.92; St. Dev. 0.27
- If BBR at supra-national level exists: Dummy, escape clause: Obs. 345; Mean. 0.44; St. Dev. 0.50
- If BBR at national level exists: Dummy, escape clause: Obs. 148; Mean. 0.30; St. Dev. 0.46
- Either BBR at national or supra-national: Dummy, escape clause: Obs. 420; Mean. 0.47; St. Dev. 0.50

- Source: IMF Fiscal Rules Database, 1985-2015.

### Regression results on design and complexity (Table 3 and Table 4)
- Table 3: The Effect of Fiscal Rule Complexity on the Size of the Treatment Effect (OLS; dependent variable: treatment effect from baseline SDD)
  - Dummy: 2 numerical rules — 4.17*** (0.81)
  - Dummy: 3 numerical rules — 1.54** (0.65)
  - Dummy: 4 numerical rules — 2.65*** (0.75)
  - Dummy: 5 numerical rules — 2.56** (1.19)
  - Dummy: 6 numerical rules — 1.21 (1.12)
  - Dummy: 7 numerical rules — 1.13 (1.83)
  - Dummy: Expenditure rule — -0.16 (0.62)
  - Dummy: Revenue rule — -0.95 (0.97)
  - Dummy: Debt rule — 2.33*** (0.81)
  - Dummy: BBR at the supranational level — 2.32** (0.80)
  - Dummy: BBR at the national × supranational level — 0.93 (0.61)
  - Constant: Column (1) -1.10*** (0.19); Column (2) -0.30 (0.61); Column (3) -0.17 (0.41)
  - Observations: 420 in each specification
  - Note: Standard errors clustered at the country-level; *** p < 0.01, ** p < 0.05, * p < 0.1.

- Table 4: The Effect of Fiscal Rules’ Design on the Size of the Treatment Effect (OLS; dependent variable: treatment effect from baseline SDD)
  - Dummy: Monitoring outside gov’t at the national level — 1.03 (0.79)
  - Dummy: Monitoring outside gov’t at the supranational level — 2.48*** (0.69)
  - Dummy: Monitoring outside gov’t at the national × supranational level — 2.46*** (0.58)
  - Dummy: Formal enforcement at the national level — -0.13 (0.75)
  - Dummy: Formal enforcement at the supranational level — 1.93*** (0.69)
  - Dummy: Formal enforcement at the national × supranational level — 1.95*** (0.47)
  - Dummy: Existence of escape clause — -2.29** (0.86)
  - Constant: Column (1) -0.64** (0.30); Column (2) -0.09 (0.35); Column (3) 2.56*** (0.67)
  - Observations: 420 in each specification
  - Note: Standard errors clustered at the country-level; *** p < 0.01, ** p < 0.05, * p < 0.1.

### Figures, fit diagnostics, and robustness notes
- Figure 2: Reports the average treatment effect obtained from a linear combination of country-specific individual treatment effects from the baseline SDD specification and Equation 5. The fiscal rule is introduced in year 0.
- Figure 3: Reports the 10th and the 90th percentiles of the actual budget balance series and the counterfactual balance series obtained from the baseline SDD specification. The fiscal rule is introduced in year 0.
- Figure 4: Shows the distribution of the Adhikari and others (2018) indices based on Equation (6) and (7). The index takes values greater or equal than 0, with smaller values indicating a better fit.
- Figure 5: Reports the weights from the baseline SDD estimation. Treated countries are ranked according to their GDP per capita in 2018 in descending order on the vertical axis, and the corresponding donor countries for each country on the horizontal axis, also ranked according to GDP per capita. Larger weights are displayed in darker red.
- Figure 6: Shows the distribution of the Adhikari and others (2018) indices based on Equation (6) and (7) in a robustness exercise where the set of donor countries is not restricted to countries of the same or adjacent income groups. The index takes values greater or equal than 0, with smaller values indicating a better fit.
- Figures 7–10: Report individual treatment effects and percentile time paths under alternative specifications including covariates and shortened pre- & post-treatment horizons. The fiscal rule is introduced in year 0 in these figures.

### Appendix (Table A.1: Countries included in the Sample and their National Fiscal Rules — selected entries)
- Antigua and Barbuda: Year of adoption 1998; Numerical Target -3; Coverage of FR GG; Included in baseline sample no,  < 1 million population in 2018
- Argentina: Year of adoption 2000; Numerical Target 0; Coverage of FR GG; Included in baseline sample Yes
- Austria: Year of adoption 1995; Numerical Target -3; Coverage of FR GG; Included in baseline sample Yes
- Belgium: Year of adoption 1992; Numerical Target -3; Coverage of FR GG; Included in baseline sample Yes
- Benin: Year of adoption 2000; Numerical Target -3; Coverage of FR CG; Included in baseline sample Yes
- Canada: Year of adoption 1998; Numerical Target 0; Coverage of FR CG; Included in baseline sample Yes
- China entry appears in donor weight matrix listing but not in Table A.1 excerpt above.
- India: Year of adoption 2008; Numerical Target -3; Coverage of FR GG; Included in baseline sample no, FR was only for 1 year in place
- Singapore: Year of adoption 1980; Numerical Target 0; Coverage of FR CG; Included in baseline sample no, FR already in place since 1962 but lack of data
- Sweden: Year of adoption 1995; Numerical Target 1; Coverage of FR GG; Included in baseline sample Yes
- United Kingdom: Year of adoption 1992; Numerical Target -3; Coverage of FR GG; Included in baseline sample Yes

- Source: IMF Fiscal Rules Dataset, 2016.

*Content unit: wpiea2020274-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020274-print-pdf.pdf_