## wpiea2021121-print-pdf — Appendix C and References

## Source details

**Canonical URL:** [wpiea2021121-print-pdf — Appendix C and References](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021121-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021121-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021121-print-pdf.pdf.json)

---

### I. Introduction — scope and key advances
- Fiscal austerity defined as reducing deficits by cutting expenditures or raising revenues.
- Conventional wisdom: deficit reduction via tax hikes or spending cuts is punished electorally.
- Key advances of this study:
  - Uses a real-time narrative dataset of ex-ante austerity plans to (i) include episodes in which the budget worsened despite austerity while excluding those in which it improved because of a booming economy; (ii) identify the precise timing of each austerity plan to match it to the government that introduced it; and (iii) distinguish between tax hikes and expenditure cuts.
- Principal empirical findings (summary):
  - Tax-based austerity carries large electoral costs; expenditure-based consolidations are neutral on average.
  - An austerity package worth 1% of GDP, carried out mostly through tax hikes, reduces the vote share of the leader’s party by about 7%.
  - Results hold controlling for growth.
  - Ideology and timing condition effects: right-leaning governments penalized more for tax hikes; expenditure cuts hurt left-leaning governments; costs higher when implemented during bad economic times.
  - Governments avoid announcing austerity close to elections and concentrate plans early in mandates; weak governments tend to implement less austerity — implying baseline estimates may be a lower bound.

### II. Dataset and variable construction
- Sample and period:
  - 16 advanced economies: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, Portugal, Spain, Sweden, the United Kingdom and the United States.
  - Period: 1978-2014.
- Austerity data source and scope:
  - Action-based dataset of Alesina et al. (2019): 3,500 fiscal measures across more than 250 austerity plans.
  - Records budgetary impact (as estimated in accompanying documents) of fiscal measures on total revenues and expenditures relative to a baseline of no policy change.
  - Captures exact month of announcement; multi-year plans; announced-but-not-yet-implemented measures.
- Construction of austerity variables:
  - For each plan, calculate overall budgetary impact expected in real-time over a five-year period as fraction of GDP.
  - Classify plan as tax- or expenditure-based when 55% or more of overall consolidation is via tax hikes or expenditure cuts respectively (plans near 50/50 excluded).
  - Sum all consolidation plans announced during the government term, distinguishing tax- and expenditure-based plans.
  - Main austerity regressors: current and expected future budgetary impact of new tax- and expenditure-based consolidation plans decided during the government’s term (in % of GDP).
  - Robustness checked with 50% and 60% thresholds and alternative constructions.
- Electoral and macro controls:
  - Election dates and government terms from Alesina et al. (2020); party composition from Wikipedia and Doring and Manow (2019).
  - Ideology normalized to range between -10 and 10 (0 = center).
  - Macroeconomic controls from IMF World Economic Outlook (Spring 2019 Edition): real per capita GDP growth, unemployment, budget balance, inflation, 10-year real government bond yield; national disposable income from OECD.
  - Macro variables linearly interpolated to monthly frequency and computed at election month, averaged over government term, and changes over 12 months and term.

### III. Stylized facts — who adopts what and when
- Timing patterns:
  - Governments avoid announcing austerity close to elections; concentrate plans at beginning of mandate.
  - Austerity larger in first year than any other year, driven mostly by tax-based consolidations which are almost four times as large in first vs fourth year of term.
  - Expenditure-based consolidations show no systematic pattern over the term on average.
- Interaction with ideology:
  - Right-leaning governments: expenditure-based consolidations increase close to an election.
  - Left-leaning governments: expenditure-based consolidations gradually decline over the term; tax-based consolidations decline as election approaches for both ideological types.
- Government strength:
  - Strong governments carry out austerity more frequently and adopt larger packages than weak ones.
  - In first year, strong governments carry out austerity packages about 2½ times larger than weak ones (statistically significant at the 95% confidence interval).

### IV. Econometric specifications (overview)
- Baseline regression (Equation (1)):
  - Dependent variable: percent change of the vote share of the coalition supporting the government or the party of the incumbent leader (%VVVVVV_i,e).
  - Main regressors: TTTT_i,e and SSTT_i,e (tax- and expenditure-based austerity plans announced during the term, in % of GDP).
  - Controls: party fixed effects, GDP growth at time of election (Y_i,e), matrix Z_i,e of macroeconomic and political controls.
  - Estimator: OLS with robust standard errors.
- Extensions:
  - Interaction with ideology (Equation (2)): include TTTT * ideology and SSTT * ideology (ideology range -10 to 10).
  - Interaction with business cycle (Equation (3)): transform standardized output growth Z into G(Z) = e^{-γZ} / (1+ e^{-γZ}) with γ = 1.5; construct austerity variables for low- and high-growth states and estimate separate effects.

### V. Main results — electoral effects and magnitudes
- Core estimates (party of incumbent leader; party fixed effects; GDP growth included):
  - Tax-based consolidations: large negative electoral effect.
    - An austerity package worth 1% of GDP, carried out mostly through tax hikes, reduces the vote share of the leader’s party by about 7% (statistically significant at the 99% confidence level).
    - Effect robust to party fixed effects and inclusion of GDP growth; controlling for growth slightly diminishes tax-based austerity effect.
  - Expenditure-based consolidations: statistically insignificant (neutral on average).
- Magnitude interpretation:
  - A tax-based consolidation of about one standard deviation (approximately 1% of GDP) would decrease incumbent party vote share by about ½ standard deviation (approximately 10%).
- Coalition vs leader’s party:
  - Negative effects of tax-based consolidations are slightly lower when dependent variable is entire governing coalition vote share.
  - Coalition subsample: percent change estimates for overall coalition and main party align with baseline; effects for rest of coalition not statistically different from zero.
- Comparison with prior approaches:
  - Using ex-ante matched fiscal data and matching plans to announcing government is crucial: ex-ante but not matched variables halve estimated negative effect; ex-post realized measures can produce a statistically positive effect unless tax/expenditure are distinguished.

### VI. Endogeneity, timing, and government strength
- Timing:
  - Focus on first-year announcements to reduce strategic selection concerns.
  - Tax-based austerity announced in first year has negative and statistically significant electoral effects; effect larger for first-year announcements than for those closer to elections.
- Government strength heterogeneity (restricted samples):
  - For a 1% of GDP tax-based package:
    - Strong governments: estimated reduction in vote share about 5% (not statistically different from zero).
    - Weak governments:
      - If austerity adopted in any year: estimated coefficient about -8% (statistically significant at 99%).
      - If adopted early in term (first year): estimated coefficient about -17% (statistically significant at 99%).
  - Expenditure-based plans have no significant effect on vote share for strong or weak governments.
  - Strong governments: not penalized for tax-based austerity; rewarded for good growth — GDP growth coefficient about 5% for strong governments.
  - Weak governments: punished when economy in recession; GDP growth coefficient ranges between 2% and 3% for weak governments.
- Choice and size of austerity:
  - Augmenting regressions with (i) a 0/1 dummy for any austerity and (ii) magnitude of overall austerity confirms tax-based austerity is more costly than expenditure-based austerity.
- Inverse propensity score weighting (selection adjustment):
  - Procedure follows Jorda and Taylor (2015): classify first-year terms as expenditure, tax, or no treatment; drop consolidations above 90th percentile to match sizes (average around 1.3% of GDP); Probit model uses macro and political covariates; compute inverse propensity weighted differences.
  - Probit marginal effects: expenditure-based plans more likely with higher debt-to-GDP; tax-based predicted by government strength.
  - Average Treatment Effects (Table 9): expenditure-based plans statistically significantly less costly than tax-based; tax-based costlier relative to no consolidation. Example ATEs: ATE Tax vs. expenditure = -11.18** (4.65); Tax vs no plan = -7.02* (4.08).

### VII. Extensions — channels, ideology, and macro conditions
A. The response of the economy to austerity
- Hypothesis: tax-based consolidations induce larger output losses than expenditure-based, potentially explaining electoral differences.
- Augmented model includes average GDP and disposable income growth over term; mean yearly changes in unemployment, inflation, and budget balance.
- Focus on first-year austerity to allow macro effects to materialize.
- Results (Table 10):
  - Control coefficients: positive for GDP and disposable income growth and fiscal balance; negative for unemployment and inflation.
  - Including fiscal balance and full macro controls only slightly reduces tax-based consolidation effect.
  - Conclusion: macroeconomic response explains only partially the electoral effects of tax-based consolidations.

B. Ideology interactions
- Hypotheses:
  - Tax-based consolidations more negatively perceived when decided by right-wing governments.
  - Expenditure-based consolidations costlier for left-leaning governments.
- Estimation via interactions with ideology (normalized so typical right-leaning government = 1; typical left = -0.7; min -1.3; max 2.1).
- Results (Table 11):
  - Right-leaning governments: more penalized from tax hikes than expenditure cuts.
  - Left-leaning governments: more penalized from expenditure cuts.
  - Tax-based consolidations very costly for right-leaning governments and not statistically significant for left-leaning ones.
  - Expenditure-based consolidations costly for left-leaning governments but beneficial for right-leaning ones.
  - F-test: difference in electoral effects statistically significant for full sample, not for first-year sample.

C. Macroeconomic conditions at announcement
- Use smooth transition function per Auerbach and Gorodnichenko (2012) to separate low vs high growth states.
- Table 12 selected estimates (baseline and growth-state interactions):
  - Baseline (no cycle distinction): Tax = -7.2*** (1.4); Expenditure = 0.2 (1.5).
  - Tax–low state = -6.5*** (1.6); Tax–high state = -8.1 (5.9) in Column 2 (no ideology).
  - Column 3 (with ideology) highlights:
    - Tax–low state = -6.1*** (2.1).
    - Expenditure–high state = 4.8** (2.3); Expenditure–high state*ideology = 8.7*** (2.0).
    - Expenditure–low state = -4.8* (2.7); Expenditure–low state*ideology = 1.1 (3.0).
  - Growth–high state coefficients around 3.0 with varying significance (e.g., 3.3* (1.8) in Column 2).
- Interpretation:
  - Austerity, particularly tax increases, reduces vote share when carried out during bad economic times regardless of ideology.
  - Announcing austerity during good times can increase vote share depending on who implements it: left-leaning governments can gain from raising taxes; right-leaning governments can gain from cutting expenditure; right-leaning governments lose from tax increases even during good times.

D. Austerity and probability of early government fall
- Logistic regression (Equation (4)): dependent dummy = 1 if government falls before last 6 months of legislature; regressors include tax- and expenditure-based consolidation variables and GDP growth in last 12 months.
- Broad result: tax-based consolidations increase probability of early fall; expenditure-based do not on average.
- Magnitude: a 1% of GDP tax-based package increases probability of early government fall by about 8% (statistically significant at 90%).
- Interaction with ideology:
  - Expenditure-based consolidations reduce early-fall probability when implemented by right-leaning parties (Expenditure*Ideology = -8.4* (4.5) in Column 2).
  - Tax-based coefficients not statistically different from zero in augmented specification though signs align with prior findings.
- Growth effect on early fall: Growth reduces probability of early fall — Growth = -4.5** (1.9) in Column 1; -4.9** (2.0) in Column 2.

### VIII. Robustness checks — stability across specifications
- Main conclusion stable: tax-based consolidations carry large electoral costs; expenditure-based consolidations neutral on average.
- Alternative dependent variables (Table C1) show consistent negative tax coefficients:
  - Baseline (% vote share): Tax -7.3*** (1.3); Expenditure 0.4 (1.5); Growth 2.7*** (0.9); Obs. 156; R-squared 0.20.
  - Percentage point change in vote share: Tax -2.6*** (0.5); Expenditure -0.4 (0.4); Growth 1.0*** (0.3).
  - % change in seats share: Tax -8.9*** (1.5); Expenditure 0.3 (1.7); Growth 3.5*** (1.2).
- Alternative explanatory variable constructions (Table C2) robust:
  - 50% threshold: Tax -6.4*** (1.3).
  - 60% threshold: Tax -7.2*** (1.3).
  - 0/1 tax/expenditure dummies: Tax -12.0*** (2.8).
  - Implemented (vs announced) austerity: Tax -9.3*** (1.5).
- Subsample analyses (Table C3):
  - Pre-2008: Tax -6.0** (2.4).
  - 1990-2014: Tax -9.0*** (1.5).
  - Minimum term length and outgoing government restrictions leave tax effect negative and significant.
- Political and macro controls (Tables C4–C5):
  - Including lagged vote change, vote share level, turnout, number of parties, term length, debt level, budget balance, 10-year yield and output gap do not overturn negative tax effect; point estimates vary but remain negative and typically significant (examples: Tax -6.1*** (1.4) with all political controls; Tax -7.7*** (1.8) with macro controls).
- Structural reform controls (Table C6): tax effect remains negative (e.g., Tax -6.8*** (1.5) with all reform controls).
- Strong vs weak government definitions (Tables C7–C8) consistently show larger negative tax effects for weak governments:
  - Strong (all years) examples: Tax -5.9 (4.0) or Tax -4.8*** (1.5) depending on definition/sample.
  - Weak (all years) examples: Tax -8.1*** (1.7) or Tax -11.5*** (3.1).
  - Tax in first year for weak governments ranges from -17.1** (7.0) to -19.8*** (4.1) depending on definition.
- Selection and robustness: excluding one party/country at a time yields results very close to baseline; inverse propensity weighting and Probit predictors confirm robustness of main inference.

### IX. Policy-relevant implications
- Tax-based consolidations impose larger electoral costs and increase risk of premature government termination — political economy constraint on tax-focused consolidation.
- Timing matters: implementing austerity during strong economic conditions mitigates or can reverse electoral costs; sequencing consolidation with favorable conditions reduces political backlash.
- Ideological alignment matters: governments are less punished when policy choices align with ideological platform (expenditure cuts for right, tax hikes less penalized for left), affecting political feasibility.
- Coalition dynamics: leader’s party bears disproportionate electoral costs from tax-based austerity as junior partners can exit to avoid fallout.

*Source: wpiea2021121-print-pdf — Appendix C, Section III–IV extracts, Tables and References.*

### References..............................................................................................................

### wpiea2021121-print-pdf - References..............................................................................................................

### References
- References....................................................................................................................... 49

### Appendix A
- Appendix A. .................................................................................................................... 52

### Appendix B
- Appendix B. .................................................................................................................... 56

*Source: wpiea2021121-print-pdf - References..............................................................................................................*

### Appendix C. ............................................................................................................

### Appendix C.

### I. INTRODUCTION
- Fiscal austerity—reducing deficits by cutting expenditures or raising revenues—is often needed to bring the government debt to sustainable levels.
- Common conventional wisdom: governments that reduce the budget deficit by raising taxes or cutting expenditures are punished at the polls.
- Key advances of this study:
  - Uses a real-time narrative dataset of ex-ante austerity plans to (i) include episodes in which the budget worsened despite austerity while excluding those in which it improved because of a booming economy; (ii) identify the precise timing of each austerity plan to match it to the government that introduced it; and (iii) distinguish between tax hikes and expenditure cuts.
- Principal empirical findings:
  - Tax-based austerity carries large electoral costs, while expenditure-based consolidations are neutral on average.
  - An austerity package worth 1% of GDP, carried out mostly through tax hikes, reduces the vote share of the leader’s party by about 7%.
  - These results hold even controlling for growth.
- Ideology matters:
  - The negative effect of tax-based consolidations is stronger for right-leaning governments.
  - Expenditure-based austerity is detrimental for left-leaning but beneficial for right-leaning governments.
- Timing matters:
  - The electoral cost of austerity—especially tax hikes—is higher (lower) when implemented during bad (good) economic times.
- Endogeneity and strategic behavior:
  - Governments tend to avoid announcing austerity close to elections and concentrate austerity plans at the beginning of their mandate.
  - Weak governments tend to implement less austerity than strong ones.
  - These patterns suggest baseline estimates may be conservative (a lower bound) of true electoral costs.

*Source: Appendix C.*

### II. DATASET
- Sample:
  - Covers 16 advanced economies: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, Portugal, Spain, Sweden, the United Kingdom and the United States.
  - Period: 1978-2014.
- Austerity data:
  - Uses the action-based dataset of Alesina et al. (2019) identifying fiscal consolidation plans via a narrative approach.
  - Dataset covers 3500 different fiscal measures adopted in the context of more than 250 austerity plans.
  - Records the budgetary impact (as estimated in accompanying documents) of fiscal measures on total revenues and expenditures relative to a baseline of no policy change.
  - Advantages:
    - Exact month of announcement of each austerity plan; allows assigning plans to the announcing government.
    - Captures multi-year plans and both immediate and future implementation measures—enables inclusion of announced-but-not-yet-implemented measures that influence voter expectations.
    - Focuses on discretionary measures whose objective is not to influence the real economy—reduces endogeneity to output.
- Construction of austerity variables:
  - For each plan, calculate the overall budgetary impact expected in real-time over a five-year period as a fraction of GDP.
  - Classify plans as tax- or expenditure-based when 55% or more of the overall consolidation effort is via tax hikes or expenditure cuts, respectively (plans near 50/50% excluded).
  - Sum all consolidation plans announced during the government term, distinguishing tax- and expenditure-based plans.
  - Two main austerity regressors: current and expected future budgetary impact of all new tax- and expenditure-based consolidation plans decided during the government’s term.
  - Robustness: results are robust to using 50% or 60% thresholds and to alternative constructions dropping the assumption that tax- and expenditure-based plans are not mutually exclusive within one term.
- Electoral and other data:
  - Election dates, start/end dates of government terms from Alesina et al. (2020).
  - Party/coalition composition from Wikipedia and Doring and Manow (2019).
  - Constructed 'party term' for same governing party/coalition within the same legislature.
  - Vote share and percent change variables from Doring and Manow (2019); ideology measure normalized to range between -10 and 10 (0 = center).
  - Macroeconomic controls from IMF World Economic Outlook (Spring 2019 Edition): real per capita GDP growth, unemployment rate, budget balance, inflation rate, 10-year real government bond yield; national disposable income from OECD.
  - Linearly interpolate macro variables to monthly frequency and compute values at election month, averages over government term, and changes over 12 months and entire term.
  - Economic liberalization indicators from Alesina et al. (2020).

### III. WHO ADOPTS WHAT TYPE OF AUSTERITY AND WHEN (STYLIZED FACTS)
- Timing patterns:
  - Governments tend to avoid announcing austerity close to elections and concentrate austerity plans at the beginning of their mandate.
  - Austerity is larger in the first than any other year of the term, driven mostly by tax-based consolidations which are almost four times as large in the first as in the fourth year of the term.
  - Expenditure-based consolidations do not follow a systematic pattern over the term on average.
- Interaction with ideology:
  - Right-leaning governments: expenditure-based consolidations increase close to an election.
  - Left-leaning governments: expenditure-based consolidations gradually decline over the term; tax-based consolidations decline as election approaches for both left- and right-leaning governments.
- Government strength:
  - Strong governments carry out austerity plans more frequently and adopt larger packages than weak ones.
  - In the first year, strong governments carry out austerity packages that are about 2½ times larger than those adopted by weak ones (statistically significant at the 95% confidence interval).
  - Strength matters for both left- and right-leaning governments, more pronounced for left-leaning and especially for expenditure-based austerity.
- Implication: strategic selection likely at play; results will explicitly address timing and strength.

### IV. ECONOMETRIC SPECIFICATIONS
- Baseline regression (Equation (1)):
  - Dependent variable: percent change of the vote share of the coalition supporting the government or the party of the incumbent leader (%VVVVVV_i,e).
  - Main regressors: TTTT_i,e and SSTT_i,e (tax- and expenditure-based austerity plans announced during the term, in % of GDP).
  - Controls: party fixed effects, GDP growth at time of election (Y_i,e), matrix Z_i,e of macroeconomic and political controls.
  - Estimator: OLS with robust standard errors.
- Extensions:
  - Interaction with ideology (Equation (2)): include TTTT * ideology and SSTT * ideology to allow effects to vary with leader’s party score (range -10 to 10).
  - Interaction with business cycle (Equation (3)): transform standardized output growth Z into G(Z) via logistic function G(Z) = e^{-γZ} / (1+ e^{-γZ}) with γ = 1.5; construct austerity variables for low- and high-growth states (superscripts L and H) and estimate separate effects for austerity announced in low vs. high growth periods.

### V. MAIN RESULTS
- Core estimates (party of incumbent leader; baseline includes party fixed effects and GDP growth):
  - Tax-based consolidations: large negative electoral effect.
    - An austerity package worth 1% of GDP, carried out mostly through tax hikes, reduces the vote share of the leader’s party by about 7% (statistically significant at the 99% confidence level).
    - Effect is robust to party fixed effects and inclusion of GDP growth.
    - Controlling for growth slightly diminishes the tax-based austerity effect, consistent with tax-based austerity having more recessionary effects.
  - Expenditure-based consolidations: statistically insignificant (neutral on average).
- Magnitude interpretation:
  - A tax-based consolidation plan of about one standard deviation (approximately 1% of GDP) would decrease the vote share of the incumbent party by about ½ standard deviation (approximately 10%).
- Coalition vs. leader’s party:
  - Negative effects of tax-based consolidations are slightly lower when the dependent variable is the entire governing coalition vote share, suggesting voters hold the leader’s party more accountable.
  - In coalition subsample, percent change estimates for the overall coalition and the main party align with baseline; effects for the rest of coalition not statistically different from zero.
- Comparison with previous literature:
  - Using ex-ante matched fiscal data is key: estimates using ex-ante but not matched austerity variables halve the estimated negative effect of tax-based consolidations and reduce statistical significance.
  - Using ex-post fiscal variables (years where cyclically adjusted budget improves by more than 1.5% of GDP) and not distinguishing tax vs. expenditure yields a statistically positive effect, consistent with Brender and Drazen (2008); distinguishing tax vs. expenditure with ex-post data shows no significant effect.
  - Conclusion: both use of ex-ante austerity data and matching policies to announcing government are crucial to identify political effects and reconcile conventional wisdom with empirical evidence.
- Robustness:
  - Results robust to excluding single political parties or countries, alternative electoral performance measures, alternative austerity measures, different subsamples, and inclusion of additional political, macroeconomic and structural controls (details in Appendix C).

### VI. ENDOGENEITY
- Timing of announcements:
  - Focus on austerity announced in the first year of the government term to reduce strategic-selection concerns.
  - Results:
    - Tax-based austerity plans have negative and statistically significant electoral effects.
    - The effect of tax hikes is larger for plans announced in the first year than for those announced closer to elections (third year), suggesting baseline estimates may be a lower bound of true electoral cost.
- Strength of government:
  - Stronger governments adopt more and larger austerity packages; weak governments avoid consolidation.
  - The paper examines separate subsamples by government strength (strong vs. weak mandate) to assess heterogeneous effects (analysis continues beyond excerpt).

*Source: Appendix C. wpiea2021121-print-pdf.*

### Section III. We also estimate Equation (1) on the two restricted samples of strong and weak

### Section III

### Differential electoral effects by government strength
- Governments with a strong mandate are much less affected by tax-based austerity than those with a weak one.
- For a tax-based austerity package worth 1% of GDP:
  - Strong governments: estimated reduction in vote share of about 5%, but the effect is not statistically different from zero.
  - Weak governments:
    - If austerity is adopted in any year of the term: estimated coefficient about -8% (statistically significant at the 99% confidence level).
    - If austerity is adopted early in the term (first year): estimated coefficient about -17% (statistically significant at the 99% confidence level).
- Expenditure-based austerity plans do not have any significant effect on the vote share for either strong or weak governments.
- Strong governments:
  - Are not penalized for tax-based austerity.
  - Are rewarded for good growth outcomes and not punished when the economy is in a recession.
  - The coefficient for GDP growth in the electoral year is about 5% for strong governments.
- Weak governments:
  - Are punished when the economy is in recession and are not rewarded in expansions.
  - The coefficient for GDP growth in the electoral year ranges between 2% and 3% for weak governments.
- Results are robust to alternative definitions of government strength and using the level (rather than the change) of the vote share to measure mandate strength produces very similar coefficients.
- Note: since weaker governments tend to refrain from doing austerity (Table 1), the estimated coefficient for the effect of tax-based austerity carried out by weak governments is likely biased upwards (smaller negative effects).

### C. Choice of Doing Austerity
- Two alternative specifications to account for the decision to carry out austerity and for package size:
  - Add a 0/1 dummy indicating whether any austerity is implemented during the government term to Equation (1).
  - Augment Equation (1) by including a variable measuring the magnitude of overall austerity (both tax hikes and expenditure cuts) implemented during the term.
- These additional variables control for the choice of carrying out austerity and for the size of the package; coefficients for tax- and expenditure-based austerity measure marginal effects of composition.
- Additional specification: include two dummies capturing terms with at least one tax- and at least one expenditure-based austerity plan—these control for the choice of adopting a tax- or expenditure-based plan; coefficients report marginal effects of plan size.
- Results (Table 7) confirm that tax-based austerity is more costly than expenditure-based austerity.

### D. Inverse propensity score weighting
- Concern: timing of announcement and type of austerity may be endogenous (timing related to proximity to elections; type related to ideology).
- Approach: adopt an inverse propensity weighted regression adjusted estimator following Jorda and Taylor (2015).
- Steps:
  1. Classify a term as receiving either an expenditure, tax or no treatment depending on whether consolidation in the first year of the term was either tax or expenditure based, if any. (Consolidations above the 90th percentile are dropped to make expenditure- and tax-based plans comparable in size; on average around 1.3% of GDP.)
  2. Estimate a Probit model of the occurrence of tax- and expenditure-based austerity on macroeconomic variables (GDP growth, potential output, inflation rate, debt-to-GDP ratio) and political variables (ideology and strength of the coalition) at the beginning of the government’s term as well as their interaction. (Following Jorda and Taylor (2015), replace extreme values of predicted probability with 0.9 and 0.1.)
  3. Perform a linear fit of the outcome of interest (growth in vote share during the term) on the same controls including treatment variables.
  4. Compute differences in outcomes between treatments by reweighting outcomes with the inverse of the propensity score from the Probit regression.
- Probit results (Table 8): expenditure-based plans are more likely to follow higher levels of debt-to-GDP ratio, while tax-based austerity seems to be predicted by the strength of the government.
- Average treatment effect (Table 9): confirms previous findings
  - Expenditure-based plans are statistically significantly less costly than tax-based ones.
  - Tax-based austerity is significantly costlier relative to years when no consolidation is announced.
  - The coefficients measure the average difference in vote share growth between different treatments; in this case the difference is about 1.3% tax- or expenditure-based plan in the first year of term.

### VII. EXTENSIONS — overview
- Extensions explore potential channels and additional outcomes:
  - (i) the response of the economy to austerity,
  - (ii) the government’s ideology,
  - (iii) the economic conditions at the time of announcement.
- Also examine the effect of austerity on the probability that the government is dissolved before the end of its mandate.

### A. The Response of the Economy to Austerity
- Hypothesis: differential macroeconomic responses to tax- versus expenditure-based consolidations could drive different electoral consequences.
  - Literature indicates tax-based consolidations induce large and relatively persistent output losses, while expenditure-based consolidations usually have only mild, if any, contractionary effects.
- Model augmented with macroeconomic indicators during the government term:
  - Average growth rate of GDP and disposable income over the term.
  - Mean yearly change of the unemployment rate, the inflation rate, and the budget balance over the term.
- Focus: austerity carried out in the first year of the government’s term to allow macroeconomic effects to materialize (literature finds effects lasting over a 5-year window).
- Results (Table 10):
  - Coefficients for control variables have expected signs and are statistically significant at standard confidence levels except for the unemployment rate:
    - Positive for GDP and disposable income growth and for the fiscal balance.
    - Negative for unemployment and inflation.
  - Inclusion of fiscal balance over the term (Column 5) and all macroeconomic variables together (Column 7) only slightly reduces the effect of tax-based consolidations.
  - Conclusion: the response of the economy explains only partially the electoral effects of tax-based consolidations.

### B. Ideology
- Hypotheses:
  - Tax-based consolidations may be more negatively perceived if decided by right-wing governments.
  - Expenditure-based consolidations should be costlier for left-leaning governments.
- Method: interact tax- and expenditure-based austerity variables with the ideology variable and estimate Equation (2); use both overall consolidation variables and first-year variables. Ideology normalized so it takes value 1 for the typical right-leaning government (after normalization, ideology takes value -0.7 for the typical left-leaning government; minimum -1.3 and maximum 2.1).
- Results (Table 11):
  - Right-leaning governments are more penalized from tax hikes than expenditure cuts.
  - Left-leaning governments are more penalized from expenditure cuts.
  - Tax-based consolidations are very costly for right-leaning governments and not statistically significant for left-leaning ones.
  - Expenditure-based consolidations are costly for left-leaning governments but beneficial for right-leaning ones.
  - Point estimates provide suggestive evidence that electoral costs tend to be larger for tax measures introduced by right governments.
  - F-test of the difference in these electoral effects is statistically significant for the full sample, but not for the first year of the government mandate.

### C. Macroeconomic Conditions at the Time of Announcements
- Question: do electoral effects of austerity depend on macroeconomic conditions prevailing at the time of announcement?
- Approach: use a smooth transition function as in Auerbach and Gorodnichenko (2012) to calculate probabilities of being in a low and in a bad state and estimate Equation (3) accordingly.

*Italic source attribution: wpiea2021121-print-pdf - Section III*

### Section IV. Table 12 reports the results. Column 1  shows the  baseline results (Equation 1),

### Section IV — Table 12 and D. Austerity and the Probability of Early Fall of the Government

### Key findings on electoral effects of austerity
- Austerity policies, particularly tax increases, reduce the vote share when carried out during bad economic times, regardless of government ideology.
- Announcing austerity during good economic times can increase the vote share, but the effect depends on “who” and “how”:
  - Left-leaning governments gain votes by raising taxes.
  - Right-leaning governments gain votes by cutting expenditure.
  - Right-leaning governments lose votes when increasing taxes during good economic times.
- Tax-based consolidations are generally more costly at the polls than expenditure-based consolidations, even after controlling for differential effects on growth.

### Austerity and the probability of an early government fall
- Specification: logistic regression (Equation (4)) where the dependent variable is a dummy equal to 1 if the government falls before the last 6 months of the legislature; regressors include tax- and expenditure-based consolidation variables and GDP growth in the last 12 months of the government term.
- Broad result: tax-based consolidations increase the probability that the government falls before the natural end of the legislature; expenditure-based consolidations do not, on average.
- Magnitude: an austerity package worth 1% of GDP, achieved mainly through tax hikes, increases the probability of an early government fall by about 8%. This effect is statistically significant at the 90% confidence level.
- When interacting austerity with ideology:
  - Expenditure-based consolidations reduce the probability of an early fall when implemented by right-leaning parties.
  - For tax-based consolidations, none of the estimated coefficients is statistically different from zero in the augmented specification, although coefficient signs align with prior findings.
- Interpretation: tax-based consolidations both decrease the vote share of the leader’s party and shorten its spell in government. Coalition dynamics: junior coalition partners may pull out when austerity is badly received, leaving the leader’s party as primarily responsible and explaining why the leader’s party is punished at the polls.

### Role of macroeconomic conditions at announcement (Table 12)
- Baseline (no cycle distinction): Tax = -7.2*** (1.4); Expenditure = 0.2 (1.5).
- Accounting for high/low growth states (Equation (3) interaction with 1- G(Z) or G(Z)):
  - Tax–low state = -6.5*** (1.6); Tax–high state = -8.1 (5.9) in Column 2 (no ideology).
  - In Column 3 (with ideology):
    - Tax–low state = -6.1*** (2.1).
    - Tax–high state*ideology = -12.7 (8.6) [not reported as statistically significant].
    - Expenditure–high state = 4.8** (2.3).
    - Expenditure–high state*ideology = 8.7*** (2.0).
    - Expenditure–low state = -4.8* (2.7).
    - Expenditure–low state*ideology = 1.1 (3.0).
  - Growth–high state coefficients: 3.3* (1.8) in Column 2; 3.0 (1.8) and 2.4 (1.9) in Columns 2 and 3 respectively for related interactions.

### Ideology and heterogeneous electoral responses (Table 11 and related text)
- Baseline tax effect (no ideology): Tax = -7.6*** (1.3).
- Interaction results (Column 2, any year):
  - Tax = -6.0*** (1.3)
  - Tax*ideology = -5.3*** (1.7)
  - Expenditure = -0.6 (1.0)
  - Expenditure*ideology = 4.8*** (1.5)
- Interpretation consistent with model: parties are punished when they implement austerity measures further from their platform (expenditure cuts for left, tax hikes for right) in a voter population whose median is the tax-payer.

### Timing, mandate strength, and the electoral cycle
- Austerity timing matters:
  - Tax (all years) = -7.3*** (1.3) (Table 5 Column 1).
  - Tax in 1st year: -8.0* (4.3) to -10.8*** (4.0) depending on sample restrictions.
  - Tax in 3rd year: -6.3** (2.3) / -6.3** (2.6).
  - Expenditure effects are generally small and not statistically significant across timing specifications.
- Government mandate strength:
  - Baseline Tax all years = -7.3*** (1.3).
  - Strong govts (all years): Tax = -5.5 (3.8).
  - Weak govts (all years): Tax = -8.2*** (1.5) when estimated in restricted samples for first-year announcements; strikingly, Tax 1st year for weak governments = -17.6*** (6.2) in Table 6 Column 5.
  - Conclusion: governments with a stronger majority are less punished by voters when engaging in tax-based austerity.

### Robustness and controls (economic response and selection)
- Controlling for macroeconomic developments (Table 10, using first-year austerity variables):
  - Column 1 baseline (no controls) — Tax = -13.0*** (4.3) in a variant that uses first-year variables and different outcome definitions (multiple columns reported different specifications).
  - Including controls sequentially (growth, unemployment, inflation, budget change, disposable income) does not overturn the negative tax effect; point estimates vary but remain negative and often significant:
    - Example: Tax = -12.1*** (4.1) when controlling for average GDP growth over term.
    - Fiscal budget change shows a positive coefficient for its own effect: 5.4*** (1.4) and 4.3*** (1.5) in Columns where budget change is included.
- Selection into austerity and propensity adjustments:
  - Table 8 shows predictors of plan type in first year (Probit marginal effects multiplied by 100); notable: Debt-to-gdp ratio, at start of term = 0.19** (0.10) for expenditure-based; Vote share = 0.56* (0.30) for tax-based.
  - Table 9 (Inverse propensity weighted ATE for plans in first year): ATE Tax vs. expenditure = -11.18** (4.65); Expenditure vs no plan = 4.16 (2.59); Tax vs no plan = -7.02* (4.08).

### Summary of key estimated effects (selected table highlights)
- Table 2 (electoral effects, baseline):
  - Tax (party) = -7.1*** (1.6) in Column 1; Tax (party, with party FE) = -8.2*** (1.4) in Column 2.
  - Expenditure (party) = -0.1 (1.4) to 0.7 (1.6); not statistically significant.
  - Growth effect: 2.8*** (0.9) and 2.6*** (0.9) in specifications with growth.
- Table 3 (coalition governments):
  - Tax (overall coalition) = -6.4*** (1.8).
  - Tax (main party) = -8.7*** (2.2); Tax (rest of coalition) = -0.4 (4.6).
- Table 4 (different austerity variable constructs):
  - Tax (ex-ante matched) = -7.3*** (1.3) in Column 2 (when distinguishing tax/expenditure).
  - Ex-post realized measures can produce different point estimates (some positive), underscoring importance of ex-ante matching.
- Table 7 (controlling for choice of doing austerity):
  - Baseline Tax = -7.3*** (1.3); when controlling for dummy tax-based plans, Dummy tax-based = -7.6* (4.1).
- Table 12 (macroeconomic state interactions — selected numbers reiterated):
  - Tax–low state = -6.5*** (1.6).
  - Expenditure–high state*ideology = 8.7*** (2.0).
- Table 13 (early fall probability, Equation (4)):
  - Tax = 8.1* (4.8) in Column 1 (no ideology).
  - In Column 2 (with ideology): Tax = 4.0 (7.5); Expenditure*Ideology = -8.4* (4.5).
  - Growth reduces probability of early fall: Growth = -4.5** (1.9) in Column 1; -4.9** (2.0) in Column 2.

### Policy-relevant implications
- Tax-based consolidations impose larger electoral costs and increase the risk of premature government termination, suggesting political economy constraints on tax-focused fiscal consolidation.
- Timing matters: implementing austerity during strong economic conditions mitigates or can reverse electoral costs; hence, sequencing fiscal consolidation with favorable economic conditions can reduce political backlash.
- Ideological alignment matters: governments are less punished when policy choices align with their ideological platform (expenditure cuts for right, tax hikes less penalized for left), implying that political feasibility of consolidation depends on ideological congruence.
- For coalition governments, the leader’s party bears disproportionate electoral costs from tax-based austerity, partly because junior partners can exit coalitions to avoid electoral fallout.

*Italic source attribution: Section IV, Tables 12–13 and related text from wpiea2021121-print-pdf.*

### REFERENCES

### REFERENCES

### Bibliographic entries
- Provides a numbered list of references cited in the work, including (but not limited to) the following works and authors exactly as listed:
  - Alesina A. & Furceri, D. & Ostry, J. D. & Papageorgiou, C. & Quinn, D. P. 2020. "Structural Ref orms and Elections: Evidence f rom a World-Wide New Dataset," NBER Working Papers 26720.
  - Alesina A. & Azzalini, G. & Favero, C. & Giavazzi, F. & Miano, A. “2018. "Is it the “How” or the “When” that Matters in Fiscal Adjustments?," IMF Economic Review, Palgrave Macmillan; International Monetary Fund, vol. 66(1), pages 144-188, March.
  - Alesina, A., Barbiero, O., Favero, C., Giavazzi, F. and Paradisi, M., 2015. Austerity in 2009–13. Economic Policy, 30(83), pp.383-437.
  - Alesina, A.F., Carloni, D. and Lecce, G., 2011. The Electoral consequences of large f iscal adjustments (No. w17655). National Bureau of Economic Research.
  - Alesina, A., Favero, C. and Giavazzi, F., 2019. Austerity: When it Works and when it Doesn't. Princeton University Press.
  - Alesina, A. and Passalacqua, A., 2016. The political economy of government debt. In Handbook of Macroeconomics (Vol. 2, pp. 2599-2651). Elsevier.
  - Alesina, A., Perotti, R., Tavares, J., Obstf eld, M. and Eichengreen, B., 1998. The political economy of fiscal adjustments. Brookings Papers on Economic Activity, 1998(1), pp.197-266.
  - Ardanaz, M., Hallerberg, M. and Scartascini, C., 2019. Fiscal Consolidations and Electoral Outcomes in Emerging Economies: Does the Policy Mix Matter? Macro and Micro Level.
  - Arias, E., & Stasavage, D. (2019). How large are the political costs of f iscal austerity? Journal of Politics.
  - Auerbach, A.J. and Gorodnichenko, Y., 2012. Measuring the output responses to fiscal policy. American Economic Journal: Economic Policy, 4(2), pp.1-27.
  - Battaglini, M., 2011. The political economy of public debt. Annu. Rev. Econ., 3(1), pp.161-189.
  - Battaglini, M. and Coate, S., 2008. A dynamic theory of public spending, taxation, and debt. American Economic Review, 98(1), pp.201-36.
  - Brender, A., 2003. The effect of fiscal performance on local government election results in Israel: 1989–1998. Journal of Public Economics, 87(9-10), pp.2187-2205.
  - Brender, Adi, and Allan Drazen. 2008. "How Do Budget Deficits and Economic Growth Affect Reelection Prospects? Evidence from a Large Panel of Countries." American Economic Review, 98 (5): 2203-20.
  - Devries, M.P., Guajardo, J., Leigh, M.D., and Pescatori, A., 2011. A new action-b ased dataset of f iscal consolidation (No. 11-128). International Monetary Fund.
  - Döring, H. and Manow, P., 2012. Parliament and government composition database (ParlGov). An infrastructure for empirical information on parties, elections and governments in modern democracies. Version, 12(10).
  - Döring, H. and Regel, S., 2019. Party Facts: A database of political parties worldwide. Party Politics, 25(2), pp.97-109.
  - Drazen, A. and Mejía Eslava, M.E., 2003. The political business cycle in Colombia on the National and Regional level. Departamento Nacional de Planeación.
  - Drazen, A. and Mejía Eslava, M.E., 2010. Electoral manipulation via voter-friendly spending: Theory and evidence. Journal of development economics 92(1), 39-52.
  - Hübscher, E., 2016. The politics of fiscal consolidation revisited. Journal of Public Policy, 36(4), pp.573-601.
  - Hübscher, E. and Sattler, T., 2017. Fiscal consolidation under electoral risk. European Journal of Political Research, 56(1), pp.151-168.
  - Imbens, Guido W. 2004. Nonparametric Estimation of Average Treatment Effects Under Exogeneity: A Review. Review of Economics and Statistics 86(1): 4–29
  - Jones, M.P., Meloni, O. and Tommasi, M., 2012. Voters as fiscal liberals: Incentives and accountability in f ederal systems. Economics & Politics, 24(2), pp.135-156.
  - Jordà, Òscar, and Alan M. Taylor. "The time for austerity: estimating the average treatment ef f ect of f iscal policy." The Economic Journal 126.590 (2016): 219-255.
  - Kreif, Noemi, Richard Grieve, Rosalba Radice, and Jasjeet S. Sekhon. 2013. Regression-Adjusted Matching and Double-Robust Methods f or Estimating Average Treatment Ef f ects in Health Economic Evaluation. Health Services and Outcomes Research Methodology 13(2–4): 174–202
  - Lunceford, Jared K., and Marie Davidian. 2004. Stratification and Weighting via the Propensity Score in Estimation of Causal Treatment Ef fects: A Comparative Study. Statistics in Medicine 23(19): 2937–60
  - Passarelli, F. and Tabellini, G., 2017. Emotions and political unrest. Journal of Political Economy 125(3), pp.903-946.
  - Peltzman, S., 1992. Voters as fiscal conservatives. The Quarterly Journal of Economics 107(2), pp.327-361.
  - Ponticelli, J. and Voth, H.J., 2020. Austerity and anarchy: Budget cuts and social unrest in Europe, 1919–2008. Journal of Comparative Economics 48(1), pp.1-19.
  - Powell, G. Bingham, and Guy D. Whitten. 1993. “A Cross-National Analysis of Economic Voting: Taking Account of the Political Context.” American Journal of Political Science 37(2): 391–414.
  - Ramey, V.A. and Zubairy, S., 2018. Government spending multipliers in good times and in bad: evidence from US historical data. Journal of Political Economy 126(2), pp.850-901.
  - Romer, C.D. and Romer, D.H., 2010. The macroeconomic effects of tax changes: estimates based on a new measure of fiscal shocks. American Economic Review 100(3), pp.763-801.
  - Weingast, B.R., Shepsle, K.A. and Johnsen, C., 1981. The political economy of benefits and costs: A neoclassical approach to distributive politics. Journal of Political Economy 89(4), pp.642-664.
  - Wooldridge, Jeffrey M. 2007. Inverse Probability Weighted M-Estimation f or General Missing Data Problems. Journal of Econometrics 141(2): 1281–1301.

### Appendix A — Description of Alesina et al. (2019) Austerity Dataset
- Overview:
  - The original dataset records fiscal adjustment plans announced between 1978 and 2014 in 16 OECD countries (Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, Portugal, Spain, Sweden, the United Kingdom, and the United States).
  - Fiscal plans are composed by a collection of measures. For each measure, the year of implementation, the budgetary impact, a broad description and classification of the type of measure are provided.
  - Measures are classified into 27 categories; the main analysis aggregates measures into two coarse categories: measures that change expenditure versus those that change taxes.
  - Measures announced at year t can be scheduled to be implemented in the same year or in the following 5.
  - Example: Table C2 provides an example of measures announced by Canada in 2011, to be implemented in the same year and in the subsequent ones (magnitudes are expressed in billions of Canadian dollars).
- Construction of independent variables:
  - For each announcement date, sum all measures announced in that period to be implemented immediately or in the future.
  - Classify the whole announcement as expenditure- or tax-based depending on whether announced changes in taxes are larger than changes in expenditures.
- Sources:
  - Fiscal measures are extracted from a variety of sources. In the earliest part of the sample, the same sources as in Devries et al. (2011) are used, including Budget Reports and Speeches, Stability and Convergence Programmes submitted by EU governments to the European Commission, Central Bank Reports, OECD Economic Surveys, and IMF Reports.
  - The dataset also relies on country specific documents such as reports by the Congressional Budget Office and the Economic Reports of the President for the United States; the Journal Officiel de la Republique Francaise for France; etc.
- Selection of exogenous measures (procedure follows Romer and Romer (2010) and Devries et al. (2011)):
  - Select fiscal adjustments motivated by the aim of improving long-run growth or reducing an inherited deficit; classify adjustments as such only if the deficit-driven adjustment is larger than the long-run growth-driven adjustment.
  - Select a measure only if the plan containing it was introduced with clear sentences attributing the approved measure to correcting dynamics of a budgetary item or addressing the dynamics of the debt-to-GDP ratio, or the deficit.
  - For plans containing both deficit increasing and deficit reducing measures, the following procedure is adopted:
    - Verify that the total amount of exogenous deficit reducing measures is smaller than the total of the deficit increasing measures. In doing this, consider all deficit increasing measures, both exogenous and endogenous.
    - If expansionary measures dominate, then the episode is considered as a fiscal expansion and is dropped.
    - If the sum of (the budgetary impact) of all expansionary measures (endogenous and exogenous) was smaller than the impact of all exogenous contractionary measures (announced at t or previously and yet to be implemented), the episode is recorded in the dataset.
    - The size of the contraction is computed as the difference between the size of the exogenous contractionary measures minus the exogenous expansionary measures.
  - Exceptionally, rare episodes of deficit increasing announcements may appear where previous contractionary announcements were large enough to overcome new expansionary measures; these cases are very rare.

### Appendix B — Construction of electoral variables
- Electoral data sources and rules:
  - Underlying electoral data come from multiple sources. The electoral dataset compiled by Alesina et al. (2020) is used to retrieve the dates of each general election, as well as the start and end dates of each government’s term in office.
  - Exclude interim and technocratic governments, and governments whose main supporting party did not run at the next election. (Footnote: There are five interim governments in our sample (Boeynants and Verhosdaft III in Belgium, Aoki and Ito in Japan and Pintasilgo in Portugal), three technocratic governments (Ciampi, Dini and Monti governments in Italy), and one case in which the main party supporting the government did not run at the next election (the Amato I cabinet in Italy). These governments are excluded.)
  - Construct a variable capturing the period featuring the same governing party (or coalition of parties) within the same legislature. (Footnote: Governing parties are defined as such if they hold Cabinet seats in the government (Ministries).)
  - For parliamentary systems (all sample countries except France and the United States), rely on Wikipedia and Doring and Manow (2019) to reconstruct the party or coalition of parties supporting the government.
  - For presidential systems: for the United States consider presidential elections; in France focus on presidential elections.
  - Treat governments within the same legislature that are not supported by the same parties as belonging to different party terms. Example: Germany 1980–1983 yields two party terms (SPD-FDP term October/1980 to September/1981; CDU-CSU-FDP term September/1981 to March/1983).
  - In cases where different governments within the same legislature are supported by the same parties, record a single party term (example: United Kingdom under Margaret Thatcher and John Major between June/1987 and April/1992).
  - For presidential systems record the party term as the term of the president.
- Electoral performance measures:
  - Rely on Doring and Manow (2019) to collect vote share obtained by the party of the chief executive at the two elections defining the start and the end of the legislature and derive its percent change.
  - Construct additional variables for robustness checks: (i) the vote share change at the previous election, (ii) the percentage point vote share change, (iii) the percent change in parliamentary seats, and (iv) the election turnout.
  - From Doring and Manow (2019) source a variable measuring political leaning (ideology), normalized to range between -10 and 10 (0 for center; higher for more right-leaning; lower for more left-leaning).
- Sample characteristics and descriptive statistics:
  - Dataset contains 157 observations (party-terms), ranging from 5 in France to 14 in Japan.
  - There is at least one austerity announcement in 80 cases—slightly more than 50% of the sample.
  - Average length of government term highest in the United Kingdom: 4.4 years; lowest in Denmark and Japan: 2.5 years.
  - Large cross-country variation in vote share of party of the incumbent leader; Belgium lowest at 17.8%, Spain highest at 42.1%.
  - Number of governing parties: Belgium and Italy largest, 4.7 and 4.6 respectively on average; Canada, Spain and the United Kingdom typically single-party governments.
  - Ideology sample average is 1.4; only Spain has a negative mean.
  - Countries with most right-leaning governments: Japan and Canada (5.2 and 2.5, respectively).
  - The austerity data coverage ends in 2014, so recent coalition governments noted are not included in the sample. (Footnote: The United Kingdom and Spain recently had coalition governments (the Cons.-Lib. Dem. 2010-2015 government in the United Kingdom and the PSOE-Podemos 2020- government in Spain). As our austerity data ends in 2014 these are not included in our sample.)
- Table B1. Descriptive Statistics (reported exactly as in source)
  - Country — Obs. Terms with austerity — Term length — Vote share — # of parties — Ideology
    - Australia 12 4 2.7 41.3 1.4 0.7
    - Austria 10 5 3.3 39.1 1.9 0.4
    - Belgium 10 7 3.3 17.5 4.7 0.2
    - Canada 10 5 3.2 40.4 1.0 2.5
    - Denmark 13 6 2.5 27.7 2.5 1.7
    - Finland 8 3 3.9 24.7 3.8 0.3
    - France 5 4 / 26.4 / 1.6
    - Germany 10 8 3.3 35.5 2.7 0.9
    - Ireland 11 4 2.8 40.7 1.7 1.9
    - Italy 8 6 3.5 30.8 4.6 1.4
    - Japan 14 7 2.5 37.5 2.0 5.2
    - Portugal 11 6 2.8 40.1 1.5 0.7
    - Spain 9 6 3.6 42.1 1.0 -0.1
    - Sweden 11 3 3.2 32.9 2.0 0.1
    - UK 7 4 4.4 41.4 1.0 2.2
    - USA 8 5 / 50.8 / /
    - All sample 157 83 3.1 35.7 2.2 1.4
  - Notes accompanying Table B1:
    - Terms with austerity are defined as government terms in which there is at least one austerity announcement.
    - “Term length” and “vote share” are the average of, respectively, the length of the government term and the vote share of the party of the incumbent leader.
    - “# of parties” indicate the average number of parties in the governing coalition.
    - “Ideology” is the average score in a variable taking more positive (negative) values the more the party of the incumbent leader is right- (left-)leaning. The variable is normalized to take value 1 for the average right-leaning party.
  - Sources: Alesina et al. (2019), Doring and Manow (2019) and own calculations.

*Source: wpiea2021121-print-pdf - REFERENCES*

### APPENDIX C.

### APPENDIX C. Robustness Checks

### Sample stability and exclusions
- Estimates excluding one party at a time and one country at a time produce results "very close to our baseline" (see Figure C1).
- Main finding: tax-based consolidations carry large electoral costs, while expenditure-based ones are neutral on average.

### Alternative dependent variables (Table C1)
- Baseline (% vote share):  
  - Tax: -7.3*** (1.3)  
  - Expenditure: 0.4 (1.5)  
  - Growth: 2.7*** (0.9)  
  - Observations: 156; R-squared: 0.20; Party FE: YES
- Percentage point change in vote share:  
  - Tax: -2.6*** (0.5)  
  - Expenditure: -0.4 (0.4)  
  - Growth: 1.0*** (0.3)  
  - Observations: 156; R-squared: 0.23; Party FE: YES
- % change in seats share:  
  - Tax: -8.9*** (1.5)  
  - Expenditure: 0.3 (1.7)  
  - Growth: 3.5*** (1.2)  
  - Observations: 142; R-squared: 0.23; Party FE: YES
- Percentage point change in seats share:  
  - Tax: -3.9*** (0.7)  
  - Expenditure: -0.4 (0.4)  
  - Growth: 1.5*** (0.5)  
  - Observations: 142; R-squared: 0.25; Party FE: YES

### Alternative explanatory variables (Table C2)
- Baseline (% vote share, 55% threshold):  
  - Tax: -7.3*** (1.3)  
  - Expenditure: 0.2 (1.5)  
  - Growth: 2.8*** (0.9)  
  - Observations: 156; R-squared: 0.20; Party FE: YES
- 50% threshold:  
  - Tax: -6.4*** (1.3)  
  - Expenditure: 0.1 (1.5)  
  - Growth: 2.7*** (0.9)  
  - Observations: 156; R-squared: 0.19
- 60% threshold:  
  - Tax: -7.2*** (1.3)  
  - Expenditure: -0.4 (1.6)  
  - Growth: 2.9*** (0.8)  
  - Observations: 156; R-squared: 0.20
- 0/1 tax/expenditure dummies:  
  - Tax: -12.0*** (2.8)  
  - Expenditure: 2.2 (3.6)  
  - Growth: 3.2*** (0.8)  
  - Observations: 156; R-squared: 0.20
- Collapsed announcements pooled over term:  
  - Tax: -4.0** (1.7)  
  - Expenditure: -1.5 (1.4)  
  - Growth: 3.0*** (0.9)  
  - Observations: 156; R-squared: 0.15
- Implemented austerity (rather than announced):  
  - Tax: -9.3*** (1.5)  
  - Expenditure: 1.5 (1.7)  
  - Growth: 2.8*** (0.9)  
  - Observations: 156; R-squared: 0.22

### Subsample analyses (Table C3)
- Full sample (baseline): Tax -7.3*** (1.3); Exp. 0.4 (1.5); Growth 2.7*** (0.9); Obs. 156; R^2 0.20
- Up to 2008 (pre-GFC): Tax -6.0** (2.4); Exp. 0.6 (2.0); Growth 2.4* (1.2); Obs. 133; R^2 0.12
  - Note: tax-based consolidations estimated to have smaller negative effects (about -6% rather than -7%).
- Post-1990 (1990-2014): Tax -9.0*** (1.5); Exp. -1.6 (1.2); Growth 3.4** (1.3); Obs. 98; R^2 0.30
  - Note: coefficient more negative at -9% for 1990-2014 subsample.
- Minimum term 12 months: Tax -7.5*** (1.5); Exp. 0.6 (1.5); Growth 3.0*** (0.9); Obs. 145; R^2 0.21
- Minimum term 36 months: Tax -7.2*** (1.4); Exp. -0.5 (1.0); Growth 1.8*** (0.6); Obs. 134; R^2 0.19
- Outgoing governments: Tax -7.5*** (1.3); Exp. 0.3 (1.5); Growth 2.6*** (0.9); Obs. 149; R^2 0.20
- Parliamentary systems (excluding U.S. and France): Tax -7.2*** (1.3); Exp. 0.6 (1.6); Growth 2.4** (0.9); Obs. 143; R^2 0.20

### Political control variables (Table C4)
- Baseline (no controls): Tax -7.3*** (1.3); Expenditure 0.4 (1.5); Growth 2.7*** (0.9); Obs. 156; R^2 0.20
- Lagged vote share change included: Tax -7.0*** (1.4); Lagged vote share change: -0.2 (0.1); Obs. 153; R^2 0.22
- Vote share level included: Tax -6.6*** (1.3); Vote share: -1.4*** (0.3); Obs. 156; R^2 0.35
- Turnout included: Tax -7.3*** (1.3); Turnout: -0.0 (0.4); Growth 2.2** (0.9); Obs. 142; R^2 0.20
- Number of parties included: Tax -7.3*** (1.3); # of parties: 2.6 (2.2); Growth 2.6*** (0.9); Obs. 156; R^2 0.21
- Term length (months) included: Tax -6.3*** (1.3); Term length: -0.3*** (0.1); Growth 2.6*** (0.8); Obs. 156; R^2 0.25
- All political controls together: Tax -6.1*** (1.4); Expenditure 0.0 (1.5); Growth 2.3** (0.8); Obs. 139; R^2 0.40

### Macroeconomic control variables (Table C5)
- Baseline (no control): Tax -7.3*** (1.3); Expenditure 0.2 (1.5); Growth 2.8*** (0.9); Obs. 156; R-squared 0.20
- Debt level at start term: Tax -7.0*** (1.4); Debt level coefficient: 0.1** (0.0); Obs. 156; R-squared 0.21
- Budget balance at start term: Tax -7.2*** (1.4); Fiscal balance: -0.7 (0.5); Obs. 156; R-squared 0.21
- 10-year real yield level at start term: Tax -8.3*** (1.9); 10-year real yield: -0.5 (1.1); Obs. 150; R-squared 0.21
- 10-year real yield change during term: Tax -8.0*** (1.7); 10-year real yield change: -1.0 (1.1); Obs. 150; R-squared 0.20
- All previous macro controls together: Tax -7.7*** (1.8); Obs. 150; R-squared 0.23
- Output gap at election: Tax -8.1*** (1.4); Output gap coefficient: 0.4 (0.5); Obs. 157; R-squared 0.12
- Unemployment rate at election: Tax -8.4*** (1.8); Unemployment rate: -1.1** (0.5); Obs. 154; R-squared 0.15

### Structural reform controls (Table C6)
- Baseline (no controls): Tax -7.3*** (1.3); Expenditure 0.2 (1.5); Growth 2.8*** (0.9); Obs. 156; R-squared 0.20
- Capital account reform included: Tax -6.4*** (1.3); KA reform: -0.0 (0.1); Obs. 142; R-squared 0.20
- Trade tariff reform included: Tax -7.3*** (1.3); Tariff reform: 0.0 (0.9) or -1.4** (0.7) in combined specification; Obs. 156; R-squared 0.20
- Domestic finance reform included: Tax -7.4*** (1.4); Finance reform: 0.6 (0.4); Obs. 156; R-squared 0.22
- Labor market reform included: Tax -7.3*** (1.3); Labor market reform: 0.0 (0.3) or -0.6 (0.4) in combined specification; Obs. 156; R-squared 0.20
- All reform controls together: Tax -6.8*** (1.5); Expenditure -0.2 (1.7); Growth 2.2*** (0.7); Obs. 142; R-squared 0.22

### Strong vs. weak governments and timing (Table C7 and Table C8)
- Table C7 (definition: strong/weak by change in vote share at election that brought government into power)
  - Baseline tax effect: -7.3*** (1.3); Obs. 156; R-squared 0.20
  - Strong governments (all years): Tax -5.9 (4.0); Obs. 76; R-squared 0.30
  - Weak governments (all years): Tax -8.1*** (1.7); Obs. 77; R-squared 0.37
  - Tax in first year (weak govts): -17.1** (7.0)  
  - Expenditure in first year (weak govts): 5.7* (3.4)
  - Growth coefficients vary by expansions/recessions and by government strength (see tables for exact values).
- Table C8 (definition: strong/weak by level of vote share at election that brought government into power)
  - Baseline tax effect: -7.3*** (1.3); Obs. 156; R-squared 0.20
  - Strong governments (all years): Tax -4.8*** (1.5); Obs. 76; R-squared 0.27
  - Weak governments (all years): Tax -11.5*** (3.1); Obs. 80; R-squared 0.31
  - Tax in first year (weak govts): -19.8*** (4.1)
  - Expenditure in first year (weak govts): 4.6 (6.6)  
  - Growth remains positive and significant in most specifications (exact coefficients reported in tables).

### Overall conclusion from robustness checks
- Across multiple robustness checks—excluding parties/countries, alternative dependent variables, alternative explanatory variables and thresholds, subsamples, political and macroeconomic controls, structural reform controls, and analyses by government strength and timing—the core result holds:
  - Tax-based consolidations are associated with large electoral costs (typical baseline estimate: Tax -7.3***).
  - Expenditure-based consolidations are, on average, neutral (coefficients close to zero and not statistically significant).

*Source: APPENDIX C. Robustness Checks (wpiea2021121-print-pdf - APPENDIX C.)*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021121-print-pdf.pdf_