## wpiea2023139-print-pdf

## Source details

**Canonical URL:** [wpiea2023139-print-pdf](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023139-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023139-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023139-print-pdf.pdf.json)

---

### Research question and motivation
- Investigates whether the timing of elections affects the duration between announcement and implementation of tax policy changes, thereby influencing the likelihood of unanticipated tax policy changes (“tax news shocks”).
- Two channels highlighted:
  - Fiscal foresight: longer announcement-to-implementation durations allow agents to adjust behavior.
  - Policy uncertainty: unanticipated tax policy changes can raise uncertainty and hinder economic activity.

### Data and identification
- Sample and period:
  - Monthly data for 22 advanced economies (AEs) and emerging markets (EMs) over 1990-2018.
  - Number of country-month observations: 7,656.
- Primary data source:
  - Tax Policy Reform Database (TPRD) constructed by Amaglobeli et al. (2018).
  - Types of taxes covered: CIT, PIT, VAT, SSC, EXE, PRO.
- Definition of tax news shock:
  - Announcement-to-implementation gap shorter than 90 days.
  - Monthly dummy = 1 if at least one announcement in a country-month has implementation lag < 90 days.
- Sample construction for “major” tax policy changes:
  - Major changes: rate changes > 1 percentage point; base changes that broadly affect groups or mobilize new resources.
  - Total major announcements: 2,736.
  - Announcements with both announcement and implementation dates: 1,616.
  - Usable events excluding retroactive implementations: 1,475 (54 percent of 2,736).
  - Direction among 1,475: 594 tax increase announcements; 881 tax decrease announcements.
  - Implementation-lag statistics (1,475 announcements): median lag = 99 days; average lag = 192 days.
  - Number with implementation lag < 90 days (baseline tax news shocks): 671 out of 1,475.

### Stylized facts and descriptive statistics
- Prevalence:
  - Country-months with tax news shocks (all tax changes): 322.
  - Average monthly probability (all tax changes): 4.2 percent (average country faces a tax news shock every 23.8 months).
  - Average monthly probability (tax increases): 2.3 percent.
  - Average monthly probability (tax decreases): 2.9 percent.
- Election-period average monthly probabilities (12-month windows):
  - Overall sample baseline: 4.2 percent.
  - Pre-election 12-month period: 3.3 percent.
  - Post-election 12-month period: 5.8 percent.
  - Tax increases: baseline 2.3 percent; pre-election 1.1 percent; post-election 3.6 percent.
  - Tax decreases: baseline 2.9 percent; pre-election 2.6 percent; post-election 3.6 percent.
- Cross-country dispersion:
  - In 19 out of 22 countries, post-election probability of tax news shocks exceeds pre-election probability (difference > 0).
- Stylized interpretation:
  - Tax news shocks become less likely before elections and more likely after elections.
  - Pre-election reduction particularly pronounced for tax increase announcements.

### Empirical strategy
- Model:
  - Panel linear probability model with fixed effects.
  - Dependent variable: monthly dummy for presence of a tax news shock (announcement→implementation < 90 days).
  - Key independent variables: dummies for 12-month pre-election period (includes election month) and 12-month post-election period.
  - Fixed effects: country-year (θ_{c,y}) and month (θ_t).
  - Standard errors clustered at country-year level.
- Identification:
  - Monthly frequency enables clearer timing effects and inclusion of country-year and month fixed effects.
  - Assumption: election dates are mostly predetermined relative to implementation lags.

### Main results (12-month window; dependent variable = dummy for announcement→implementation < 90 days)
- Key descriptive averages reiterated:
  - Average monthly probability (all changes): 4.2 percent.
  - Average monthly probability (tax increases): 2.3 percent.
  - Average monthly probability (tax decreases): 2.9 percent.
- Main regression coefficients (point estimate and clustered standard error):
  - Pre-election 12-month period:
    - Change (all tax changes): -0.026*** (0.009).
    - Increase: -0.024*** (0.007).
    - Decrease: -0.016** (0.008).
    - Interpretation: pre-election decline of 2.6 percentage points (all changes); 2.4 pp for increases; 1.6 pp for decreases.
  - Post-election 12-month period:
    - Change (all tax changes): 0.030*** (0.009).
    - Increase: 0.023*** (0.007).
    - Decrease: 0.018** (0.008).
    - Interpretation: post-election increase of 3.0 percentage points (all changes); 2.3 pp for increases; 1.8 pp for decreases.
- Fit and sample:
  - Observations: 7,656 per regression column.
  - R-squared: range 0.144–0.146 across main columns.

### Robustness and alternative definitions
- Narrower window (6-month):
  - Pre-election (Change): -0.022*** (0.009).
  - Post-election (Change): 0.019** (0.010).
  - Observations: 7,656; R-squared range: 0.142–0.145.
- Multiple announcements handled by average implementation lag (12-month window):
  - Pre-election (Change): -0.021** (0.009).
  - Post-election (Change): 0.023*** (0.009).
  - Effect on tax decreases less pronounced under this definition.
- Dependent variable = number of tax news shocks per month (12-month window):
  - Pre-election (Change): -0.062** (0.026).
  - Post-election (Change): 0.056** (0.024).
- Sensitivity to lag thresholds (12-month window):
  - Threshold < 30 days: Pre-election (Change): -0.018*** (0.006); Post-election (Change): 0.017*** (0.006).
  - Threshold < 60 days: Pre-election (Change): -0.022*** (0.008); Post-election (Change): 0.024*** (0.008).
  - Threshold < 120 days: Pre-election (Change): -0.028*** (0.010); Post-election (Change): 0.031*** (0.010).
  - Threshold = median (99 days): Pre-election (Change): -0.031*** (0.010); Post-election (Change): 0.034*** (0.010).
- Controlling for monthly institutional and economic covariates (lagged):
  - Election window coefficients remain similar (Table 9):
    - Pre-election (Change): -0.030*** (0.010).
    - Post-election (Change): 0.031*** (0.010).
  - Institutional strength, industrial production change, and inflation not statistically significant in these specifications.
  - Observations reduced to 6,676 due to data availability.
- Controlling for number of tax announcements (Table 10):
  - Election coefficients attenuate but largely persist:
    - Pre-election (Change): -0.012* (0.007).
    - Post-election (Change): 0.019*** (0.007).
  - Announcements variable strongly positive: 0.126*** (0.009) (Change column).
  - Observations: 7,656; R-squared increased (e.g., 0.403 in Change column).
- Nonlinear binary model (Probit with bias correction; average partial effects):
  - Pre-election (Change): -0.089*** (0.025).
  - Post-election (Change): 0.115*** (0.032).
  - Observations: varying subsets reported (e.g., 1,817 for Change pre/post in Table 11).

### Explanatory power (ROC / AUC analysis)
- Baseline model with fixed effects:
  - AUC = 0.617 with a standard error of 0.018.
  - Statistically different than 0.5 at the 1 percent significance level.
- Model with 12-month pre-election dummy:
  - AUC = 0.678 with a standard error of 0.015.
  - Statistically different than 0.5 at the 1 percent significance level.
  - Statistically different from the fixed-effects-only model at the 5 percent significance level.
- Model with 12-month post-election dummy:
  - AUC = 0.670 with a standard error of 0.016.
  - Statistically different than 0.5 at the 1 percent significance level.
  - Statistically different from the fixed-effects-only model at the 10 percent significance level.
- Conclusion: Pre- and post-election dummies add predictive information for tax news shocks though they are not perfect predictors.

### Heterogeneity: tax types and country groups
- Changes in tax base versus tax rates (12-month window; announcement→implementation < 90 days):
  - Average monthly probabilities (footnote):
    - Tax base: 3.1 percent.
    - Tax rates: 2.2 percent.
  - Election window coefficients:
    - Pre-election — Tax base: -0.019** (0.008).
    - Pre-election — Tax rates: -0.018*** (0.007).
    - Post-election — Tax base: 0.018** (0.008).
    - Post-election — Tax rates: 0.024*** (0.007).
  - Interpretation: Pre- and post-election effects hold for both tax-base and tax-rate changes.
- Consumption versus income taxes (12-month window):
  - Average monthly probabilities (footnote):
    - Consumption taxes: 1.2 percent.
    - Income taxes: 3.2 percent.
  - Election window coefficients:
    - Pre-election — Consumption: -0.010* (0.005).
    - Pre-election — Income: -0.021** (0.008).
    - Post-election — Consumption: 0.015*** (0.005).
    - Post-election — Income: 0.021*** (0.008).
  - Interpretation: Effects present for both consumption and income taxes; larger pre-election reduction for income taxes.
- Advanced economies (AEs) versus emerging markets (EMs) (12-month window):
  - Observations: AEs 5,916; EMs 1,740.
  - Election window coefficients:
    - Pre-election — AEs: -0.017* (0.010).
    - Pre-election — EMs: -0.058** (0.024).
    - Post-election — AEs: 0.022** (0.009).
    - Post-election — EMs: 0.038 (0.028).
  - Interpretation: Confirmed patterns in AEs (pre-election decrease; post-election increase). In EMs, pre-election decrease observed; post-election increase not statistically significant.

### Mechanisms and interpretation
- Political economy channels discussed:
  - Pre-election postponement: governments delay unpopular tax increases to improve re-election chances; strategic timing for tax cuts may vary.
  - Post-election implementation: newly elected governments implement announced changes faster due to political capital or reduced electoral risk.
- Relation to literature:
  - Connects to opportunistic political business cycles and literature on anticipated versus unanticipated tax shocks and policy uncertainty.

### Conclusion and policy implications
- Main empirical conclusions:
  - Implementation lags are significantly longer for tax policy change announcements made in pre-election periods, reducing likelihood of tax news shocks by between 2 to 3 percentage points.
  - Implementation lags are shorter for announcements in post-election periods, increasing likelihood of tax news shocks by about 3 percentage points.
  - Patterns are robust across tax types and categories, specifications, and alternative definitions of tax news shocks.
- Macroeconomic relevance:
  - Anticipated tax changes enable intertemporal substitution; unanticipated changes contribute to policy uncertainty with potential adverse macroeconomic effects.
  - Elections shape a channel through which tax-policy timing affects macroeconomic outcomes.
- Policy recommendations presented:
  - Reduce fiscal policy uncertainty especially around elections.
  - Consider mechanisms to reduce incentives for opportunistic timing after elections, for example:
    - Procedural fiscal rules.
    - Reforms to budget processes that promote transparency and ensure predictability on tax policy changes.
  - Potential benefits: enhanced credibility and increased effectiveness that could translate into lower spreads and risk premia.
- Theoretical caveat:
  - Under some circumstances unanticipated policy changes might be less detrimental than anticipated ones (example cited in text), but empirical support for positive effects of surprises is limited.

*IMF Working Paper — "Unpleasant Surprises? Elections and Tax News Shocks" (excerpt from wpiea2023139-print-pdf)*

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

### 1. Introduction

### Research question and motivation
- Investigates whether the timing of elections affects the duration between announcement and implementation of tax policy changes, thereby influencing the likelihood of unanticipated tax policy changes (“tax news shocks”).
- Highlights two channels by which tax policy timing matters:
  - Fiscal foresight: longer announcement-to-implementation durations allow agents to adjust behavior (Leeper et al. 2012, 2013), potentially altering macroeconomic impacts of tax changes.
  - Policy uncertainty: unanticipated tax policy changes can raise uncertainty and hinder economic activity (Bernanke 1983, Bloom et al. 2007; Baker et al. 2016).

### Data and identification
- Uses monthly data for 22 advanced economies (AEs) and emerging markets (EMs) over the period 1990-2018.
- Employs the Tax Policy Reform Database (TPRD) constructed by Amaglobeli et al. (2018) with detailed information on announcement and implementation dates, types and directions of tax changes, and economic significance.
- Defines a tax news shock as a policy change whose announcement-to-implementation duration is shorter than 90 days.
- Uses monthly frequency to better capture pre- and post-election effects and improve identification by controlling for country-year economic and institutional factors and month-specific common shocks.

### Main empirical findings
- Average monthly probability of tax news shocks (encompassing both tax increases and cuts) in the whole sample is 4.2 percent.
- Election-period effects:
  - The average monthly probability of tax news shocks decreases by 2.6 percentage points during the 12-month period running up to the elections.
  - The average monthly probability of tax news shocks increases by 3.0 percentage points in the first 12 months following elections.
- Directional and tax-type patterns:
  - The lower pre-election probability is driven by decreased likelihood of tax news shocks for both tax increases and tax cuts.
  - Post-election increases in the probability of tax news shocks apply to both tax increases and tax cuts.
  - These patterns hold for changes in the tax base and tax rates, and for consumption and income taxes.
  - Post-election effects on the likelihood of tax news shocks are more pronounced in the case of AEs.

### Interpretation and theoretical mechanisms
- Political economy mechanisms considered:
  - Pre-election postponement: Governments may delay unpopular tax increases before elections to improve re-election chances; may accelerate pre-election tax cuts or delay their implementation depending on strategic considerations.
  - Post-election implementation: Newly elected governments may implement announced changes faster due to political capital or reduced electoral risk.
  - Strategic timing may differ by direction of change (tax increases versus tax cuts) and by whether governments seek immediate electoral gains or to time benefits for future elections.
- Relates to opportunistic political business cycles (PBCs) and electoral cycles in fiscal policy (Nordhaus 1975; Persson and Tabellini 1990; Alesina et al. 1997), but focuses on pace of implementation rather than frequency of announcements.

### Robustness and additional analyses
- Results are robust to:
  - Alternative definitions of tax news shocks.
  - Accounting for differences in institutional capacity or constraints to implement announced tax measures.
- Economic controls:
  - Examines the role of economic activity and inflation on the likelihood of tax news shocks.
  - Finds that neither economic activity nor inflation seems to affect the occurrence of tax news shocks in the sample.
  - Controlling for these variables does not change the main result regarding election-period effects.
- Identification assumption:
  - Relies on election dates being mostly predetermined relative to implementation lags; the authors note they confirm this empirically by focusing on elections with pre-determined dates.

### Contribution to literature
- Novelty:
  - First study, to the authors’ knowledge, exploring the role of election timing in the occurrence of tax news shocks (determinants of unanticipated tax policy changes).
  - One of the first cross-country studies using monthly-frequency data on election timing to study electoral cycles in policy implementation timing.
- Situates findings within literature on:
  - Effects of anticipated versus unanticipated tax policy changes (Mertens and Ravn 2011, 2012, 2014; Leeper et al. 2012, 2013; Ramey 2016).
  - Policy uncertainty and macroeconomic outcomes (Rodrik 1991; Baker et al. 2016).
  - Electoral cycles in fiscal policy and tax reforms (Alesina et al. 1992; Fuest et al. 2021; David and Sever 2022).

*IMF Working Paper — Introduction (excerpt) from wpiea2023139-print-pdf*

### Section 5 concludes.

### Section 5 concludes.

### Data and Stylized Facts: Tax Policy Changes
- Definition and identification
  - Tax news shock: announcement-to-implementation gap shorter than 90 days.
  - Monthly dummy variable: equals 1 if at least one announcement in a given country-month has implementation lag < 90 days, 0 otherwise.
  - Measure applied separately for: all policy changes, tax increases, and tax decreases.
  - Focus on announcement date (behavior of government and election-motivated signaling).
- Data source and scope
  - Primary source: the most recent version of the TPRD (Tax Policy Reform Database) constructed by Amaglobeli et al. (2018) using >950 OECD Economic Surveys and 53,000 tax-related documents.
  - Types of taxes covered: corporate income tax (CIT), personal income tax (PIT), value-added and sale taxes (VAT), social security contributions (SSC), excise tax (EXE), and property tax (PRO).
  - Information available: direction, announcement and implementation dates (exact days for majority of events), and economic significance (“major” or “minor”).
  - Countries in sample: Australia, Austria, Brazil, Canada, China, Czechia, Denmark, France, Greece, Germany, India, Ireland, Italy, Japan, Korea, Luxembourg, Mexico, Poland, Portugal, Spain, Turkiye, United Kingdom, United States. China is dropped from the analytical sample due to peculiarities of its electoral system.
  - Time period analyzed: 1990-2018.
- Sample construction and major-change focus
  - Analysis restricted to “major” tax policy changes:
    - Rate changes: larger than 1 percentage point.
    - Base changes: broadening likely affects a large group or has large potential for mobilizing new resources.
  - Total major tax policy change announcements in sample: 2,736.
  - Announcements with both announcement and implementation dates: 1,616.
  - Excluding retroactive implementations (negative lags): usable events for baseline analysis: 1,475 (54 percent of 2,736).
  - Direction of those 1,475 announcements: 594 tax increase announcements; 881 tax decrease announcements.
  - Implementation lag statistics (for 1,475 announcements): median lag = 99 days; average lag = 192 days.
  - Number of announcements with implementation lag < 90 days (tax news shocks per baseline definition): 671 out of 1,475.
- Robustness and alternative definitions
  - Alternative time windows tested: implementation lag < 30, < 60, < 120 days.
  - Alternative threshold: implementation lag below sample median (99 days).
  - Alternatives to handle multiple announcements in a month:
    - Use average implementation lag of all announcements in a country-month and compare to 90 days.
    - Use count (number) of tax news shocks in a country-month instead of a dummy.

### Data: Elections and Other Variables
- Elections data
  - Months of elections collected manually for 22 countries for 1990-2018.
  - Election type chosen based on political system: presidential elections for presidential systems (e.g., United States), parliamentary elections for parliamentary systems (e.g., Germany).
  - Cross-checked with other datasets where applicable.
- Additional monthly controls for robustness checks
  - Institutional proxies from ICRG:
    - Bureaucratic quality index (range 0–4).
    - Law and order index (range 0–6).
  - Economic proxies from IMF IFS:
    - Industrial production index change (year-on-year).
    - Inflation (consumer prices, year-on-year).

### Stylized Facts: Events and Election Timing
- Sample size and prevalence
  - Sample: monthly data for 22 AEs and EMs over 1990-2018.
  - Number of country-month observations: 7,656.
  - Country-months with tax news shocks (all tax changes): 322.
  - Average monthly probability of tax news shocks (all tax changes): 4.2 percent.
    - Interpretation: average country faces a tax news shock every 23.8 months.
  - Country-months with tax news shocks for tax increases: 176 (average monthly probability 2.3 percent).
  - Country-months with tax news shocks for tax decreases: 225 (average monthly probability 2.9 percent).
- Temporal and cross-country patterns
  - Annual counts: multiple shocks per year, with a notable decrease in prevalence in the last few years of the sample.
  - Cross-country dispersion: the number of shocks differs across countries (example cited: US lower number, Turkiye highest number).
- Relationship with election periods (12-month windows)
  - Average monthly probability (overall sample, percent):
    - Overall sample baseline: 4.2 percent.
    - Pre-election 12-month period: 3.3 percent.
    - Post-election 12-month period: 5.8 percent.
  - Tax increases (12-month windows, percent):
    - Overall sample baseline: 2.3 percent.
    - Pre-election: 1.1 percent.
    - Post-election: 3.6 percent.
  - Tax decreases (12-month windows, percent):
    - Overall sample baseline: 2.9 percent.
    - Pre-election: 2.6 percent.
    - Post-election: 3.6 percent.
  - Cross-country differences in pre- vs post-election probability:
    - In 19 out of 22 countries, the post-election probability of tax news shocks (all changes) exceeds the pre-election probability (difference > 0).
- Stylized interpretation
  - Tax news shocks become less likely before elections and more likely after elections.
  - The pre-election reduction in likelihood is particularly pronounced for tax increase announcements; less pronounced for pre-election tax decreases.
  - Evidence suggests governments act relatively slowly to implement tax policy changes announced before elections, and implement announced changes after elections more quickly.

### Empirical Strategy
- Objective
  - Estimate whether likelihood of tax news shocks differs for announcements made before and after elections versus other periods.
- Main econometric approach
  - Panel linear probability model with fixed effects.
  - Dependent variable: monthly dummy indicating presence of a tax news shock in a country-month (measured at announcement date).
  - Key independent variables: dummies for 12-month pre-election period (includes election month) and for 12-month post-election period; also examined with 6-month windows.
  - Fixed effects included:
    - Country-year fixed effects (θ_{c,y}) to absorb macroeconomic and institutional factors varying at country-year frequency.
    - Month fixed effects (θ_t) to absorb global monthly shocks common across countries.
  - Standard errors clustered at the country-year level.
- Identification assumptions and advantages
  - Monthly frequency enables:
    - Clearer identification of timing effects before and after elections compared to annual data.
    - Inclusion of country-year and month fixed effects to mitigate omitted-variable concerns.
  - Assumption: election dates are mostly predetermined and unlikely to be driven by tax news shocks within a 12-month window (reverse causality addressed later).
- Interpretation of coefficient of interest (β)
  - Negative β for pre-election dummy implies lower probability (governments act more slowly) of tax news shocks in pre-election months.
  - Positive β for post-election dummy implies higher probability (governments act more quickly) of tax news shocks in post-election months.
  - Based on stylized facts, expectation: β < 0 for pre-election; β > 0 for post-election.

*Source: IMF Working Paper "Unpleasant Surprises? Elections and Tax News Shocks" (Section presented: Data, Stylized Facts, and Empirical Strategy).*

### 4. Results

### 4. Results

### 4.1. Main Results
- Key descriptive averages:
  - The average monthly probability of tax news shocks (all tax change announcements) in the sample is 4.2 percent.
  - The average monthly probability of tax news shocks related to tax increase announcements is 2.3 percent.
  - The average monthly probability of tax news shocks related to tax decrease announcements is 2.9 percent.
- Main regression findings (12-month window around elections; dependent variable = dummy for months with a tax news shock where announcement→implementation < 90 days):
  - Pre-election 12-month period (Columns 1–3, Table 1):
    - Election window coefficient (Change, all tax changes): -0.026*** (0.009).
    - Election window coefficient (Increase): -0.024*** (0.007).
    - Election window coefficient (Decrease): -0.016** (0.008).
    - Interpretation: The average monthly probability of tax news shocks (all changes) declines by 2.6 pp in the pre-election period; declines are driven by both tax increases and tax decreases (2.4 pp lower for increases; 1.6 pp lower for decreases).
  - Post-election 12-month period (Columns 4–6, Table 1):
    - Election window coefficient (Change, all tax changes): 0.030*** (0.009).
    - Election window coefficient (Increase): 0.023*** (0.007).
    - Election window coefficient (Decrease): 0.018** (0.008).
    - Interpretation: The average monthly probability of tax news shocks (all changes) increases by 3.0 pp in the first 12 months following elections; increases hold for both tax increases and decreases.
- Table 1 sample and fit statistics:
  - Country-year F.E.: Yes; Month F.E.: Yes.
  - Observations: 7656 (for each column).
  - R-squared: 0.146, 0.144, 0.145, 0.146, 0.144, 0.146 (by column).

- Robustness to narrower window:
  - Table 2 (6-month window) shows broadly similar patterns:
    - Pre-election (Change): -0.022*** (0.009).
    - Post-election (Change): 0.019** (0.010).
    - Observations: 7656; R-squared range: 0.142–0.145.

### 4.2. Alternative Definitions of Tax News Shocks
- Multiple announcements in a month — average implementation lag:
  - Definition: When multiple announcements occur in a month, take the average implementation lag; identify shock if average < 90 days.
  - Table 3 (12-month window):
    - Pre-election (Change): -0.021** (0.009).
    - Pre-election (Increase): -0.017*** (0.006).
    - Pre-election (Decrease): -0.005 (0.006).
    - Post-election (Change): 0.023*** (0.009).
    - Post-election (Increase): 0.015*** (0.006).
    - Post-election (Decrease): 0.009 (0.006).
    - Observations: 7656; R-squared: 0.151, 0.143, 0.140, 0.152, 0.143, 0.140.
  - Note: Effect on tax decreases becomes less pronounced under this alternative.

- Dependent variable = number of tax news shocks per month:
  - Table 4 (12-month window):
    - Pre-election (Change): -0.062** (0.026).
    - Pre-election (Increase): -0.051*** (0.013).
    - Pre-election (Decrease): -0.012 (0.017).
    - Post-election (Change): 0.056** (0.024).
    - Post-election (Increase): 0.040*** (0.015).
    - Post-election (Decrease): 0.016 (0.015).
    - Observations: 7656; R-squared: 0.140, 0.142, 0.134, 0.140, 0.141, 0.134.
  - Note: Similar overall results; less pronounced pattern for tax decreases.

- Sensitivity to chosen implementation-lag threshold:
  - Threshold = 30 days (Table 5, 12-month window):
    - Pre-election (Change): -0.018*** (0.006).
    - Post-election (Change): 0.017*** (0.006).
    - Observations: 7656; R-squared: 0.153, 0.147, 0.150, 0.153, 0.147, 0.150.
  - Threshold = 60 days (Table 6, 12-month window):
    - Pre-election (Change): -0.022*** (0.008).
    - Post-election (Change): 0.024*** (0.008).
    - Observations: 7656; R-squared: 0.145, 0.141, 0.146, 0.145, 0.141, 0.146.
  - Threshold = 120 days (Table 7, 12-month window):
    - Pre-election (Change): -0.028*** (0.010).
    - Post-election (Change): 0.031*** (0.010).
    - Observations: 7656; R-squared: 0.143, 0.147, 0.140, 0.143, 0.147, 0.140.
  - Threshold = sample median = 99 days (Table 8, 12-month window):
    - Pre-election (Change): -0.031*** (0.010).
    - Post-election (Change): 0.034*** (0.010).
    - Observations: 7656; R-squared: 0.144 (across columns).

- Summary: Findings are broadly similar across alternative definitions and thresholds; effects for tax decreases are sometimes less pronounced depending on definition.

### 4.3. Other Robustness Checks
- Controlling for monthly country-level covariates (Table 9; 12-month window; dependent variable dummy where announcement→implementation < 90 days):
  - Election window coefficients remain similar:
    - Pre-election (Change): -0.030*** (0.010).
    - Pre-election (Increase): -0.027*** (0.007).
    - Pre-election (Decrease): -0.017** (0.008).
    - Post-election (Change): 0.031*** (0.010).
    - Post-election (Increase): 0.021*** (0.007).
    - Post-election (Decrease): 0.021** (0.008).
  - Added controls (lagged values):
    - Institutional strength (ICRG bureaucratic quality): 0.003 (0.042); -0.011 (0.044); 0.001 (0.045); 0.007 (0.043); -0.008 (0.044); 0.003 (0.045).
    - Economic activity (industrial production index change): -0.000 (0.001); -0.000 (0.001); -0.001 (0.001); -0.000 (0.001); -0.000 (0.001); -0.001 (0.001).
    - Inflation: 0.000 (0.000) across columns.
  - Observations: 6676 (reduced due to data availability); R-squared: 0.155, 0.152, 0.156, 0.155, 0.151, 0.156.
  - Interpretation: Monthly institutional and economic controls do not materially change the main election-period effects; the controls themselves are not statistically significant.

- Controlling for number of tax policy change announcements (Table 10; 12-month window):
  - Election window coefficients attenuate but remain statistically significant in many cases:
    - Pre-election (Change): -0.012* (0.007).
    - Pre-election (Increase): -0.008* (0.005).
    - Pre-election (Decrease): -0.012** (0.006).
    - Post-election (Change): 0.019*** (0.007).
    - Post-election (Increase): 0.009** (0.005).
    - Post-election (Decrease): 0.015** (0.006).
  - Announcements variable (number of announcements) is strongly positive and significant for all columns:
    - Announcements coefficients: 0.126*** (0.009); 0.197*** (0.020); 0.155*** (0.013); 0.126*** (0.009); 0.197*** (0.020); 0.155*** (0.013).
  - Observations: 7656; R-squared: 0.403, 0.446, 0.405, 0.404, 0.446, 0.406.
  - Interpretation: Election-period effects persist even after accounting for the number of announcements, although the number of announcements strongly predicts tax news shocks.

- Nonlinear binary models and additional checks:
  - Probit model with analytical bias correction for incidental parameters (Table 11; average partial effects reported; fixed effects included):
    - Pre-election (Change): -0.089*** (0.025).
    - Pre-election (Increase): -0.209*** (0.038).
    - Pre-election (Decrease): -0.078** (0.036).
    - Post-election (Change): 0.115*** (0.032).
    - Post-election (Increase): 0.285*** (0.055).
    - Post-election (Decrease): 0.106** (0.045).
    - Observations: 1817, 769, 1085 (pre-election columns) and same for post-election; Pseudo R-squared: 0.116, 0.135, 0.112, 0.118, 0.142, 0.114.
  - Additional unreported robustness exercises (available upon request) include:
    - Excluding countries with extreme numbers of shocks.
    - Weighted regressions (weights: real GDP, real GDP per capita, inverse).
    - Dropping last four years and restricting to later years (e.g., starting from the 2000s).
    - Removing snap elections and using only pre-determined-date elections.
    - Alternative fixed-effects specifications (no fixed effects; country fixed effects instead of country-year).
    - Non-clustered standard errors.
    - Logit model instead of Probit.
    - Oster (2019) omitted variable bias test: Oster’s delta > 1 (range 5–24) for all regressions except one (likelihood of tax news shocks related to tax decreases in the pre-election period), suggesting robustness to omitted variable bias.

### 4.4. Explanatory Power
- Approach: Receiver operating characteristic (ROC) curve analysis using Probit model to assess predictive power for binary classification of tax news shocks.
- Conceptual summary:
  - x-axis: false positive rate (how often model predicts a shock when none occurs).
  - y-axis: true positive rate (how often model predicts a shock when a shock occurs).
  - Area under the curve (AUC): measure of predictive power; AUC = 0.5 indicates no information (coin toss), AUC = 1 is perfect prediction.
- Implementation:
  - Analysis uses tax news shocks defined based on all tax policy change announcements (both increases and decreases).
  - Gray lines in Figure 3: goodness-of-fit from Probit models in columns 1 and 4 of Table 11 (pre- and post-election Probit models with election window dummy and fixed effects).
  - Black lines in Figure 3: Probit models dropping the election window dummy but keeping fixed effects.
- Reported qualitative conclusion (from text):
  - The AUC value of the model that only includes the fixed effects is [text truncated in source before numeric AUC reported], but the ROC discussion underscores that the election-window variables materially affect predictive fit relative to fixed-effects-only specifications.

*Source: IMF Working Paper — "Unpleasant Surprises? Elections and Tax News Shocks", Section 4: Results.*

### 0.617 with a standard error of 0.018. Being statistically different than 0.5 at the 1 percent significance level, it

### wpiea2023139-print-pdf - 0.617 with a standard error of 0.018. Being statistically different than 0.5 at the 1 percent significance level, it

### Predictive performance of models (ROC / AUC)
- Baseline model with fixed effects: AUC = 0.617 with a standard error of 0.018.
  - Interpretation: Statistically different than 0.5 at the 1 percent significance level; model performs significantly better than tossing a coin in predicting tax news shocks.
- Model with 12-month pre-election dummy: AUC = 0.678 with a standard error of 0.015.
  - Interpretation: AUC is statistically different from 0.5 at the 1 percent significance level.
  - Comparison: Statistically different from the model excluding the pre-election dummy at the 5 percent significance level.
- Model with 12-month post-election dummy: AUC = 0.670 with a standard error of 0.016.
  - Interpretation: AUC is statistically different from 0.5 at the 1 percent significance level.
  - Comparison: Statistically different from the model excluding the post-election dummy at the 10 percent significance level.
- Conclusion: Pre- and post-election dummies add predictive information for tax news shocks though they are not perfect predictors.

### Changes in the tax base versus tax rates (Table 12)
- Dependent variable: dummy = 1 in months with a tax news shock regarding changes in the tax base or tax rates (announcement-to-implementation < 90 days).
- Key coefficient estimates (Election window; 12-month pre-election and 12-month post-election):
  - Pre-election — Tax base: -0.019** (0.008)
  - Pre-election — Tax rates: -0.018*** (0.007)
  - Post-election — Tax base: 0.018** (0.008)
  - Post-election — Tax rates: 0.024*** (0.007)
- Controls and sample:
  - Country-year F.E.: Yes (columns 1–4)
  - Month F.E.: Yes (columns 1–4)
  - Observations: 7656 (columns 1–4)
  - R-squared: 0.143, 0.140, 0.143, 0.141 (columns 1–4 respectively)
  - Standard errors clustered at the country-year level.
  - Significance notation preserved: *** p<0.01, ** p<0.05, * p<0.1.
- Interpretation:
  - Pre-election period associated with lower probability of tax news shocks for both tax-base and tax-rate announcements.
  - Post-election period associated with higher probability of tax news shocks for both categories.
- Caveat: More granular categorization reduces event frequency in the sample.
- Average monthly probabilities (footnote):
  - Tax base: 3.1 percent
  - Tax rates: 2.2 percent

### Changes in consumption versus income taxes (Table 13)
- Dependent variable: dummy = 1 in months with a tax news shock regarding consumption or income taxes (announcement-to-implementation < 90 days).
- Key coefficient estimates (Election window; 12-month pre-election and 12-month post-election):
  - Pre-election — Consumption: -0.010* (0.005)
  - Pre-election — Income: -0.021** (0.008)
  - Post-election — Consumption: 0.015*** (0.005)
  - Post-election — Income: 0.021*** (0.008)
- Controls and sample:
  - Country-year F.E.: Yes (columns 1–4)
  - Month F.E.: Yes (columns 1–4)
  - Observations: 7656 (columns 1–4)
  - R-squared: 0.129, 0.143, 0.129, 0.143 (columns 1–4 respectively)
  - Standard errors clustered at the country-year level.
  - Significance notation preserved: *** p<0.01, ** p<0.05, * p<0.1.
- Interpretation:
  - Pre-election associated with reduced likelihood of tax news shocks for both consumption and income taxes (larger effect for income taxes).
  - Post-election associated with increased likelihood for both categories.
- Caveat: Granular analysis reduces event frequency for each tax type.
- Average monthly probabilities (footnote):
  - Consumption taxes: 1.2 percent
  - Income taxes: 3.2 percent

### Advanced economies (AEs) versus emerging markets (EMs) (Table 14)
- Dependent variable: dummy = 1 in months with a tax news shock (announcement-to-implementation < 90 days), including tax decreases and increases.
- Subsample estimates (Election window; 12-month pre-election and 12-month post-election):
  - Pre-election — AEs: -0.017* (0.010)
  - Pre-election — EMs: -0.058** (0.024)
  - Post-election — AEs: 0.022** (0.009)
  - Post-election — EMs: 0.038 (0.028)
- Controls and sample:
  - Country-year F.E.: Yes (columns 1–4)
  - Month F.E.: Yes (columns 1–4)
  - Observations: AEs 5916; EMs 1740
  - R-squared: 0.169, 0.272, 0.169, 0.269 (columns 1–4 respectively)
  - Standard errors clustered at the country-year level.
  - Significance notation preserved: *** p<0.01, ** p<0.05, * p<0.1.
- Interpretation:
  - Results confirmed in the subsample of AEs (pre-election decrease; post-election increase).
  - In EMs: pre-election decrease observed; post-election increase not statistically significant in this sample.

### Conclusion and policy implications
- Data and scope:
  - Monthly data from 22 advanced economies and emerging markets over the period 1990-2018.
- Main empirical findings:
  - Implementation lags are significantly longer for tax policy change announcements made in pre-election periods, leading to a decrease in the likelihood of tax news shocks.
    - Typical decrease in the probability of a tax news shock in pre-election periods: between 2 to 3 percentage points.
  - Implementation lags are shorter for announcements in post-election periods, increasing the likelihood of tax news shocks.
    - Typical increase in the probability of tax news shocks in post-election periods: 3 percentage points.
  - Patterns are similar across types of tax measures and tax categories; specifications absorb country-specific annual factors and common monthly developments.
  - Findings robust to alternative definitions of tax news shocks and to controlling for economic variables and institutional factors.
- Macroeconomic and policy relevance:
  - Anticipated tax changes cause intertemporal substitution of taxable activities; unanticipated changes can contribute to economic policy uncertainty with undesired macroeconomic outcomes.
  - The change in likelihood of tax news shocks around elections identifies a channel through which elections can shape macroeconomic outcomes.
- Policy recommendations (as presented in the text):
  - Reduce fiscal policy uncertainty especially around elections.
  - Consider imposing mechanisms to reduce incentives to behave opportunistically in the aftermath of elections, for example:
    - Procedural fiscal rules.
    - Reforms to budget processes that promote transparency and ensure predictability on tax policy changes.
  - Potential benefits: enhanced credibility (by anchoring of expectations) and increased effectiveness (Leeper 2009, End and Hong 2022) that could translate into lower spreads and risk premia.
- Theoretical caveat noted in the paper:
  - Some literature suggests under certain circumstances unanticipated policy changes might have less detrimental effects than anticipated ones (example cited: Korinek and Stiglitz (2009) on dividend tax rates), but empirical evidence of positive effects of policy surprises remains elusive.

*Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023139-print-pdf.pdf*

### References

### wpiea2023139-print-pdf - References

### References list
- Comprehensive bibliography covering political business cycles, tax policy, uncertainty, investment dynamics, fiscal foresight, and related empirical methods.
- Includes working papers, journal articles, books, databases, and institutional publications cited in the main text, e.g.:
  - Alesina, Alberto, and Andrea Passalacqua. 2016. "The political economy of government debt." Handbook of Macroeconomics 2: 2599-2651.
  - Amaglobeli, David, Valerio Crispolti, Era Dabla-Norris, Pooja Karnane, and Florian Misch. 2018. “Tax policy measures in advanced and emerging economies: a novel database”. International Monetary Fund Working Paper No 18/110.
  - Baker, Scott R., Nicholas Bloom, and Steven J. Davis. 2016. "Measuring economic policy uncertainty." The Quarterly Journal of Economics 131: 1593-1636.
  - Mertens, Karel, and Morten O. Ravn. 2012. "Empirical evidence on the aggregate effects of anticipated and unanticipated US tax policy shocks." American Economic Journal: Economic Policy 4: 145-81.
  - Sirbu, Anca-Ioana. 2019. "News About Taxes and Expectations-Driven Business Cycles." Macroeconomic Dynamics 23: 1340-1370.
- Methodological and data sources cited include:
  - Comparative Political Data Set 1960-2020 (Armingeon, Klaus, Sarah Engler and Lucas Leemann. 2022).
  - International Financial Statistics (IFS) Database (International Monetary Fund).
  - The International Country Risk Guide (ICRG) Database (Political Risk Services Group).

### Appendix — Elections in the sample (Table A.1)
- Table A.1 enumerates election dates covered in the sample for multiple countries across regions.
- Countries listed include: Australia, Austria, Brazil, Canada, Czechia, Denmark, France, Germany, Greece, India, Ireland, Italy, Japan, Korea, Luxembourg, Mexico, Poland, Portugal, Spain, Turkiye, UK, USA, and others.
- Election-date format preserved as month/year entries (e.g., 3/1990; 6/1990; 12/1990; 2/1990; 10/1991; 4/1992).

### Appendix — Implementation lags and election periods (Figure A.1)
- Analysis focus:
  - Distribution of implementation lags for announced tax policy changes during pre- and post-election 12-month periods.
  - Implementation lags are regressed on country fixed effects and the residuals are reported (residuals include both negative and positive values).
- Key findings described:
  - The chart on the first row shows distributions around election periods using tax policy changes including both tax increases and decreases.
  - The second row shows separate distributions for announcements regarding tax increases (left-hand chart) and tax decreases (right-hand chart).
  - The distributions tend to shift right in pre-election periods (dotted red lines) compared to the post-election windows (solid black lines), pointing to longer implementation lags before elections.
- Notes and definitions preserved exactly:
  - "The chart on the first row displays the distribution of implementation lags of the tax policy change announcements (including both tax increase and decrease) that are made during the 12-month periods before (dotted red line) and after elections (solid black line)."
  - "The left-hand side (right-hand side) chart on the second row illustrates the distribution of implementation lags of the tax policy change announcements with a tax increase (decrease) that are made during the 12-month periods before and after elections."
  - "Implementation lags are net of country fixed effects."
  - "We note that the pre-election period includes 12 months running up to the elections and the election months as with the regression analysis above."

*Content derived from "References" and Appendix (Table A.1 and Figure A.1) of wpiea2023139-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023139-print-pdf.pdf_