## wp18110 - References

## Source details

**Canonical URL:** [wp18110 - References](https://www.imf.org/-/media/files/publications/wp/2018/wp18110.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18110.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18110.pdf.json)

---

### Overview and purpose
- Introduces the Tax Policy Reform Database (TPRD), a comprehensive database of tax policy measures adopted in 23 advanced and emerging market economies over the last four decades.
- Motivation: detailed study of past tax reforms and their impacts to inform future reforms and policy advice.

### Data sources and scope
- Primary documentary sources:
  - "953 OECD Economic Surveys examined for 23 countries."
  - "37,943 IBFD news clips" (tax-related news from the International Bureau of Fiscal Documentation).
- Time coverage and timing precision:
  - Announcement and implementation dates (day and/or month and year) documented where available, with especially detailed coverage for 1988-2014 for PIT, CIT, and VAT measures.
  - For measures before 1988 timing primarily drawn from OECD “Calendar” or “Chronology”.
  - Exact announcement and implementation dates (day, month, year) reported for most PIT, CIT and VAT measures announced starting from 1988.
- Tax types covered (six categories):
  - PIT (Personal income taxes)
  - CIT (Corporate income taxes)
  - VAT (Value added and sales taxes)
  - SSC (Social security contributions)
  - EXE (Excises)
  - PRO (Property taxes)
- Data format and dissemination:
  - Database currently available in excel format.
  - A dedicated web portal with a search engine was expected to go live in July, 2018.

### Key features and innovations of the TPRD
- Granular documentation of direction of changes in rates and tax bases across the six tax types.
- Records whether measures represented major tax changes and whether phased over multiple years.
- Includes governments’ estimates of intended revenue yield for fiscal consolidation years when available (t0–t8 variables for intended revenue yields).
- Captures announcement and implementation dates enabling analysis of implementation lags and anticipation effects.
- Raw textual excerpts preserved to permit user-specific reclassification and customized definitions of “major”.

### Advantages relative to existing databases and methodologies
- Classifies precise nature of discretionary tax policy measures, including changes in exemptions, tax thresholds, and capital gains taxation.
- Decomposes tax policy changes into rate and base effects across multiple tax types.
- Longer time horizon and broader country coverage relative to some existing datasets; complements narrative and survey-based datasets.
- Facilitates applications such as:
  - Assessing whether declines in standard tax rates were accompanied by base broadening.
  - Estimating dynamic macroeconomic effects of tax policy packages.
  - Investigating implementation lags and political economy of timing.
  - Identifying case studies for specific policy changes.

### Stylized facts and empirical findings
- Database coverage and totals:
  - "The database documents 3,285 tax policy measures, equivalent to an average of 5 tax measures per country year."
- Composition and nature of measures:
  - "Most measures entailed a change in the tax base (about 60 percent of total identified measures); ⅔ of these changes implied a decrease in tax liabilities."
  - "Rate increases account for almost half of identified rate changes."
  - "More than 70 percent of all identified tax measures involved “major” tax changes or reforms in a single year."
- Co-occurrence and correlation:
  - "Changes to the tax rate are more frequent when the tax base also changes: 423 country years out of 520 country years in which there a rate change, versus 97 country years where the tax base does not change."
  - "Decreases in tax liabilities occur more frequently when there are also increases in tax liabilities than when tax liabilities are not increased."
- Timing patterns (implementation and anticipation):
  - Average and median implementation lags (major CIT, PIT, VAT): "Average implementation lag: 153 days." "Median implementation lag: 78 days."
  - Anticipation (Mertens and Ravn (2012) cutoff at 90 days):
    - "Over 50 percent of CIT and PIT changes in the sample have been unanticipated."
    - "60 percent of VAT changes in the sample have been unanticipated."
    - "Only ⅓ of all VAT rate changes were anticipated."
    - "Almost 60 percent of tax decreases were unanticipated."
    - "About 40 percent of tax increases were anticipated."
    - "One in every two tax measures that increases tax liabilities was anticipated."
    - "More than ⅔ of VAT decreases were unanticipated."
- Business cycle, consolidation, and election timing:
  - Recessions vs expansions:
    - "Tax reforms were more frequent when economies were in a recession as opposed to an expansion."
    - During expansions: "Decreases in tax liability occurred more often than increases (41 versus 15 percent of country years)."
    - During expansions: "Average measures that increase tax liabilities: 1.7 measures per country year." "Average measures that decrease tax liabilities: 2.7 measures per country year."
  - Fiscal consolidations:
    - "Average number of major tax changes in consolidation periods: 4.3 measures per country year." "Average number during normal fiscal times: 3.9 measures per country year."
    - "Average number of measures that increase tax liability in consolidation periods: 2.0 measures per country year." "Average number in normal times: 1.6 measures per country year."
    - "During consolidation episodes, decreases in tax liabilities happen in conjunction with increases in tax liability in 55 percent of total country years."
  - Elections:
    - "Major tax measures were more common in the twelve months following an election (137 country years) than in the twelve months preceding an election (100 country years)."
    - Average numbers: "Post-election years: 1.9 measures per country year." "Pre-election years: 2.7 measures per country year."
    - Direction: "Tax measures that increase tax liability are more likely in post-electoral periods: 43 percent of 137 country years." "Tax measures that increase tax liability in pre-electoral periods: 15 percent of 100 country years." "Decreases in tax liability in pre-electoral periods: 63 percent of country years."

### Box 1 — Developments in Tax Rates (selected trends)
- Statutory tax rate trends:
  - CIT:
    - Advanced economies (16 countries): "declined from about 44 percent in 1970s to around 26 percent in 2017."
    - Emerging market economies (6 countries): "declined from about 33 percent in 1980s to 22 percent in 2017."
  - PIT:
    - Advanced economies: "declined from nearly 70 percent to 44 percent."
    - Emerging market economies: "declined from 60 percent to 36 percent."
  - VAT:
    - Advanced economies: "average VAT rate increased from 14 to 18 percent."
    - Emerging market economies: "average VAT rate increased from 13 to 16 percent."
- Typical tax package composition and frequency:
  - "A typical annual tax package comprised about 4 tax measures."
  - "Included both rate and base changes in approximately 60 percent of total country years."
  - "Rate changes announced alone occurred in 11 percent of cases."
  - "Countries, on average, announced less than a tax package per year or an average of about 25 tax packages over the entire sample period."
  - "G7 countries (except Japan) announced approximately between 35 and 45 packages over the sample period."
- Implementation lag aggregate statistics:
  - "Average implementation lag for major CIT, PIT, and VAT measures: 153 days." "Median implementation lag: 78 days."
  - Tax-type differences: "VAT measures showed much lower median implementation lags and cross-country variation."

### Limitations and data quality considerations
- Quality of information varies across countries, time periods, and tax types.
- The database:
  - "Does not attempt to measure and compile policy settings across countries."
  - "Does not provide an exhaustive accounting of all tax policy measures introduced by a country over the sample period."
  - Binary coding of changes (except consolidation episodes) limits measurement of magnitude.
  - The exact motivation underlying a tax change is not identified (except for consolidation episodes where expected yields are available).
  - "TPRD is preliminary and work in progress; quality and accuracy vary across tax types, countries, and over time."

### Cross-checks, validation, and coverage special cases
- External databases used for cross-checks:
  - IMF tax rate database (1980-2014)
  - European Commission tax indicator database (1995-2015)
  - Global KPMG tax rates database (2006-2015)
  - USAID Collecting Taxes Database (2007-2012)
- Narrative databases used for country-level checks: Romer and Romer (2009), Cloyne (2013), Gil and others (2017), Pereira and Wemans (2013).
- China and India special handling:
  - Limited OECD Surveys coverage; information integrated with IMF internal documents, IBFD archives, and national sources to expand coverage.
- Cross-check outcomes and categories:
  - Match; Dating problem; Coding problem; Database only (TPRD-only measures not present in external top-rate databases due to TPRD covering many rate types).

### Coding rules, variable definitions, and processing steps
- Coding conventions:
  - Introduction/removal of a tax coded as BASE measure; replacement coded as two BASE measures.
  - Income bracket changes coded as BASE unless specified otherwise.
  - Reduction in number of tax brackets coded as base broadening (Tax_reformtype=BASE and Tax_change=INC).
  - Changes in per unit taxation coded as RATE.
  - Extension or postponement of a measure coded as an actual tax change (e.g., postponement of a rate reduction coded as RATE INC).
- Major rate change threshold:
  - "A rate change larger than 1 percentage point in absolute value was considered a major reform."
- Major base change rules by tax type:
  - CIT major base changes: R&D promotion; investment promotion; loss-carry rules; thin capitalization; capital gains; other base changes considered major if OECD/IBFD language indicates a large change.
  - PIT major base changes: standard relief; child relief; capital gains relief; interest relief; relief for SSC, insurance premiums, and private pensions; other base changes judged by OECD/IBFD language.
  - VAT major base changes: exemptions on food items; exemptions on medical supplies; exemptions on education; other VAT base changes judged by OECD/IBFD language.
  - SSC, EXE, PRO: major if OECD/IBFD suggest change affects large groups or mobilizes significant resources.
- Textual and structural variables:
  - Sentence_OECD, Paragraph_OECDs, Tax_type, Tax_reform (RATE/BASE), Tax_change (INC/DEC), Tax_major (dummy), Base_category (coded 1–6 with specific mappings for CIT, PIT, VAT), Base_category_label, Tax_package, Tax_multiyear, Announcement_date (MM/DD/YYYY with "x" for missing components), Implementation_date (MM/DD/YYYY with "x").
- IBFD matching:
  - Merge_MATCH: 1 = "perfect" match; 2 = "imperfect" match; 3 = "no match".
  - IBFD_Measure: ID of IBFD news clip (accessible to IBFD subscribers).
- Consolidation variables:
  - Consolidation_year dummy (Alesina and others (2017) definition).
  - t0–t8 variables report intended revenue yield for announcement year and up to eight years ahead; Currency_consolidation indicates the currency.

### Text mining and validation pipeline
- Step 1: Text mining of 953 OECD Surveys and 37,943 IBFD news clips using Provalis Prosuit® and a defined tax vocabulary to extract candidate sentences.
- Step 2: Manual assessment of extracted excerpts to validate actual policy changes and classify measures across dimensions (tax type, rate vs base, direction, major, package, multiyear, consolidation).
- Step 3: Document announcement and implementation dates using IBFD, OECD calendars, or the excerpts.
- Step 4: Cross-check PIT, CIT, VAT information against alternative quantitative and qualitative sources; identify gaps for future versions.
- Additional processing notes:
  - Sentences with multiple measures separated so each database line is a unique measure; manual coding processed in Stata and exported to country-specific Excel files.
  - Text-mining proximity rule: maximum distance between keywords set to 5 words in rules used to identify measures.

### Database summary tables and appendix highlights (selected statistics and country patterns)
- Total documented measures and aggregates:
  - Grand totals and final aggregate values reported across appendix tables; overall total reported: "3,285" documented tax policy measures.
- Cross-country heterogeneity:
  - Emerging market averages: Brazil, China, Mexico, and Poland average "3 to 4 tax measures per year."
  - Advanced economy averages: France, United Kingdom, Germany, United States, and Italy average "more than 6 tax measures per year."
- Composition of major reforms:
  - "PIT, CIT, and VAT account for around 80 percent of total major tax changes in the sample."
  - "SSC measures account for 7 percent of total major measures."
  - "About 2/3 of all PIT, CIT, and PRO consisted of base changes."
- Country-level composition examples (BASE vs RATE for major reforms; selected entries):
  - Australia: "CIT BASE 36% / RATE 64% ; EXE BASE 70% / RATE 30% ; PIT BASE 100%."
  - Germany: "CIT BASE 73% / RATE 27% ; EXE BASE 26% / RATE 74%."
  - United Kingdom: "CIT BASE 78% / RATE 22% ; EXE BASE 21% / RATE 79%."
  - United States: "CIT BASE 67% / RATE 33% ; EXE BASE 100%."
- Directional composition examples (DEC vs INC for major reforms; selected entries):
  - Australia: "CIT DEC 45% / INC 55%."
  - Japan: "CIT DEC 94% / INC 6%."
  - Spain: "CIT DEC 75% / INC 25%."
  - Turkey: "CIT DEC 18% / INC 82%."

### Implications and recommended use
- Key caution for empirical work:
  - "Ignoring tax base changes when analyzing rate changes can bias estimated economic effects of rate changes."
- Recommended applications:
  - Use the TPRD to analyze base vs rate changes, anticipation effects via announcement vs implementation timing, political economy timing (elections, recessions, consolidations), and to select case studies for policy design and sequencing.
- User customization:
  - Database contains underlying tax excerpts and fields enabling users to customize definitions of “major” and to generate tailored datasets for specific research questions.
- Forward work suggested:
  - Extending country and time coverage.
  - Adding details on included or currently excluded tax measures.
  - Closing identified information gaps and updating SSC, EXE, PRO timing documentation in future versions.

*Source: wp18110 - References (Tax Policy Reform Database overview and appendices).*

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

### wp18110 - References

### Overview and purpose
- Introduces the Tax Policy Reform Database (TPRD), a comprehensive database of tax policy measures adopted in 23 advanced and emerging market economies over the last four decades.
- Motivation: detailed study of past tax reforms and their impacts to inform future reforms and policy advice.

### Data sources and scope
- Primary documentary sources:
  - "more than 900 Organization for Economic Cooperation and Development (OECD) Economic Surveys"
  - 37,000 tax-related news from the archives of the International Bureau of Fiscal Documentation (IBFD)
  - Explicit count: 953 OECD Economic Surveys examined for 23 countries.
- Time coverage highlights:
  - Announcement and implementation dates (day and/or month and year) documented where available, with especially detailed coverage for 1988-2014 for PIT, CIT, and VAT measures.
  - For periods preceding 1988, timing information primarily drawn from the OECD “Calendar” or “Chronology” of main economic events.
- Tax types covered (six categories):
  - Personal income taxes (PIT)
  - Corporate income taxes (CIT)
  - Value added and sales taxes (VAT)
  - Social security contributions (SSC)
  - Excises (EXE)
  - Property taxes (PRO)
- Data format and dissemination:
  - Database currently available in excel format.
  - A dedicated, user-friendly web portal with a search engine was expected to go live in July, 2018.

### Key features and innovations of the TPRD
- Granular documentation of the direction of changes in rates and tax bases for the six tax types listed above.
- Records whether measures represented major tax changes (e.g., tax reforms) and if they were phased over multiple years.
- Includes governments’ estimates of intended revenue yield for fiscal consolidation years when available.
- Captures announcement and implementation dates to enable analysis of implementation lags and anticipation effects.
- Raw data structured to permit more granular categorizations as needed.

### Advantages relative to existing databases and methodologies
- Enables classification of precise nature of discretionary tax policy measures, including changes in exemptions, tax thresholds, and capital gains taxation.
- Decomposes tax policy changes into rate and base effects across multiple tax types, extending approaches used in related studies.
- Longer time horizon and broader country coverage compared with some existing datasets (e.g., European Commission TEDB), and complements other narrative and survey-based datasets.
- Facilitates applications such as:
  - Assessing whether declines in standard tax rates were accompanied by base broadening.
  - Estimating dynamic macroeconomic effects of tax policy packages and synergies across reforms.
  - Investigating effects of differing implementation lags and the political economy of timing.
  - Identifying case studies for specific policy changes (e.g., VAT or PIT thresholds).

### Stylized facts and empirical findings
- Composition and nature of measures:
  - The majority of tax policy measures in the sample affect the tax base rather than tax rates.
  - Many measures are part of broader tax policy reform packages.
- Timing and political/economic context regularities:
  - Tax increases occur relatively more frequently in periods of economic recessions and post-election years than in expansions and in the run-up to elections, respectively.
  - In episodes of fiscal consolidation (as defined in Alesina and others (2017) and Dabla-Norris and Lima (forthcoming)), tax increases are often offset by tax decreases, suggesting policymakers attenuate distortionary and/or distributional effects of higher taxes.
- Coverage and heterogeneity:
  - Significant heterogeneity across countries and tax types in the composition, timing, and direction of measures.
  - PIT, CIT, and VAT measures introduced between 1988-2014 offer the most comprehensive coverage in the database.

### Limitations and data quality considerations
- Quality of information varies across countries, time periods, and types of tax measures.
- The database:
  - Does not attempt to measure and compile policy settings across countries.
  - Does not provide an exhaustive accounting of all tax policy measures introduced by a country over the sample period.
- Forward work suggested:
  - Extending country and time coverage.
  - Adding details on included or currently excluded tax measures.
  - Closing identified information gaps.

### Applications and research potential
- Use cases highlighted:
  - Empirical analysis of tax policy effects accounting for base vs. rate changes.
  - Studies of anticipation effects leveraging announcement vs. implementation timing.
  - Political economy analysis of timing (election cycles, recessions, consolidations).
  - Selection of case studies for policy design and reform sequencing.

*Source: wp18110 - References (Tax Policy Reform Database overview and appendices).*

### annex of OECD Surveys until 2003-2005. Compared to the IBFD, OECD calendars typically provide

### wp18110 - annex of OECD Surveys until 2003-2005. Compared to the IBFD, OECD calendars typically provide

### Data sources and timing precision
- OECD calendars typically provide less detail on the timing of tax changes (i.e., the day of implementation/announcement is often unavailable), resulting in lower precision in dating measures before 1988 than for measures adopted after 1988.
- When timing information is not found in IBFD or OECD calendars, timing is retrieved by assessing textual fragments of the OECD Surveys, which typically allows identification of announcement and implementation years.
- In the current version of the database:
  - Exact announcement and implementation dates (i.e., day, month, year) are reported for most PIT, CIT and VAT measures announced starting from 1988.
  - CIT, PIT and VAT measures in 1988-2014 typically feature exact announcement and implementation dates (i.e., day, month, year) based on IBFD information, while CIT, PIT, and VAT measures prior to 1988 generally display less precise announcement and implementation dates (i.e., month, year) based on OECD calendars.
  - SSC, EXE, and PRO for the period 1988-2014 are being documented for precise timing and will feature in the next version of the database; current SSC, EXE, and PRO measures typically report announcement and implementation dates based on OECD calendars or the excerpts themselves and are hence less precise than IBFD-derived information.

### Cross-checks and validation
- Identified tax measures were cross-checked against external indicators to ensure accuracy and detect information gaps.
- Databases used for cross-checks include:
  - IMF tax rate database (1980-2014)
  - European Commission tax indicator database (1995-2015)
  - Global KPMG tax rates database (2006-2015)
  - USAID collecting taxes database (2007-2012)
- For PIT, CIT, and VAT base measures dated with IBFD news, checks involved comparing OECD Survey information with detailed information from the IBFD archives.
- Some country information was checked against well-established narrative databases.
- Cross-checks for SSC, EXE, and PRO taxes are ongoing; potential future revisions in coding and/or timing will be reflected in the next database version.
- The checks improve quality but do not rule out the risk of inconsistencies and/or omissions in the database.

### Coverage considerations and special cases
- Countries covered in the database include: Australia, Austria, Brazil, Canada, China, Czech Republic, Denmark, France, Greece, Germany, India, Ireland, Italy, Japan, Korea, Luxembourg, Mexico, Poland, Portugal, Spain, Turkey, United Kingdom, and the United States.
- China and India are special cases because only a handful number of OECD Surveys is available for these countries. To increase coverage:
  - Information from the Surveys was integrated with hand-picked information from alternative sources, such as IMF internal documents, IBFD archives, and national sources.
  - This integration significantly expanded coverage of tax policy measures for these countries.
- OECD Country Surveys are available at http://www.oecd-ilibrary.org/economics/oecd-economic-surveys_16097513 and are covered by standard copyrights.

### Steps involved in constructing the Tax Policy Reform Database (TPRD)
- Step 1: Process information contained in the 953 OECD Surveys and 37, 943 IBFD news clips to identify excerpts describing changes in any of six taxes (PIT, CIT, VAT, SSC, EXE, PRO) using a system of text-based rules.
  - The text-mining software Provalis Prosuit® was used to perform this task.
  - A “tax vocabulary” codified how policy changes are typically described in OECD Surveys (see Appendix C for details).
- Step 2: Manually assess which extracted OECD excerpts constitute actual policy changes; validated changes were classified along dimensions including:
  - Type of tax (e.g., CIT, VAT)
  - Type of change (rules governing the tax base and tax rates)
  - Direction of change (increase, decrease)
  - Whether the measure represented a major tax change (“reform”)
  - Whether the measure was announced as part of a package
  - Whether the measure was phased over several years
  - Whether the measure was introduced in a consolidation year
- Step 3: Document announcement and implementation dates of each classified measure using IBFD archives, OECD calendars, or the excerpts themselves if IBFD/OECD were unavailable.
- Step 4: Check accuracy of PIT, CIT, and VAT information against quantitative and qualitative information on rate and base changes from alternative data sources (see Appendix A and Appendix E). These checks identify information gaps to be addressed in future versions.

### Identifying tax policy measures and classification rules
- The database includes heterogeneous measures ranging from major reforms (e.g., introduction of VAT in France; overhaul of PIT in Italy and Poland; reform of CIT in Ireland) to minor measures (small rate changes, adoption of tax expenditures).
- Measures were differentiated between major tax changes with potentially large macro-fiscal implications and measures with limited economic effects by:
  - Applying a threshold for rate changes: a rate change larger than 1 percentage point in absolute value was considered a major reform.
    - This threshold balances accuracy and parsimony despite being arbitrary.
  - Relying on informed judgement for base changes using all available information from OECD Surveys, IBFD archives, and analytical studies.
- For excise taxes (per unit), determination of whether a change is major is based on assessment of the language used in the OECD report.
- For base changes:
  - A major tax reform is identified when the change in the tax base affects a large group of taxpayers or has the potential to mobilize significant resources.
  - A change that broadens the tax base (defined as increasing tax revenue holding constant the statutory tax rate, other tax base aspects and the behavior of economic agents) is coded as an “increase”; a change that reduces the tax base is denoted as a “decrease”.
  - Detailed information on specific tax base changes and the legislative/regulatory actions is captured in text format in the database.
- A number of conventions were followed in classifying different tax changes (further details in Appendices).

*Source: wp18110 - annex of OECD Surveys until 2003-2005. Compared to the IBFD, OECD calendars typically provide (IMF Working Paper annex text provided).*

### introduction/removal of a tax was coded as a base measure and so were changes in income

### wp18110 - introduction/removal of a tax was coded as a base measure and so were changes in income brackets (unless specified otherwise)

### Coding rules for measures and interpretation of timing
- Introduction/removal of a tax was coded as a base measure and so were changes in income brackets (unless specified otherwise).
- A reduction in the number of tax brackets was coded as a base broadening measure following the assumption that simplification can boost compliance.
- The extension or postponement of a tax measure (e.g., a temporary surcharge is maintained for an additional year, the reduction in PIT rate is delayed) were coded as an actual tax change aimed at avoiding the effects of the planned tax change.
  - Example: the postponement of a rate reduction was coded as a rate increase because absent such a postponement, the rate would have been lower.

### Classification approach for rate changes (PIT, CIT, VAT)
- PIT rate change categories considered (will be included in next TRPD version):
  - 1) Statutory rates
  - 2) Top rate
  - 3) Bottom rate
  - 4) Surcharges
  - 5) Capital gains
  - 6) Dividends
  - 7) Other
- CIT rate change categories:
  - 1) Statutory rates
  - 2) Top rate
  - 3) Surcharges
  - 4) Capital gains
  - 5) Dividends
  - 6) Other
- VAT rate change categories:
  - 1) Standard rate
  - 2) Reduced rates
  - 3) Other

### Definition and classification of major base changes by tax type
- CIT base change pertains to any of the following:
  - R&D promotion (e.g., tax credit)
  - Investment promotion (e.g., depreciation rules)
  - Loss-carry rules
  - Thin capitalization
  - Capital gains
  - If not in these categories (e.g., generic exemptions), classified as “other base changes”; considered “major” only if OECD Survey or IBFD archives suggest a significant reform.
- PIT base change pertains to any of the following:
  - Standard relief (e.g., single person or family deductions, tax credits)
  - Child relief (e.g., tax credit, deductions)
  - Relief on capital gains
  - Interest relief
  - Relief for SSC, insurance premiums, and private pensions
  - Other base changes (e.g., deductions for special purposes) classified as “other base changes”; may or may not be considered major depending on OECD/IBFD information.
- VAT base change pertains to any of the following:
  - Exemptions on food items
  - Exemptions on medical supplies
  - Exemptions on education
  - All other VAT base changes (e.g., introduction of VAT, generic exemptions) classified as “other base changes”; considered major only if OECD/IBFD corroborate.
- SSC, EXE, or PRO base change:
  - Classified as “major” when OECD Surveys and/or IBFD archives suggest the change affects large groups of taxpayers or could potentially mobilize significant resources.
  - The criterion is arbitrary and reflects lack of consensus in literature on what constitutes a major SSC, EXE, or PRO base change.
- The database contains all underlying tax excerpts, allowing users to customize definitions of “major” change.

### Additional coding fields and package/phasing information
- A dummy variable (1 or 0) identifies whether the coded measure is part of a package if explicitly mentioned in the OECD report or if measures share the same announcement date.
- A dummy identifies whether the measure was phased over several years (i.e., multiyear).
- For tax measures announced during a consolidation period (as defined in Alesina and others (2017)), the TPRD provides expected revenue yield of each measure as reported in Dabla-Norris and Lima (forthcoming) by associating consolidation measures and related expected revenue yields with corresponding measures in the tax measures database (see Appendix B for details).

### Caveats and limitations of the TPRD
- The approach does not rely on a single common metric to identify tax base changes; more judgement involved in deeming a measure a major base change compared with rate-change-based studies.
- The database provides no information regarding the stance of current (or past) tax policy or tax structure.
- Tax changes (other than in consolidation episodes) are identified as a binary dummy rather than continuous variables, which does not allow measurement of the size of the change.
- The exact motivation underlying a tax change is not identified (except for consolidation episodes where expected yields are available).
- TPRD is preliminary and work in progress; quality and accuracy vary across tax types, countries, and over time, with more detail available for recent decades.

### Examples of different rate and base changes by tax (selected excerpts from OECD/IBFD sources)
- Examples include:
  - DNK: "A tax credit for r&d activities of some enterprises has been proposed in the fiscal bill for 2012" — CIT BASE DEC 11/8/24/2011 1/1/2012
  - CHN: "Withholding tax on dividends cut from 5% to zero" — CIT RATE DEC 11 12/25/2006 1/1/2008
  - LUX: "An increase in the solidarity levy on corporate income tax" — CIT RATE INC 0 15/2/2006 1/1/2007
  - MEX: "The tax reform approved in 2013 included the introduction of taxes on high-caloric foods and sweetened beverages." — EXE BASE INC 10 x /x /2013 x /x /2013
  - TUR: "Petroleum consumption tax rate is increased from 26 to 31.5 per cent." — EXE RATE INC 10 x /x /1990 x /x /1990
  - GRC: "The tax reform in 2008 included further cuts in personal income taxes, reducing the two middle marginal rates of 29% and 39% by four percentage points between 2007 and 2009, to 25% and 35% respectively, and measures to broaden the tax base, such as the imposition of a 10% tax rate on dividends and capital gains." — PIT BASE INC 11 8/28/2008 1/1/2009
  - USA: "The social security tax rate was increased from 11,7 per cent to 12,1 per cent, the taxable earnings base was increased from $16 500 to $17000" — SSC RATE INC 00 x /x /1977 1/x /1978
  - IRL: "An income levy was introduced and new taxes imposed... the vat rate was increased by 0.5 percentage points." — VAT RATE INC 01 10/14/2008 12/1/2008
- Note: announcement and implementation dates use an “x” when day or month is not available (e.g., 4/x/1989).

### Stylized facts and database summary statistics
- The database documents 3,285 tax policy measures, equivalent to an average of 5 tax measures per country year.
- Cross-country heterogeneity:
  - Emerging market economies (Brazil, China, Mexico, and Poland) average 3 to 4 tax measures per year.
  - Advanced economies (France, United Kingdom, Germany, United States, and Italy) average more than 6 tax measures per year.
- Composition of measures (Table 2 summary):
  - Most measures entailed a change in the tax base (about 60 percent of total identified measures); ⅔ of these changes implied a decrease in tax liabilities.
  - Rate changes: rate increases account for almost half of identified rate changes.
  - More than 70 percent of all identified tax measures involved “major” tax changes or reforms in a single year.
  - Among major reforms:
    - A decrease in the tax base introduced as part of a policy package in a single year accounted for 20 percent of all identified tax measures.
    - A base increase accounted for 12 percent of total identified measures.
    - Major rate decreases introduced as part of a package in a single year represented 10 percent of total identified measures.
- Country-year distribution (Table 3 summary):
  - Base measures occurred more often than rate measures in the whole sample (575 versus 520 country years out of 672 country years).
  - Base decreases outnumbered base increases.
  - Rate increases were more frequent in the case of minor measures (132 versus 101 country years) and measures not announced as part of a tax package (183 versus 167 country years).
- Co-occurrence (Table 4 summary):
  - Changes to the tax rate are more frequent when the tax base also changes: 423 country years out of 520 country years in which there a rate change, versus 97 country years where the tax base does not change.
  - Decreases in tax liabilities occur more frequently when there are also increases in tax liabilities than when tax liabilities are not increased.
  - These patterns align with Kawano and Slemrod (2016) for CIT base and rate changes and highlight potential bias if base changes are ignored when assessing rate-change effects.
- Major tax measures (over 80 percent of observed tax policy changes or 95 percent of country years occurrences):
  - PIT, CIT, and VAT account for around 80 percent of total major tax changes in the sample.
  - SSC measures account for 7 percent of total major measures.
  - EXE and PRO measures occur less frequently.
  - Most PIT, CIT, and PRO measures comprise base changes; rate changes are more salient for VAT, SSC, and EXE.
    - About 2/3 of all PIT, CIT, and PRO consisted of base changes.
  - Breakdown by direction:
    - Reduction in tax liability was more common for major CIT and PIT reforms.
    - Increases in tax liabilities were more pronounced for SSC, VAT, EXE, and PRO reforms.
  - Geographic/country patterns:
    - In most countries major PIT, CIT, and PRO measures were dominated by base changes, except Japan and China for PIT; Ireland and Luxemburg for CIT; China, Portugal, Denmark, and Czech Republic for PRO.
    - Major VAT base changes outnumbered rate changes in approximately 1/3 of the sample.
    - Major SSC and EXE measures were typically dominated by rate change, except Spain, Italy, France, and Turkey for SSC; and Portugal for excises.
  - Distributional consistency:
    - Frequency of major reductions in tax liabilities of PIT and CIT common across all countries, except Poland and Portugal for PIT, and Austria and Greece for CIT.
    - SSC and PRO measures more evenly distributed between increases or decreases in tax liabilities, with some countries experiencing only increases or only decreases for these taxes (examples noted in source).

### Implications and recommended use
- Ignoring tax base changes when analyzing rate changes can bias estimated economic effects of rate changes.
- The TPRD provides a comprehensive database of occurrences of changes in multiple aspects of tax bases for different tax types, enabling users to:
  - Customize definitions of “major” change.
  - Generate new databases tailored to specific research questions.
- Users should account for:
  - Binary coding of changes (except consolidation episodes) which limits measurement of magnitude.
  - Variation in data quality across tax types, countries, and time.

*Source: Tax Policy Reform Database, OECD, IBFD; excerpts from wp18110 (TPRD documentation).*

### Box 1. Developments in Tax Rates

### Box 1. Developments in Tax Rates

### Trends in statutory tax rates
- Corporate income tax (CIT) rates:
  - Advanced economies (16 countries): declined from about 44 percent in 1970s to around 26 percent in 2017.
  - Emerging market economies (6 countries): declined from about 33 percent in 1980s to 22 percent in 2017.
- Personal income tax (PIT) rates:
  - Advanced economies: declined from nearly 70 percent to 44 percent (same period).
  - Emerging market economies: declined from 60 percent to 36 percent (same period).
- Value-added tax (VAT) rates increased:
  - Advanced economies: average VAT rate increased from 14 to 18 percent.
  - Emerging market economies: average VAT rate increased from 13 to 16 percent.

### Composition of tax packages and reforms
- A typical annual tax package:
  - Comprised about 4 tax measures.
  - Included both rate and base changes in approximately 60 percent of total country years.
  - Rate changes announced alone occurred in 11 percent of cases.
  - 24 percent of total country years combined at least one measure to increase a tax rate with at least one base narrowing measure.
  - Approximately 26 percent of total country years combined at least one measure to decrease a tax rate with at least a base broadening tax change.
- Frequency of tax packages:
  - Countries, on average, announced less than a tax package per year or an average of about 25 tax packages over the entire sample period.
  - G7 countries (except Japan) announced approximately between 35 and 45 packages over the sample period.

### Implementation lags and anticipation of measures
- Implementation lag definition:
  - Measured in number of days: implementation date minus announcement day.
  - Positive lag reflects lead time for legislative process; negative lag indicates retroactive implementation.
- Aggregate statistics:
  - Average implementation lag for major CIT, PIT, and VAT measures: 153 days.
  - Median implementation lag for major CIT, PIT, and VAT measures: 78 days.
  - Average implementation lag and variation reflect presence of retroactive introductions (negative lags).
- Tax-type differences:
  - CIT measures featured higher variation in implementation lags than PIT measures across countries, though CIT and PIT showed similar median lags.
  - VAT measures showed much lower median implementation lags and cross-country variation.
  - Since the global financial crisis: median implementation lag and variation declined for PIT and CIT (less pronounced for CIT); median implementation lag and variation for VAT increased.
- Country heterogeneity (implementation effectiveness and dispersion):
  - China and Turkey: average implementation lag below 40 days for PIT and CIT.
  - Czech Republic and Germany: faced longer implementation delays.
  - Australia, Germany, and Japan: higher dispersion consistently across different tax types.
  - Japan, India, and Australia: significantly high variation in implementation lags for VAT.
- Anticipated versus unanticipated (following Mertens and Ravn (2012) cutoff at 90 days):
  - Over 50 percent of CIT and PIT changes in the sample have been unanticipated.
  - 60 percent of VAT changes in the sample have been unanticipated.
  - Only ⅓ of all VAT rate changes were anticipated.
  - Almost 60 percent of tax decreases were unanticipated.
  - About 40 percent of tax increases were anticipated.
  - One in every two tax measures that increases tax liabilities was anticipated.
  - More than ⅔ of VAT decreases were unanticipated.

### Timing of tax measures: recessions, consolidations, and elections
- General timing patterns:
  - Most CIT and PIT measures were announced between the late 1980s and early 2000s; VAT changes more evenly distributed across years (except early 1990s peak).
- Average frequency and composition:
  - On average, countries announced 3.8 major CIT, PIT, and VAT changes per country year; half of these changes typically consisted of PIT measures.
- Recessions versus expansions:
  - Tax reforms were more frequent when economies were in a recession as opposed to an expansion (Table 6).
  - The average number of tax measures announced in periods of economic expansion was slightly higher than in periods of recession.
  - Decreases in tax liability occurred more often than increases during expansions (41 versus 15 percent of country years).
  - During expansions, average measures that increase tax liabilities: 1.7 measures per country year.
  - During expansions, average measures that decrease tax liabilities: 2.7 measures per country year.
  - These patterns suggest a possible pro-cyclical bias in tax policymaking in the sample.
- Fiscal consolidation episodes:
  - Major measures are less frequent in consolidation periods than in normal times, but consolidation periods show more pronounced tax policy activism:
    - Average number of major tax changes in consolidation periods: 4.3 measures per country year.
    - Average number during normal fiscal times: 3.9 measures per country year.
    - Average number of measures that increase tax liability in consolidation periods: 2.0 measures per country year.
    - Average number of measures that increase tax liability in normal times: 1.6 measures per country year.
  - During consolidation episodes, decreases in tax liabilities happen in conjunction with increases in tax liability in 55 percent of total country years.
  - Interpretation: governments may use decreases as “sweeteners” to buy political support for fiscally consolidating measures.
- Electoral timing:
  - Major tax measures were more common in the twelve months following an election (137 country years) than in the twelve months preceding an election (100 country years).
  - Average number of measures:
    - Post-election years: 1.9 measures per country year.
    - Pre-election years: 2.7 measures per country year.
  - Directional composition:
    - Tax measures that increase tax liability are more likely in post-electoral periods: 43 percent of 137 country years.
    - Tax measures that increase tax liability in pre-electoral periods: 15 percent of 100 country years.
    - Decreases in tax liability in pre-electoral periods: 63 percent of country years.
  - Interpretation: incumbent governments avoid announcing unpopular tax increases before elections and may announce decreases prior to elections.

### Key stylized facts and conclusions
- Database characteristics:
  - The Tax Policy Reform Database (TPRD) identifies tax policy changes in six types of taxes (CIT, PIT, VAT, SSC, EXE, and PRO) for 23 advanced and emerging countries.
  - Classifications include rate vs base change, increase vs decrease in tax liabilities, major tax change designation, whether part of a broader tax package, phased measures, and measures announced in the context of fiscal consolidation.
- Stylized facts demonstrated using the database:
  - Changes to the tax base are frequent and typically accompany rate changes.
  - Tax policy measures are often part of broader reform packages.
  - In most advanced and emerging economies, PIT changes are most frequent, followed by CIT and VAT changes.
  - The average implementation lag of PIT, CIT, and VAT measures is around 2-5 months, though this differs across tax types and countries.
  - Significant cross-country heterogeneity exists in the nature and timing of tax reforms.
  - Timing matters:
    - Number and direction of tax policy changes differ in recessions vs expansions, during fiscal consolidations, and around elections.
    - Number of announced tax policy changes tends to be markedly lower before elections.
    - Average number of measures that decrease tax liabilities is higher during expansions than during recessions, potentially indicating pro-cyclical bias.
    - During fiscal consolidations, governments tend to adopt measures that both increase and decrease tax liabilities.
    - Decreases in tax liability far outnumber increases in pre-electoral periods.

*Source: Box 1, "Developments in Tax Rates", wp18110.*

### REFERENCES

### REFERENCES

### References cited
- Alesina A., O. Barbiero, C. Favero, F. Giavazzi, and M. Paradisi, 2017, "The Effects of Fiscal Consolidations: Theory and Evidence," NBER Working Papers 23385 (Cambridge, Massachusetts: National Bureau of Economic Research, Inc.).
- Alesina, A., C. Favero, and F. Giavazzi, 2015, “The Output Effect of Fiscal Consolidation Plans”, Journal of International Economics, Vol. 96, pp. S19-S42.
- Auerbach, A. J. and Y. Gorodnichenko, 2012, "Measuring the Output Responses to Fiscal Policy," American Economic Journal: Economic Policy, Vol. 4, pp. 1-27.
- Cloyne, J. S., 2013, “Discretionary Tax Changes and the Macroeconomy: New Narrative Evidence from the United Kingdom,” American Economic Review, 103, 4, 1507-28.
- Dabla-Norris, E., and F. Lima, Forthcoming, “The Macroeconomic Effects of Tax Changes: Evidence from Fiscal Consolidations, IMF Working Paper (Washington: International Monetary Fund).
- Duval, R., D. Furceri, J. Jalles, B. Hu and H. Nguyen, 2018, “A Narrative Database of Major Labor and Product Market Reforms in Advanced Economies,” IMF Working Paper No. 18/19 (Washington: International Monetary Fund).
- Gentzkow, M., B. T. Kelly, and M. Taddy, 2017, "Text as Data," NBER Working Papers 23276 (Cambridge, Massachusetts: National Bureau of Economic Research, Inc.).
- Gil, P., F. Martí, R. Morris, J. J. Pérez, and R. Rarnos, 2017, “The Output Effects of Tax Changes: Narrative Evidence from Spain,” Documentos de Trabajo No. 1721 (Madrid: Banco de España).
- IMF, Forthcoming, “Tax Policy Assessment Framework (TPAF)” (Washington: International Monetary Fund).
- IMF, 2017, “IMF Fiscal Monitor: Achieving More with Less,” (Washington: International Monetary Fund).
- IMF, 2016, “IMF Fiscal Monitor: Acting Now, Acting Together,” (Washington: International Monetary Fund).
- Leeper, E. M., T. B. Walker, and S. S. Yang, 2013, "Fiscal Foresight and Information Flows," Econometrica, Vol. 81, pp. 1115-1145.
- Kawano, L. and J. Slemrod, 2016, “How do Corporate Tax Bases Change when Corporate Tax Rates Change? With Implications for the Tax Rate Elasticity of Corporate Tax Revenues,” International Tax and Public Finance, Vol. 23, pp. 401–433.
- Kremer, J., Braz, C.R., Brosens, T., Langenus, G., Momigliano, S. and Spolander, M., 2006, “A disaggregated framework for the analysis of structural developments in public finances,” ECB Working Paper Series, No. 579.
- Mertens, K. and M. O. Ravn, 2012, "Empirical Evidence on the Aggregate Effects of Anticipated and Unanticipated US Tax Policy Shocks," American Economic Journal: Economic Policy, Vol. 4, pp. 145-181.
- OECD, 2006, “Fundamental Reform of Personal Income Tax”, OECD Tax Policy Studies No. 13, (Paris: Organization for Economic Co-operation and Development).
- OECD, 2010, “Tax Policy Reform and Economic Growth,” OECD Tax Policy Studies No. 20, (Paris: Organization for Economic Co-operation and Development).
- OECD, 2016a, “Consumption Tax Trends 2016: VAT/GST and Excise Rates, Trends and Policy Issues,” (Paris: Organization for Economic Co-operation and Development).
- OECD, 2016b, “Taxing Wages,” (Paris: Organization for Economic Co-operation and Development).
- Pereira, M. C. and L. Wemans, 2013, “Output Effects of a Measure of Tax Shocks based on Changes in Legislation for Portugal,” Working Papers 15 (Lisbon: Banco de Portugal).
- Ramey, V.A., 2011, “Identifying Government Spending Shocks: It's all in the Timing,” The Quarterly Journal of Economics, Oxford University Press, Vol. 126, pp. 1-50.
- Ramey, V. A. and M. D. Shapiro, 1998, “Costly Capital Reallocation and the Effects of Government Spending,” Carnegie-Rochester Conference Series on Public Policy, Vol. 48, pp. 145-194.
- Ramey, V. A. and S. Zubairy, 2014, “Government Spending Multipliers in Good Times and in Bad: Evidence from U.S. Historical Data,” NBER Working Papers 20719 (Cambridge, Massachusetts: National Bureau of Economic Research, Inc.).
- Riera-Crichton, D., C. A. Vegh, and V. Guillermo, 2015, “Procyclical and Countercyclical Fiscal Multipliers: Evidence from OECD Countries,” Journal of International Money and Finance, Vol. 52, pp. 15-31.
- Romer, C. D. and D. H. Romer, 2010, "The Macroeconomic Effects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks," American Economic Review, Vol. 100, pp. 763-801.
- Romer, C. D. and D. H. Romer, 2009, “A Narrative Analysis of Postwar Tax Changes,” (Berkeley: University of California, Berkeley).
- Vegh, C. A. and G. Vuletin, 2015, “How is Tax Policy Conducted over the Business Cycle?" American Economic Journal: Economic Policy, Vol. 7, pp. 327-70.

*This document also contains the Appendix describing the TPRD data sources and variable definitions (see sections below).*

---

### Data Sources and Coverage

### Countries and primary sources
- Database covers tax policy reforms in 23 countries: Australia, Austria, Brazil, Canada, China, Czech Republic, Denmark, France, Greece, Germany, India, Ireland, Italy, Japan, Korea, Luxembourg, Mexico, Poland, Portugal, Spain, Turkey, United Kingdom, and the United States.
- Primary sources:
  - OECD Economic Surveys (publicly available for download from the OECD website).
  - IBFD archives (news clips; detailed tax information from 1988 onwards).

### Tax rate databases used for checks
- IMF Tax Rate Database: statutory top rates for CIT, PIT, and VAT; covers 1980 to 2015.
- European Commission Tax Indicator Database: top statutory CIT and PIT rates, standard and reduced VAT rates, labor-implicit tax rates, consumption-implicit tax rates; covers 1995 to 2015.
- Global KPMG Tax Rates: secondary source for CIT, PIT, VAT, social security rates for employees and indirect rates; covers 2006 to 2015.
- USAID Collecting Taxes Database: supplementary source for CIT, PIT, VAT; covers 2007 to 2012.

### Other narrative databases used for cross-checks
- Country-level narrative databases considered: United States (Romer and Romer, 2009), United Kingdom (Cloyne, 2013), Spain (Gil and others, 2017), Portugal (Pereira and Wemans, 2013).
- Figure Appendix E.3 (in source) lists countries and years with possible information gaps.

---

### Definitions and Variable Structure

### Textual variables
- Sentence_OECD: textual excerpt from OECD Surveys mentioning changes in any of six tax types (PIT, CIT, VAT, SSC, EXE, PRO). Excerpts vary in specificity (e.g., generic statements vs. explicit rate changes such as "increased the VAT standard rate from 15% to 20%").
- Paragraph_OECDs: paragraph from which Sentence_OECD is taken.

### Tax types (following GFSM 2014)
- Tax_type covers six taxes defined per GFSM 2014:
  - PIT: GFSM items 1111 and portion of 1113 referring to personal taxation.
  - CIT: GFSM item 1112 and portion of 1113 referring to corporate taxation.
  - VAT: GFSM items 11411, 11412, 11413.
  - EXE: GFSM item 1142.
  - SSC: GFSM items 112 and 12.
  - PRO: GFSM item 113.
- Excluded GFSM items: 11414, 1143, 1144, 1145, 1146, 115, 116, 13, and 14.

### Tax measure classification variables
- Tax_reform type: categorical; RATE or BASE. Changes in per unit (specific) taxes classified as RATE.
- Tax_change: categorical; INC or DEC.
- Tax_major: dummy = 1 if change is major.
  - Major rate change: ΔRATE ≥ 1pp in absolute terms, or textual description explicitly calling it "major".
  - For per unit taxes, majority based on sentence/paragraph language.
- Major base change rules (by tax):
  - CIT major base changes include changes affecting: R&D promotion; investment promotion; loss-carry rules; thin capitalization; capital gains. Other CIT base changes classified as 'other base changes' and deemed major if language indicates large change or large potential revenue.
  - PIT major base changes include changes affecting: standard relief; child relief; relief on capital gains; interest relief; relief for SSC, insurance premiums, and private pensions. Other PIT base changes classified as 'other base changes' and deemed major if language indicates large impact.
  - VAT major base measures: exemptions on food items, medical supplies, and education (selected from common exemptions in OECD countries). Other VAT base changes classified as 'other base changes'. Exemptions defined as "input taxed supplies".
  - SSC, EXE, PRO: major if they affect large groups of taxpayers or have potential to mobilize significant resources, assessed from OECD Surveys and IBFD archives.

### Base category variables
- Base_category: numeric variable relevant for PIT, CIT, VAT base changes; values 1 to 6 with specific mappings:
  - CIT base changes:
    - 1: R&D promotion
    - 2: Investment promotion
    - 3: Loss-carry rules
    - 4: Thin capitalization
    - 5: Capital gains
    - 6: Other CIT base changes
  - PIT base changes:
    - 1: Standard relief
    - 2: Child relief
    - 3: Capital gains
    - 4: Interest relief
    - 5: SSC, pension, and insurance relief
    - 6: Other PIT base changes
  - VAT base changes:
    - 1: Exemptions on food
    - 2: Exemptions on medical supply
    - 3: Exemptions on education
    - 4: Other VAT base changes
- Base_category_label: textual labels corresponding to Base_category values.

### Package and timing variables
- Tax_package: dummy = 1 if measure is part of a package (announced with other tax measures or same announcement date).
- Tax_package_id: identifier numbering packages sequentially within a country.
- Tax_multiyear: dummy = 1 if measure is phased over several years.
- Announcement_date: format MM/DD/YYYY; missing components indicated with "x" (e.g., 4/x/1989).
- Implementation_date: format MM/DD/YYYY; missing components indicated with "x" (e.g., x/x/1995); measures announced but never implemented show "x/x/x".

### IBFD matching and identification
- Merge_MATCH (for PIT, CIT, VAT measures with IBFD dates, 1988-2014):
  - 1 = "perfect" match between Sentence_OECD and IBFD news clip.
  - 2 = "imperfect" match (e.g., similar measure, different size, missing implementation date).
  - 3 = "no match".
- IBFD_Measure: ID of IBFD news clip reporting announcement/implementation (accessible only to IBFD subscribers).

### Consolidation and revenue yield variables
- Consolidation_year: dummy = 1 if announcement year is a consolidation year as defined in Alesina and others (2017) and Dabla-Norris and Lima (forthcoming).
- t0–t8: numeric variables reporting intended revenue yield of a measure or set of measures in:
  - t0 = year of announcement,
  - t1 = one year ahead,
  - t2 = two years ahead,
  - ...
  - t8 = eight years ahead,
  - Applicable in consolidation years (when Consolidation_year = 1).
- Currency_consolidation: variable indicating currency of t0–t8.

---

### Coding Conventions and Conventions for Specific Cases

- Introduction/removal of a tax: coded as BASE measures. Replacement of taxes with a new tax = two BASE measures (removal + introduction).
- Income bracket changes: coded as BASE unless text specifies otherwise.
- Reduction in number of tax brackets: coded as base broadening (Tax_reformtype=BASE and Tax_change=INC), following assumption that simplification boosts compliance.
- Changes in per unit taxation: coded as RATE change.
- Extension or postponement of a tax measure (e.g., temporary surcharge maintained for an additional year, reduction in PIT rate delayed): coded as an actual tax change aimed at avoiding effects of planned change (e.g., delay in reduction of PIT rates coded as RATE INC).

---

### Data Construction Steps

- Step 1. Text Mining:
  - OECD Economic Surveys downloaded in PDF.
  - Surveys divided into paragraphs to retain raw text.
  - Processed using Provalis Prosuit® software with text-based rules to extract relevant sentences.
  - Selected sentences matched to origin paragraphs in Stata and exported to country-specific Excel files for manual verification.

- Step 2. Manual Coding:
  - All software-preselected sentences were examined to determine whether they described actual tax policy changes.
  - Sentences describing tax changes were classified along the variables and dimensions described above.

*Source: IMF working paper appendix text (TPRD data sources and variable definitions).*

### Appendix A (e.g., type of tax, type of change, direction of change, etc.). In this round, each

### wp18110 - Appendix A (e.g., type of tax, type of change, direction of change, etc.). In this round, each

### Steps in building and validating the TPRD
- Data processing pipeline:
  - Sentences with multiple measures were separated so that each line in the database referred to a unique measure; the manually coded information was processed in Stata and exported to Excel files.
  - The merge_MATCH variable was used to indicate whether the IBFD archives contained information for the tax measure.
- Step 3. Dating Tax Measures:
  - Information from IBFD archives and OECD calendars was used to determine the exact announcement and implementation dates for each measure.
  - News clips from the IBFD were compiled into country-specific files and matched to measures identified in Step 2.
  - For measures outside the span of the IBFD (1988 and onwards), the OECD’s Calendar or Chronology of main economic events was used for exact dates.
  - Multiple rounds of checks minimized the number of measures for which exact announcement and implementation dates could not be found.
- Step 4. Cross-checks Against Other Databases:
  - Annual change in tax rate levels from external sources was calculated and compared to instances in the TPRD; because the TPRD does not quantify rate change magnitudes, comparisons focused on direction of change.
  - Four possible outcomes of cross-checks:
    - Match – there was a change in the tax rate according to the external sources, and the TPRD also indicated a rate change in the same direction.
    - Dating problem – the TPRD indicated a rate change in a certain direction, but external databases indicated the rate changed in the same direction during the previous or following year.
    - Coding problem – the TPRD contained a rate change in a certain year, but external databases indicated the rate changed in the opposite direction. If no change in the same year but an opposite change in the previous or following year, this was also marked as a coding problem. Coding problem also indicated when external sources contained a rate change in a certain year but the TPRD did not have any rate change in that year, the previous year, or the following year.
    - Database only – the TPRD indicated a rate change, but external databases did not indicate any rate change in that year, the previous, or the following year. This category was not seen as problematic because external sources contained information on top statutory CIT, PIT, and VAT rates, whereas the TPRD covered many different types of rates (such as reduced and marginal rates).
  - Measures marked as dating or coding problems were checked against IBFD news clips for errors.
- Step 5. Add Information on Intended Revenue Yields during Consolidation Episodes:
  - Information from Dabla-Norris and Lima (forthcoming) on intended revenue yield for tax measures announced during a consolidation year (as defined in Alesina and others (2017)) was merged when possible.
  - Pairing used unique identifiers for validated measures in the TPRD and unique identifiers in the consolidation database.
  - Pairing often associated one TPRD measure with several consolidation-database measures; intended revenue yield for such TPRD measures was given by the sum of intended revenue yields of all associated consolidation-database measures.

### Text extraction and the tax vocabulary
- Tooling:
  - Extraction of textual fragments used Provalis Prosuit® software.
  - Software allowed defining sets of keywords and text-based rules (categories) to analyze and classify document content.
- Tax vocabulary framework:
  - Models how different tax measures are discussed in OECD and IBFD documents: terms typically used to identify a specific tax and how those terms relate to terms identifying actions.
  - Intended to identify passages likely to discuss changes in six taxes of interest: PIT, CIT, VAT, SSC, EXE, PRO.
  - Two keyword sets:
    - Set identifying taxes (examples): “corporate tax”, “value added tax”, “income tax”, “social security contributions”, “excise”, “property tax”.
    - Set identifying actions (examples): announcement, postponement, implementation, abolishment, adoption, change.
  - Interaction of keyword sets through text-based rules operating at the single-sentence level in each OECD and IBFD document.
- Proximity and rule thresholds:
  - Concept of proximity expressed as maximum distance between words belonging to different keyword sets.
  - Maximum distance allowed between keywords was set to 5 words.
  - Rationale: documents are technical and concise; a 5-word limit balanced minimizing missed measures and minimizing selection of irrelevant information.
- Limitations noted:
  - Effectiveness depends on adequacy of keywords, precision of rules, and quality of documents processed.

### Example rules to identify adoption of CIT measures (structure and keyword sets)
- Overall rule logic:
  - ACTION 1: FIND ALL SENTENCE IN PDF FILES THAT SATISFY RULE 1 AND RULE 2
- RULE 1:
  - (AT LEAST ONE WORD OF SET 1 IS NEAR TO AT LEAST ONE WORD OF SET 2) OR (INCLUDE THE WORD "CIT").
  - SET 1 examples: BANK*, BUSINES*, CAPITAL?, CIT, COMPAN*, CORPORATE*, ENTERPRISE*, GAIN*, PROFIT*, INVESTMENT*
  - SET 2 examples: CONTRIBUTION*, DEDUCTION*, DUTIES, DUTY, EXEMPTION*, LEVIES, LEVY, NONTAX*, RATE?, RELIEF, REVENUE*, SURCHARGE*, SURTAX*, TARIFF*, TAX*
- RULE 2:
  - (AT LEAST ONE KEY WORD IN SENTENCES THAT SATISFY RULE 1 IS NEAR TO AT LEAST ONE WORD OF SET 3).
  - KEY WORDS IN SENTENCES THAT SATISFY RULE 1 are then checked for proximity to SET 3.
  - SET 3 examples: ADOPT*, APPROV*, COM*_INTO, DECLAR*, GRANT*, IMPLEMENT*, C?ME_INTO, ENACT*, INTRODUC*, LEGISLAT*, PASS*, RATIF*, TAK*_INTO, SIGN*, TOOK_INTO
- Legend for wildcards:
  - *, ?, # and wildcards stand respectively for any character(s), any letter, and any number. The character "_" is equivalent to space.

### Quality checks, redundancies, and information gaps
- Redundancy checks:
  - Text mining selects any passage likely to discuss changes in the six taxes, including repetitions; identified duplicates were examined and typically only one excerpt retained (preferably the one with more detail).
- Metadata accuracy and consistency checks:
  - Confirmed dimensions of each tax policy change (e.g., type of tax, type of change, direction of change, importance of the change) against OECD calendars and IBFD news archives.
  - For tax packages, additional checks verified consistency of timing (announcement and implementation dates) across tax types.
- Identifying information gaps and their sources:
  - Information gaps may arise from:
    - Automatic text-extraction limitations, especially for old scanned PDF files versus digitally produced PDFs.
    - Source coverage: OECD Economic Surveys were biannual in many countries, which may omit less macro-critical tax measures relative to annual Surveys.
    - Human error in validating extracted excerpts as valid tax measures.
- Cross-checks versus external rate databases:
  - Accuracy of PIT, CIT, and VAT rate-direction information was verified by comparing TPRD direction of rate changes with magnitude of annual rate changes available from external databases.
  - Strategy for inconsistencies:
    - Misrepresentations of measures in OECD Economic Surveys were corrected.
    - Inconsistencies due to coding conventions or broader coverage of the TPRD relative to external databases were documented and left unchanged.
      - Example: rate changes phased over several years are recorded only once in the TPRD when the first tax change occurs (no entry for planned subsequent years) unless OECD Surveys explicitly discuss subsequent implementations.
    - Inconsistencies originating from information gaps (rate/base changes included in external databases but not in TPRD) were documented and subjected to further analysis using OECD and IBFD documents and alternative sources, including national sources; measures with sufficient retrieved information will be added to the next TPRD version.
- Base-change checks:
  - Verification of PIT, CIT, and VAT base changes was possible only for base measures dated using IBFD archives due to lack of cross-country quantitative information on tax base changes.
  - Identified inconsistencies were reviewed to determine misrepresentation versus information gap and addressed following the same logic as for rate changes.
- Additional country-level cross-checks:
  - For the United States, the United Kingdom, Spain, and Portugal, checks were performed against narrative databases to verify major tax measures were included in the TPRD and to document major information gaps for further retrieval and addition to the next version.

*Source: wp18110 - Appendix A (content unit provided).*

### Appendix D.

### Appendix D.

### Tables and Figures: Inventory
- Page references and figure/table labels included in the appendix: Figures Appendix E.2 and E.3; Tables Appendix E.1 through E.7.
- Sources cited across figures and tables: IMF, OECD, IBFD, national sources; Romer and Romer (2009); Cloyne (2013); Gil and others (2017); Pereira and Wemans (2013); Tax Policy Reform Database, OECD, IBFD.

### Identified Information Gaps from OECD Surveys (Figure Appendix E.2)
- Note: Solid bars indicate that there is at least one tax policy measure in the external rate databases that is not included in the current version of the TPRD for a specific year.

### Identified Information Gaps Based on Narrative Databases (Figure Appendix E.3)
- Sources: Romer and Romer (2009), Cloyne (2013), Gil and others (2017), and Pereira and Wemans (2013).
- Note: Solid bars indicate that there is at least one tax policy measure in the narrative databases that is not included in the current version of the TPRD for a specific year.
- Color coding in figure: dark blue = CIT changes; blue = PIT changes; light blue = VAT changes; light green = EXE changes; green = SSC changes; dark green = PRO changes.

### Distribution of Tax Policy Measures by Type and Direction of Change (Table Appendix E.1)
- Source: Tax Policy Reform Database, OECD, IBFD.
- Country-year counts and averages (selected excerpts as presented in table):
  - Country year counts (first block): 32, 35, 23, 23, 22, 20, 19, 19, 15, 26, 42, 27, 24, 20, 35, 37, 32, 12, 18, 35, 11, 15, 33, 18, 18, 11, 15, 14, 13, 16, 12, 6.0, 14, 24, 15, 14, 13, 20, 24, 21, 9.0, 10, 16, 5.0, 8.0, 23, 26, 30, 17, 20, 17, 14, 11, 12, 15, 19, 40, 21, 21, 13, 29, 35, 23, 6.0, 15, 31, 10, 11, 25, 25, 32, 19, 19, 12, 19, 18, 19, 21, 26, 36, 19, 20, 19, 33, 33, 33, 17, 18, 28, 8.0, 14, 32, 19, 22, 14, 10, 9.0, 10, 9.0, 11, 9.0, 15, 30, 10, 11, 13, 25, 26, 26, 11, 4.0, 17, 6.0, 7.0, 17, 16, 29, 13, 18, 8.0, 12, 13, 13, 18, 17, 20, 13, 17, 11, 23, 24, 19, 10, 17, 18, 4.0, 7.0, 23.
  - Aggregate counts (selected blocks): 136, 137, 53, 89, 61, 58, 46, 54, 39, 53, 174, 67, 73, 50, 118, 171, 96, 35, 43, 123, 36, 52, 109, 36, 35, 23, 29, 23, 29, 30, 25, 7.0, 21, 68, 29, 25, 22, 47, 47, 52, 15, 14, 36, 9.0, 24, 45, 100, 102, 30, 60, 38, 29, 16, 29, 32, 32, 106, 38, 48, 28, 71, 124, 44, 20, 29, 87, 27, 28, 64, 77, 101, 68, 47, 37, 42, 32, 43, 45, 56, 109, 53, 40, 46, 128, 132, 90, 30, 40, 83, 16, 24, 73, 39, 49, 49, 14, 15, 20, 11, 17, 16, 24, 70, 16, 15, 27, 65, 61, 49, 15, 6.0, 45, 8.0, 10, 27, 38, 52, 19, 33, 22, 22, 21, 26, 29, 32, 39, 37, 25, 19, 63, 71, 41, 15, 34, 38, 8.0, 14, 46, 6.1, 6.1, 4.8, 5.0, 4.1, 4.2, 3.5, 4.0, 3.5, 3.1, 6.3, 4.1, 4.5, 4.0, 6.3, 7.6, 4.9, 3.4, 3.5, 5.3, 4.3, 4.2, 4.4.
  - Overall totals (end of block): 213, 238, 121, 136, 98, 100, 78, 97, 84, 109, 283, 120, 113, 96, 246, 303, 186, 65, 83, 206, 52, 76, 182, 35, 39, 25, 27, 24, 24, 22, 24, 24, 35, 45, 29, 25, 24, 39, 40, 38, 19, 24, 39, 12, 18, 41.

### Frequency of Tax Policy Measures (Table Appendix E.2)
- Source: Tax Policy Reform Database, OECD, IBFD.
- Grand totals and category totals (as appearing in the table):
  - Grand Total values: 1,658; 42; 79; 78; 187; 741; 531; 143; 10; 17; 7; 10; 48; 51; 61; 4; 10; 6; 6; 15; 20; 82; 6; 7; 1; 4; 33; 31; 180; 7; 33; 53; 22; 35; 30; 103; 2; 27; 44; 11; 9; 10; 77; 5; 6; 9; 11; 26; 20; 323; 17; 50; 60; 32; 83; 81; 1,981; 59; 129; 138; 219; 824; 612; 1,435; 537; 898; 900; 433; 467; 2,335; 227; 871; 403; 141; 831; 315; 412; 876; 32; 4; 6; 1; 3; 7; 11; 4; 1; 1; 1; 1; 28; 3; 5; 1; 2; 7; 10; 46; 1; 9; 2; 8; 8; 18; 19; 1; 8; 1; 4; 3; 2; 27; 1; 1; 4; 5; 16; 78; 5; 15; 3; 11; 15; 29; 8; 2; 2; 2; 2; 1; 1; 7; 2; 2; 2; 1; 9; 6; 2; 1; 6; 4; 1; 1; 3; 2; 1; 17; 2; 8; 2; 3; 2; 95; 7; 23; 5; 11; 18; 31; 148; 1; 3; 8; 73; 63; 56; 3; 4; 31; 18; 92; 1; 4; 42; 45; 131; 1; 5; 2; 10; 51; 62; 21; 5; 2; 7; 2; 5; 110; 1; 3; 49; 57; 279; 2; 8; 2; 18; 124; 125; 23; 4; 1; 8; 10; 6; 2; 1; 2; 1; 17; 2; 6; 9; 12; 1; 3; 3; 5; 6; 1; 2; 3; 6; 1; 5; 35; 5; 3; 4; 13; 10; 314; 7; 11; 6; 18; 137; 135; 180; 601; 201; 774; 0; 137; 357; 317; 241; 129; 524; 09.
  - Final aggregate values at end of table: 1,615; 597; 1,018; 1,077; 473; 604; 2,692; 258; 941; 643; 351; 951; 405; 933; 3,285.

### Correlation and Co-occurrence of Tax Policy Measures (Tables Appendix E.3 and E.4)
- Source: Tax Policy Reform Database, OECD, IBFD.
- Table Appendix E.3: Correlation among different tax policy measures (table provided in appendix; numeric values reported in original table).
- Table Appendix E.4: Co-occurrence of different tax policy measures in the sample (reported in country years; numeric values reported in original table).

### Sample Coverage during Economic Cycles and Election Cycles (Table Appendix E.5)
- Source: Tax Policy Reform Database, OECD, IBFD.
- Expansion/Recession and Consolidation/Normal times coverage — minimum and maximum years by country (selected entries as given):
  - Minimum Year (Expansion/Recession block): Australia 2003; Austria 2002; Brazil 2014; Canada 2006; China 2011; Czech Republic 2007; Denmark 2012; France 2007; Germany 2007; Greece 1996; India 2007; Ireland 2007; Italy 2011; Japan 2006; Luxembourg 2007; Mexico 2007; Poland 2007; Portugal 2011; South Korea 2008; Spain 2014; Turkey 2001; United Kingdom 2008; United States 1970.
  - Maximum Year (Expansion/Recession block): Australia 1970; Austria 1970; Brazil 1985; Canada 1988; China 1970; Czech Republic 1989; Denmark 1988; France 1987; Germany 1970; Greece 1970; India 1970; Ireland 1970; Italy 1970; Japan 1981; Luxembourg 1988; Mexico 1990; Poland 1970; Portugal 1970; South Korea 1970; Spain 1991; Turkey 1971; United Kingdom 1988; United States 1973.
  - Consolidation/Normal times coverage — minimum and maximum years by country (selected entries): Australia minimum 2011, maximum 2013; Austria minimum 2013, maximum 2014; Canada minimum 2014, maximum 2013; Denmark minimum 2013, maximum 2013; France minimum 2006, maximum 2006; Germany minimum 2014, maximum 2014; Ireland minimum 2011, maximum 2011; Italy minimum 2011, maximum 2011; Japan minimum 2014, maximum 2014; Portugal minimum 2012, maximum 2013; Spain minimum 2013, maximum 2012; United Kingdom minimum 1997, maximum 1978; United States minimum 1979, maximum 1989. 
  - Pre-/Post-election coverage — minimum and maximum years by country (selected entries): Australia minimum 2001, maximum 2010; Austria minimum 2012, maximum 2011; Canada minimum 2012, maximum 2011; Czech Republic minimum 2009, maximum 2014; Denmark minimum 2009, maximum 1969; France minimum 1974, maximum 1989; Germany minimum 1986, maximum 1991; Greece minimum 1973, maximum 1990; Ireland minimum 1969, maximum 1988; Italy minimum 1988, maximum 1988; Japan minimum 1973, maximum 1973; Luxembourg minimum 1972, maximum 1992; Poland minimum 1972, maximum 1972; Portugal minimum 1986, maximum 1972.

### Composition of Major Tax Reforms by Tax Type, Type of Change, and Country (Table Appendix E.6)
- Source: Tax Policy Reform Database, OECD, IBFD.
- Composition percentages by tax type (BASE vs RATE) for selected countries (as presented):
  - Australia: CIT BASE 36% / RATE 64% ; EXE BASE 70% / RATE 30% ; PIT BASE 100% .
  - Austria: CIT BASE 41% / RATE 59% ; EXE BASE 75% / RATE 25% ; PIT BASE 18% / RATE 82% .
  - Brazil: CIT BASE 57% / RATE 43% ; EXE BASE 100% .
  - Canada: CIT BASE 21% / RATE 79% ; EXE BASE 100% .
  - China: CIT BASE 35% / RATE 65% .
  - Czech Republic: CIT BASE 22% / RATE 78% ; EXE BASE 50% / RATE 50% .
  - Denmark: CIT BASE 22% / RATE 78% ; EXE BASE 50% / RATE 50% .
  - France: CIT BASE 20% / RATE 80% ; EXE BASE 42% / RATE 58% .
  - Germany: CIT BASE 73% / RATE 27% ; EXE BASE 26% / RATE 74% .
  - Greece: CIT BASE 69% / RATE 31% ; EXE BASE 40% / RATE 60% .
  - India: CIT BASE 42% / RATE 58% ; EXE BASE 60% / RATE 40% .
  - Ireland: CIT BASE 60% / RATE 40% ; EXE BASE 60% / RATE 40% .
  - Italy: CIT BASE 75% / RATE 25% ; EXE BASE 47% / RATE 53% .
  - Japan: CIT BASE 63% / RATE 38% ; EXE BASE 78% / RATE 22% .
  - Luxembourg: CIT BASE 22% / RATE 78% ; EXE BASE 63% / RATE 38% .
  - Mexico: CIT BASE 100% / RATE 25% ; EXE BASE 75% .
  - Poland: CIT BASE 100% / RATE 42% ; EXE BASE 58% .
  - Portugal: CIT BASE 46% / RATE 54% ; EXE BASE 56% / RATE 44% .
  - South Korea: CIT BASE 67% / RATE 33% ; EXE BASE 33% / RATE 67% .
  - Spain: CIT BASE 30% / RATE 70% ; EXE BASE 80% / RATE 20% .
  - Turkey: CIT BASE 38% / RATE 63% ; EXE BASE 24% / RATE 76% .
  - United Kingdom: CIT BASE 78% / RATE 22% ; EXE BASE 21% / RATE 79% .
  - United States: CIT BASE 67% / RATE 33% ; EXE BASE 100% .
- Note: Table lists composition for tax types CIT, EXE, PIT, PRO, SSC, VAT for each country, with BASE and RATE percentages as shown.

### Composition of Major Tax Reforms by Tax Type, Direction of Change, and Country (Table Appendix E.7)
- Source: Tax Policy Reform Database, OECD, IBFD.
- Directional composition (DEC vs INC) by tax type for selected countries (as presented):
  - Australia: CIT DEC 45% / INC 55% .
  - Austria: CIT DEC 60% / INC 40% ; EXE DEC 100% .
  - Brazil: CIT DEC 28% / INC 72% ; EXE DEC 25% / INC 75% .
  - Canada: CIT DEC 38% / INC 62% .
  - China: CIT DEC 57% / INC 43% .
  - Czech Republic: CIT DEC 50% / INC 50% ; EXE DEC 32% / INC 68% .
  - Denmark: CIT DEC 100% / INC 52% .
  - France: CIT DEC 48% / INC 22% / other entries as shown in table.
  - Germany: CIT DEC 50% / INC 50% ; EXE DEC 22% / INC 78% .
  - Greece: CIT DEC 100% / INC 13% .
  - India: CIT DEC 37% / INC 63% .
  - Ireland: CIT DEC 73% / INC 27% .
  - Italy: CIT DEC 24% / INC 76% .
  - Japan: CIT DEC 94% / INC 6% .
  - Luxembourg: CIT DEC 38% / INC 63% .
  - Mexico: CIT DEC 29% / INC 71% .
  - Poland: CIT DEC 29% / INC 71% .
  - Portugal: CIT DEC 20% / INC 80% .
  - South Korea: CIT DEC 20% / INC 80% .
  - Spain: CIT DEC 75% / INC 25% .
  - Turkey: CIT DEC 18% / INC 82% .
  - United Kingdom: CIT DEC 67% / INC 33% .
  - United States: CIT DEC 50% / INC 50% .
- Note: Table provides DEC/INC percentage splits for tax categories CIT, EXE, PIT, PRO, SSC, VAT by country as presented.

*Source: wp18110 - Appendix D. (Tax Policy Reform Database, OECD, IBFD; figures and tables as listed in Appendix D.)*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18110.pdf_