## wp17145 - REFERENCES

## Source details

**Canonical URL:** [wp17145 - REFERENCES](https://www.imf.org/-/media/files/publications/wp/2017/wp17145.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17145.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17145.pdf.json)

---

### Overview
- Research objective: assess causal relationship from fiscal policy to growth using country-case studies and the Synthetic Control Method (SCM).
- Sample focus: high- and upper-middle-income countries; reform episodes during 1975-2010 (or latest available observation).
- Number of indicators used for country-case selection: nine.

### Methodology
- Identification method: Synthetic Control Method (SCM) developed in Abadie and Gardeazabal (2003), and Abadie, Diamond, and Hainmueller (2010).
- Selection approach improvements versus IMF (2015a):
  - Use only quantitative criteria (no qualitative elements) to reduce discretion and selection bias.
  - Restrict candidate pool to high- and upper-middle-income countries to improve homogeneity of synthetic units and data consistency.
  - Explicitly account for non-fiscal structural reforms (business regulation, financial market, labor market, legal and product markets).
  - Include income inequality as an additional outcome dimension.
- Data handling:
  - Five-year non-overlapping averages starting in 1975 to 2010 or latest available observation to smooth business-cycle fluctuations.
  - Reform durability requirement: change in relevant fiscal variable must occur over at least two consecutive five-year periods.
  - Countries reporting less than 20 years of data for a relevant fiscal indicator were excluded (VAT standard rate threshold lowered to 15 years due to data limitations).

### Country selection criteria and indicators
- Three fiscal reform areas covered and the nine quantitative indicators used for selection:
  - Tax mix (direct-to-indirect tax ratio): Negative change over at least two consecutive five-year periods (source: OECD Tax Revenue Statistics and WEO).
  - VAT standard rate (percent): Positive change over at least two consecutive five-year periods (source: Tax Policy Division of the IMF's Fiscal Affairs Department).
  - Top corporate income tax rate (percent): Negative change over at least two consecutive five-year periods.
  - Top individual income tax rate (percent): Negative change over at least two consecutive five-year periods.
  - Public capital spending (percent of GDP): Positive change over at least two consecutive five-year periods (source: WEO).
  - Public health spending (percent of GDP): Positive change over at least two consecutive five-year periods (source: GFS and WEO).
  - Public education spending (percent of GDP): Positive change over at least two consecutive five-year periods.
  - Public social protection spending (percent of GDP): Positive change over at least two consecutive five-year periods.
  - Overall fiscal balance (percent of GDP): Positive change over at least two consecutive five-year periods (source: WEO).
- Selection procedure:
  - Countries ranked by number of reform indicators satisfied (maximum of nine).
  - Within-region selection following IMF regional classification across EUR, APD, WHD to choose countries with the largest number of reforms per region.
  - Only countries with at least five reform episodes included in initial OECD universe for Table 2.

### Selected reform countries and comparator selection
- Seven countries selected as reformers: Australia, New Zealand, Belgium, Ireland, Netherlands, Germany, and Chile.
- Note: Some countries experienced more than one reform event (examples: Australia and Chile).
- Comparator pool:
  - Control group constructed from high- and upper-middle-income countries (World Bank classification), excluding oil exporters and small states.
  - Only countries with five reforms or less included in control group.
  - Comparator country types:
    - Type A: all qualifying comparator countries in the whole pool.
    - Type B: subset in the same region as the treated country.
  - Baseline chosen by lower root mean squared error (RMSE) of pre-event real GDP match:
    - Type A chosen for Australia, Germany, and New Zealand.
    - Type B chosen for Belgium, Chile, Ireland, and the Netherlands.

### Main empirical findings
- Cases identified: nine fiscal reform episodes in seven countries.
- Growth effects (average annual GDP growth rate 10 years after reform start, treated minus synthetic):
  - Average effect across episodes: about 1 percentage point.
  - Range across episodes: 0.1 percentage points (Germany, reform started in 2003) to 4.3 percentage points (Chile, reform started in 1983).
  - Excluding Chile 1983 and Ireland 1987 reduces the average growth impact to 0.3 percentage point.
- Timing and heterogeneity:
  - Speed of materialization varied: quick in Chile and Ireland; more modest in first few years in other cases.
  - Reform episodes in countries with relatively lower initial level of development were associated with higher growth after the fiscal reform started (examples: Chile 1983 and Ireland 1987).
- Placebo experiments and robustness:
  - Placebo experiments confirm baseline results for four out of nine cases.
  - Results are borderline robust in three cases.
  - Not robust in two cases.
  - Two-thirds of events are either robust or borderline robust to placebo assessment.
- Non-fiscal structural reforms:
  - Main results hold after controlling for pre-reform non-fiscal structural reforms.
  - No definitive evidence that non-fiscal structural reforms were stronger in reformer countries relative to their synthetic units after fiscal reforms started.
  - Caveat: many structural reforms could take place after fiscal reforms started; attribution remains cautious.
- Inequality outcomes:
  - No clear-cut evidence that fiscal reform countries had different inequality outcomes relative to a global trend or their synthetic units.

### Synthetic Control Method: description and limitations
- Purpose:
  - Provides a data-driven methodology to quantify the effects of a particular event in comparative case studies by creating an artificial counterfactual (synthetic unit) matched to the unit of interest prior to the event.
- Interpretation:
  - The divergence between the treated unit and its synthetic unit after the event is interpreted as the quantitative estimate of the event’s effect over a specified number of periods.
- Advantages:
  - Captures heterogeneity in effects by producing case-specific estimates (contrast with average estimates from panel regressions).
  - Addresses omitted-variable bias from time-variant fixed effects.
- Limitations and potential biases:
  - Can suffer from reverse causation if the decision to undertake fiscal reforms was affected by expectations on future growth.
  - Estimates can be biased by subsequent events that affect the treated unit and its counterfactual differently (example: radical political revolutions after natural disasters).
  - Standard inference techniques are not directly applicable; placebo experiments are used for robustness.

### Synthetic Control Method: implementation steps
- Step 1: Choose comparator countries and explanatory variables
  - Comparator countries: similar to the country of interest but unaffected by the event within the sample period.
  - Predictor variables in initial country selection: level of GDP per capita (at the beginning, half-way, and the end of the pre-reform period), trade openness, inflation rate, terms of trade, index of human capital.
  - Outcome variable used: real GDP at constant 2005 prices (in million 2005 US$). (Note: alternative specification using real GDP per capita yields similar main results, but growth effect is negative in two cases: Chile 1974 and New Zealand 1986.)
- Step 2: Construct the synthetic series via nested optimization
  - Country weights vector w* chosen to minimize distance between pre-event predictors for unit of interest and comparators, subject to weights between zero and one.
  - Variable weight matrix V* chosen among positive definite diagonal matrices to minimize mean square prediction error (MSPE) of the outcome variable over the pre-event period.
  - Iterative optimization updates V* and w* to construct the synthetic post-event path.
- Step 3: Compare actual and post-event outcome variables
  - The difference between actual and synthetic post-event outcome series is the estimated impact (assuming other factors are controlled).

### Baseline results: growth effects of fiscal structural reforms
- Aggregate finding:
  - Fiscal reforms led to higher growth relative to the synthetic unit in all reform episodes, with heterogeneity across episodes.
- Key statistics (10-year horizon):
  - Average annual GDP growth for the 10 years after the fiscal reform started was higher in the reform country relative to its synthetic units by about 1 percentage point.
  - Range across episodes: from 0.1 percentage point (Germany 2003) to 4.3 percentage points (Chile 1983).
  - Excluding Chile 1983 and Ireland 1987 reduces the average growth impact to 0.3 percentage point.
- Placebo experiments (robustness specifics at 10 years):
  - Very few placebo permutations exceed treated-country growth effects:
    - Belgium 1992: 1/4
    - Chile 1983: 0/8
    - Ireland 1987: 0/3
    - Netherlands 1983: 0/6
  - Baseline estimates not robust for:
    - Chile 1974: 4/7
    - Germany 2003: 20/21
  - Borderline cases:
    - Australia 1985: 4/20
    - Australia 1998: 4/18
    - New Zealand 1986: 5/19
  - Overall: two-thirds of events are either robust or borderline robust to placebo assessment.

### Non-fiscal structural reforms: methods to control and main findings
- Rationale:
  - Structural reforms in product market, labor market, business regulations, legal environment, financial reforms can affect long-term growth and potentially bias estimates.
- Approaches:
  - Pre-reform differences: include non-fiscal structural reform indices one at a time as additional predictors in SCM.
  - Concurrent/post-reform differences: document ten-year changes in structural reform indices and compare treated vs synthetic; construct “non-fiscal structural reform gaps” in placebo permutations and assess p-values.
- Controlled structural reform indicators:
  - Business regulations, financial reforms, legal reforms, labor market reforms, product market reforms (added one at a time).
- Results controlling for pre-reform non-fiscal structural reforms:
  - Baseline estimates generally confirmed.
  - Weighted-average finding: on average, output growth was higher by 0.9 percentage points relative to control units, ranging from -0.7 percentage point (New Zealand 1983) to 4.3 percentage points (Chile 1983).
  - Some point estimates for Australia 1985, Australia 1998, and New Zealand 1986 show negative growth effects after including structural reform predictors.
  - Exception: New Zealand 1986 growth effect turns negative when controlling for non-fiscal structural reforms (possible omitted variable bias in baseline).
- Concurrent non-fiscal reforms assessment:
  - Two exercises:
    1. Ten-year changes in non-fiscal structural reform indicators: compare whether treated or synthetic had more reforms.
    2. Non-fiscal structural reform gaps in placebo permutations: assess p-values to gauge whether treated unit had unusually strong concurrent structural reform progress.
  - General summary:
    - Non-fiscal reforms were more pronounced in treated units but not predominant.
    - p-values for non-fiscal structural reform gaps are often large, suggesting a subdued role of non-fiscal reforms; p-values were larger than those for growth effects in almost all cases, highlighting fiscal reforms as likely key contributors.
  - Caveat:
    - Magnitude of growth effects caused by non-fiscal structural reforms varies across countries; quantitative assessment remains an open question.

### Non-fiscal structural reform indicators: summary statistics (Table 3)
- Table 3: Average Percentage Change Difference in Index (Difference = treated unit minus synthetic unit, ten years after start of fiscal reform)
  - Australia 1985:
    - Business Regulation: 10.2
    - Financial: 34.4
    - Labor Market: n.a.
    - Legal: 1.9
    - Product Market: 1.3
  - Australia 1998:
    - Business Regulation: 21.1
    - Financial: 0.0
    - Labor Market: -17.6
    - Legal: -7.0
    - Product Market: 2.2
  - Belgium 1992:
    - Business Regulation: 3.1
    - Financial: -40.3
    - Labor Market: n.a.
    - Legal: -23.8
    - Product Market: -6.1
  - Chile 1974:
    - Business Regulation: n.a.
    - Financial: 46.7
    - Labor Market: n.a.
    - Legal: -59.2
    - Product Market: n.a.
  - Chile 1983:
    - Business Regulation: 14.7
    - Financial: 32.4
    - Labor Market: n.a.
    - Legal: 2.8
    - Product Market: n.a.
  - Germany 2003:
    - Business Regulation: 11.0
    - Financial: n.a.
    - Labor Market: 13.1
    - Legal: -3.3
    - Product Market: -4.4
  - Ireland 1987:
    - Business Regulation: -18.5
    - Financial: 18.0
    - Labor Market: n.a.
    - Legal: 5.0
    - Product Market: -4.3
  - Netherlands 1983:
    - Business Regulation: 1.7
    - Financial: 5.0
    - Labor Market: n.a.
    - Legal: 1.2
    - Product Market: -18.7
  - New Zealand 1986:
    - Business Regulation: 9.6
    - Financial: -24.5
    - Labor Market: n.a.
    - Legal: 7.9
    - Product Market: -25.2
- “More Reform in Treated/Total” counts:
  - Business Regulation: 7/8
  - Financial: 6/8
  - Labor Market: 1/2
  - Legal: 5/9
  - Product Market: 5/7

### Non-fiscal structural reform gaps vs growth effect (Table 4)
- For six cases where baseline growth effect was robust or borderline robust, comparison of growth placebo permutations and non-fiscal reform gaps:
  - Belgium 1992:
    - Growth effect placebos: 1/4 better than treated
    - Business Regulations: 2/4 better than treated
    - Legal System in Property Rights: 3/4 better than treated
  - Chile 1983:
    - Growth effect placebos: 0/8 better than treated
    - Business Regulations: 2/8 better than treated
    - Legal System in Property Rights: 4/8 better than treated
  - Ireland 1987:
    - Growth effect placebos: 0/3 better than treated
    - Business Regulations: 2/3 better than treated
    - Legal System in Property Rights: 1/3 better than treated
  - Netherlands 1983:
    - Growth effect placebos: 0/6 better than treated
    - Business Regulations: 1/4 better than treated
    - Legal System in Property Rights: 3/6 better than treated
  - Australia 1985:
    - Growth effect placebos: 4/20 better than treated
    - Business Regulations: 7/20 better than treated
    - Legal System in Property Rights: 8/20 better than treated
  - Australia 1998:
    - Growth effect placebos: 4/18 better than treated
    - Business Regulations: 1/18 better than treated
    - Legal System in Property Rights: 17/18 better than treated

### Inequality: methods and key findings
- Data and measures:
  - Use Solt (2014)’s Standardized World Income Inequality Database (SWIID).
  - Track market income Gini and net income Gini (net after fiscal redistribution).
- Scope:
  - Seven cases with available data (Chile 1974 and Netherlands 1983 dropped).
  - Use baseline SCM weights to construct synthetic unit Gini series (documentary, not causal).
- Findings relative to global trend (ten-year averages after reform started):
  - Both market income Gini and net income Gini increased in five out of seven cases (exceptions: Chile 1983 and Ireland 1987).
  - Net income Gini increased less or decreased more than market income Gini in most cases (exceptions: Belgium 1992 and Ireland 1987), suggesting fiscal redistribution partially offset market-income inequality increases.
  - For Germany 2003 nine-year averages were used as the Gini series ends in 2012.
- Findings relative to synthetic units:
  - Market income Gini increased more in four out of seven reform cases relative to synthetic units: Australia 1985, Australia 1998, Germany 2003, New Zealand 1986.
  - Fiscal redistribution increased more in those four cases as well.
  - Cases where market income Gini increased less than synthetic unit and fiscal redistribution also increased less: Belgium 1992 and Ireland 1987.
  - Chile 1983: market Gini increased less than synthetic unit while fiscal redistribution increased more.
  - Overall: no consistent pattern that fiscal reforms led to higher market-income inequality or a systematic weakening of fiscal redistribution in reforming countries.
- Caveats:
  - Results are associations, not causal claims.
  - Data limitations prevented re-estimating SCM using both growth and inequality as outcome variables.

### Concluding remarks and policy-relevant implications
- Summary of evidence:
  - Countries identified as exposed to growth-friendly fiscal reforms experienced higher growth relative to their synthetic units, with larger impacts for countries at lower initial development levels.
  - Results broadly robust to placebo permutations and controlling for pre-reform non-fiscal structural reforms.
  - No definitive evidence that non-fiscal structural reforms were systematically stronger in reform countries relative to their synthetic units after reforms started.
  - No clear-cut evidence of a systematic trade-off between growth-friendly fiscal reforms and increases in inequality at the aggregate level; no evidence pointing to causality from structural fiscal reforms to more unequal income distribution.
- Policy-relevant implications and suggested avenues for further research:
  - A well-designed, durable package of fiscal structural reforms can be associated with positive long-term growth effects that persist after accounting for pre-existing non-fiscal reforms.
  - Quantitative, transparent selection criteria and restricting analysis to more homogeneous income groups can improve credibility of SCM-based country-case inference.
  - Robustness checks (placebo experiments) are essential: results vary across episodes and warrant case-by-case policy caution.
  - Monitoring distributional outcomes remains important since no systematic inequality effects were detected across the sampled reform episodes.
  - Further research:
    - Identify whether broadening tax bases (rather than changes in standard rates) yields more desirable growth benefits.
    - Assess the impact of structural expenditure measures controlling for quality across spending categories.
    - Explore these questions using the synthetic control method where data availability permits.

### Appendix 1. Country Selection — Quantitative Reform Indicators (selected highlights)
- European countries (EUR) — Selected countries marked with * and counts:
  - Netherlands* — High Income: OECD — 8
  - Ireland* — High Income: OECD — 8
  - Germany* — High Income: OECD — 8
  - Belgium* — High Income: OECD — 8
  - Sweden — High Income: OECD — 7
  - United Kingdom — High Income: OECD — 7
  - Denmark — High Income: OECD — 7
  - Finland — High Income: OECD — 7
  - Hungary — Upper Middle Income: OECD — 7
  - Portugal — High Income: OECD — 7
  - Italy — High Income: OECD — 7
  - Turkey — Upper Middle Income: OECD — 7
  - Spain — High Income: OECD — 7
  - Luxembourg — High Income: OECD — 7
  - Austria — High Income: OECD — 6
  - France — High Income: OECD — 5
  - Poland — High Income: OECD — 4
  - Switzerland — High Income: OECD — 4
  - Czech Republic — High Income: OECD — 4
  - Bulgaria — Upper Middle Income: OECD — 4
  - Slovak Republic — High Income: OECD — 2
  - Estonia — High Income: OECD — 2
  - Lithuania — High Income — 1
  - Malta — High Income — 1
- Asian-Pacific countries (APD) — Selected:
  - Australia* — High Income: OECD — 6
  - New Zealand* — High Income: OECD — 6
  - Korea — High Income: OECD — 5
  - Thailand — Upper Middle Income — 5
  - Singapore — High Income — 5
  - Japan — High Income: OECD — 4
  - Malaysia — Upper Middle Income — 3
  - China — Upper Middle Income — 2
  - Hong Kong SAR — High Income — 1
- Western Hemisphere countries (WHD) — Selected:
  - Chile* — High Income: OECD — 6
  - Canada — High Income: OECD — 5
  - Uruguay — High Income — 5
  - Panama — Upper Middle Income — 5
  - Costa Rica — Upper Middle Income — 4
  - United States — High Income: OECD — 4
  - Brazil — Upper Middle Income — 2

### Appendix 2. Comparator Countries — summary
- Type A: Countries with less than 5 reforms. Cases: Australia, Germany, and New Zealand
  - Albania, Argentina, Belarus, Bosnia and Herzegovina, Brazil, Bulgaria, Canada, China, Colombia, Costa Rica, Croatia, Cyprus, Czech Republic, Dominican Republic, Estonia, France, Germany, Hong Kong SAR, Iceland, Jamaica, Japan, Korea, Latvia, Lithuania, Malta, Malaysia, Malta, Panama, Peru, Poland, Serbia, Singapore, Slovak Republic, Switzerland, Thailand, United States, Uruguay
- Type B: Countries with less than 5 reforms and are in the same region. Cases: Chile
  - Argentina, Brazil, Canada, Colombia, Costa Rica, Dominican Republic, Jamaica, Panama, Peru, United States, Uruguay
- Type B: Cases: Belgium, Ireland, and Netherlands
  - Albania, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, Cyprus, Czech Republic, Estonia, France, Germany, Iceland, Latvia, Lithuania, Malta, Poland, Serbia, Slovak Republic, Switzerland

*Source: wp17145 - REFERENCES, IMF working paper content provided.*

### REFERENCES _____________________________________________________________________________________  44

### wp17145 - REFERENCES

### Overview
- Research objective: assess causal relationship from fiscal policy to growth using country-case studies and the Synthetic Control Method (SCM).
- Sample focus: high- and upper-middle-income countries; reform episodes during 1975-2010 (or latest available observation).
- Number of indicators used for country-case selection: nine.

### Methodology
- Identification method: Synthetic Control Method (SCM) developed in Abadie and Gardeazabal (2003), and Abadie, Diamond, and Hainmueller (2010).
- Selection approach improvements versus IMF (2015a):
  - Use only quantitative criteria (no qualitative elements) to reduce discretion and selection bias.
  - Restrict candidate pool to high- and upper-middle-income countries to improve homogeneity of synthetic units and data consistency.
  - Explicitly account for non-fiscal structural reforms (business regulation, financial market, labor market, legal and product markets).
  - Include income inequality as an additional outcome dimension.
- Data handling:
  - Five-year non-overlapping averages starting in 1975 to 2010 or latest available observation to smooth business-cycle fluctuations.
  - Reform durability requirement: change in relevant fiscal variable must occur over at least two consecutive five-year periods.
  - Countries reporting less than 20 years of data for a relevant fiscal indicator were excluded (VAT standard rate threshold lowered to 15 years due to data limitations).

### Country selection criteria and indicators
- Three fiscal reform areas covered:
  - Tax mix (direct-to-indirect tax ratio): Negative change over at least two consecutive five-year periods (source: OECD Tax Revenue Statistics and WEO).
  - VAT standard rate (percent): Positive change over at least two consecutive five-year periods (source: Tax Policy Division of the IMF's Fiscal Affairs Department).
  - Top corporate income tax rate (percent): Negative change over at least two consecutive five-year periods.
  - Top individual income tax rate (percent): Negative change over at least two consecutive five-year periods.
  - Public capital spending (percent of GDP): Positive change over at least two consecutive five-year periods (source: WEO).
  - Public health spending (percent of GDP): Positive change over at least two consecutive five-year periods (source: GFS and WEO).
  - Public education spending (percent of GDP): Positive change over at least two consecutive five-year periods.
  - Public social protection spending (percent of GDP): Positive change over at least two consecutive five-year periods.
  - Overall fiscal balance (percent of GDP): Positive change over at least two consecutive five-year periods (source: WEO).
- Selection procedure:
  - Countries ranked by number of reform indicators satisfied (maximum of nine).
  - Within-region selection following IMF regional classification across EUR, APD, WHD to choose countries with the largest number of reforms per region.
  - Only countries with at least five reform episodes included in initial OECD universe for Table 2.

### Selected reform countries and comparator selection
- Seven countries selected as reformers: Australia, New Zealand, Belgium, Ireland, Netherlands, Germany, and Chile.
- Note: Some countries experienced more than one reform event (examples: Australia and Chile).
- Comparator pool:
  - Control group constructed from high- and upper-middle-income countries (World Bank classification), excluding oil exporters and small states.
  - Only countries with five reforms or less included in control group.
  - Comparator country types:
    - Type A: all qualifying comparator countries in the whole pool.
    - Type B: subset in the same region as the treated country.
  - Baseline chosen by lower root mean squared error (RMSE) of pre-event real GDP match:
    - Type A chosen for Australia, Germany, and New Zealand.
    - Type B chosen for Belgium, Chile, Ireland, and the Netherlands.

### Main empirical findings
- Cases identified: nine fiscal reform episodes in seven countries.
- Growth effects (average annual GDP growth rate 10 years after reform start, treated minus synthetic):
  - Average effect across episodes: about 1 percentage point.
  - Range across episodes: 0.1 percentage points (Germany, reform started in 2003) to 4.3 percentage points (Chile, reform started in 1983).
- Relationship with initial development:
  - Reform episodes in countries with relatively lower initial level of development were associated with higher growth after the fiscal reform started.
- Placebo experiments and robustness:
  - Placebo experiments confirm baseline results for four out of nine cases.
  - Results are borderline robust in three cases.
  - Not robust in two cases.
  - Two-thirds of events are either robust or borderline robust to placebo assessment.
- Non-fiscal structural reforms:
  - Main results hold after controlling for non-fiscal structural reforms that took place prior to fiscal reforms.
  - No definitive evidence found that non-fiscal structural reforms were stronger in reformer countries relative to their synthetic units after fiscal reforms started.
  - Caveat: many structural reforms could take place after fiscal reforms started; attribution remains cautious.

### Inequality outcomes
- Despite theoretical potential tradeoffs between growth and inequality, the study did not observe a clear tradeoff in the analyzed fiscal reform episodes.
- No clear-cut evidence that fiscal reform countries had different inequality outcomes relative to a global trend or their synthetic units.

### Policy-relevant implications
- A well-designed, durable package of fiscal structural reforms can be associated with positive long-term growth effects that persist after accounting for pre-existing non-fiscal reforms.
- Quantitative, transparent selection criteria and restricting analysis to more homogeneous income groups can improve credibility of SCM-based country-case inference.
- Robustness checks (placebo experiments) are essential: results vary across episodes and warrant case-by-case policy caution.
- Monitoring distributional outcomes remains important since no systematic inequality effects were detected across the sampled reform episodes.

*Source: wp17145 - REFERENCES, IMF working paper content provided.*

### Appendix 2 provides the full list of countries that are available as part of the comparator group

### wp17145 - Appendix 2 provides the full list of countries that are available as part of the comparator group

### Synthetic Control Method: description and limitations
- Purpose:
  - Provides a data-driven methodology to quantify the effects of a particular event in comparative case studies by creating an artificial counterfactual (synthetic unit) matched to the unit of interest prior to the event.
- Interpretation:
  - The divergence between the treated unit and its synthetic unit after the event is interpreted as the quantitative estimate of the event’s effect over a specified number of periods.
- Advantages:
  - Captures heterogeneity in effects by producing case-specific estimates (contrast with average estimates from panel regressions).
  - Addresses omitted-variable bias from time-variant fixed effects.
- Limitations and potential biases:
  - Can suffer from reverse causation if the decision to undertake fiscal reforms was affected by expectations on future growth.
  - Estimates can be biased by subsequent events that affect the treated unit and its counterfactual differently (example: radical political revolutions after natural disasters).
  - Standard inference techniques are not directly applicable; placebo experiments are used for robustness.

### Synthetic Control Method: implementation steps
- Step 1: Choose comparator countries and explanatory variables
  - Comparator countries: similar to the country of interest but unaffected by the event within the sample period.
  - Predictor variables in initial country selection: level of GDP per capita (at the beginning, half-way, and the end of the pre-reform period), trade openness, inflation rate, terms of trade, index of human capital.
  - Outcome variable used: real GDP at constant 2005 prices (in million 2005 US$). (Note: alternative specification using real GDP per capita yields similar main results, but growth effect is negative in two cases: Chile 1974 and New Zealand 1986.)
- Step 2: Construct the synthetic series via nested optimization
  - Country weights vector w* chosen to minimize distance between pre-event predictors for unit of interest and comparators, subject to weights between zero and one.
  - Variable weight matrix V* chosen among positive definite diagonal matrices to minimize mean square prediction error (MSPE) of the outcome variable over the pre-event period.
  - Iterative optimization updates V* and w* to construct the synthetic post-event path.
- Step 3: Compare actual and post-event outcome variables
  - The difference between actual and synthetic post-event outcome series is the estimated impact (assuming other factors are controlled).

### Baseline results: growth effects of fiscal structural reforms
- Aggregate finding:
  - Fiscal reforms led to higher growth relative to the synthetic unit in all reform episodes, with heterogeneity across episodes.
- Key statistics (10-year horizon):
  - Average annual GDP growth for the 10 years after the fiscal reform started was higher in the reform country relative to its synthetic units by about 1 percentage point.
  - Range across episodes: from 0.1 percentage point (Germany 2003) to 4.3 percentage points (Chile 1983).
  - Excluding Chile 1983 and Ireland 1987 reduces the average growth impact to 0.3 percentage point.
- Timing:
  - Speed of materialization varied: quick in Chile and Ireland; more modest in first few years in other cases.
- Development-stage association:
  - Episodes in countries with relatively lower development at time of reform were associated with higher subsequent growth effects (e.g., Chile 1983 and Ireland 1987).
- Placebo experiments (robustness):
  - About half of baseline results are robust to placebo experiments.
  - Very few placebo permutations exceed treated-country growth effects at 10 years for:
    - Belgium 1992 (1/4),
    - Chile 1983 (0/8),
    - Ireland 1987 (0/3),
    - Netherlands 1983 (0/6).
  - Baseline estimates not robust for:
    - Chile 1974 (4/7),
    - Germany 2003 (20/21).
  - Borderline cases:
    - Australia 1985 (4/20),
    - Australia 1998 (4/18),
    - New Zealand 1986 (5/19).
  - Overall: two-thirds of events are either robust or borderline robust to placebo assessment.

### Non-fiscal structural reforms: methods to control and main findings
- Rationale:
  - Structural reforms in product market, labor market, business regulations, legal environment, financial reforms can affect long-term growth and potentially bias estimates if the treated unit had more non-fiscal reforms than its synthetic unit.
- Approaches:
  - Pre-reform differences: include non-fiscal structural reform indices one at a time as additional predictors in SCM to ensure treated and synthetic units were similar prior to fiscal reform.
  - Concurrent/post-reform differences: document ten-year changes in structural reform indices and compare treated vs synthetic; construct “non-fiscal structural reform gaps” in placebo permutations and assess p-values.
- Controlled structural reform indicators:
  - Business regulations, financial reforms, legal reforms, labor market reforms, product market reforms (added one at a time).
- Results controlling for pre-reform non-fiscal structural reforms:
  - Baseline estimates generally confirmed.
  - Weighted-average finding: on average, output growth was higher by 0.9 percentage points relative to control units, ranging from -0.7 percentage point (New Zealand 1983) to 4.3 percentage points (Chile 1983).
  - Some point estimates for Australia 1985, Australia 1998, and New Zealand 1986 show negative growth effects after including structural reform predictors.
  - Exception: New Zealand 1986 growth effect turns negative when controlling for non-fiscal structural reforms (possible omitted variable bias in baseline).
- Concurrent non-fiscal reforms assessment:
  - Two exercises undertaken:
    1. Ten-year changes in non-fiscal structural reform indicators: compare whether treated or synthetic had more reforms.
    2. Non-fiscal structural reform gaps in placebo permutations: assess p-values to gauge whether treated unit had unusually strong concurrent structural reform progress.
  - General summary:
    - Non-fiscal reforms were more pronounced in treated units but not predominant.
    - p-values for non-fiscal structural reform gaps are often large, suggesting a subdued role of non-fiscal reforms; p-values were larger than those for growth effects in almost all cases, highlighting fiscal reforms as likely key contributors.
  - Caveat:
    - Magnitude of growth effects caused by non-fiscal structural reforms varies across countries; quantitative assessment remains an open question.

### Non-fiscal structural reform indicators: summary statistics (Table 3)
- Table 3: Average Percentage Change Difference in Index (Difference = treated unit minus synthetic unit, ten years after start of fiscal reform)
  - Australia 1985:
    - Business Regulation: 10.2
    - Financial: 34.4
    - Labor Market: n.a.
    - Legal: 1.9
    - Product Market: 1.3
  - Australia 1998:
    - Business Regulation: 21.1
    - Financial: 0.0
    - Labor Market: -17.6
    - Legal: -7.0
    - Product Market: 2.2
  - Belgium 1992:
    - Business Regulation: 3.1
    - Financial: -40.3
    - Labor Market: n.a.
    - Legal: -23.8
    - Product Market: -6.1
  - Chile 1974:
    - Business Regulation: n.a.
    - Financial: 46.7
    - Labor Market: n.a.
    - Legal: -59.2
    - Product Market: n.a.
  - Chile 1983:
    - Business Regulation: 14.7
    - Financial: 32.4
    - Labor Market: n.a.
    - Legal: 2.8
    - Product Market: n.a.
  - Germany 2003:
    - Business Regulation: 11.0
    - Financial: n.a.
    - Labor Market: 13.1
    - Legal: -3.3
    - Product Market: -4.4
  - Ireland 1987:
    - Business Regulation: -18.5
    - Financial: 18.0
    - Labor Market: n.a.
    - Legal: 5.0
    - Product Market: -4.3
  - Netherlands 1983:
    - Business Regulation: 1.7
    - Financial: 5.0
    - Labor Market: n.a.
    - Legal: 1.2
    - Product Market: -18.7
  - New Zealand 1986:
    - Business Regulation: 9.6
    - Financial: -24.5
    - Labor Market: n.a.
    - Legal: 7.9
    - Product Market: -25.2
- “More Reform in Treated/Total” counts (number of episodes where structural reform preceded more in treated unit / total episodes with data):
  - Business Regulation: 7/8
  - Financial: 6/8
  - Labor Market: 1/2
  - Legal: 5/9
  - Product Market: 5/7

### Non-fiscal structural reform gaps vs growth effect (Table 4)
- For six cases where baseline growth effect was robust or borderline robust, comparison of growth placebo permutations and non-fiscal reform gaps:
  - Belgium 1992:
    - Growth effect placebos: 1/4 better than treated
    - Business Regulations: 2/4 better than treated
    - Legal System in Property Rights: 3/4 better than treated
  - Chile 1983:
    - Growth effect placebos: 0/8 better than treated
    - Business Regulations: 2/8 better than treated
    - Legal System in Property Rights: 4/8 better than treated
  - Ireland 1987:
    - Growth effect placebos: 0/3 better than treated
    - Business Regulations: 2/3 better than treated
    - Legal System in Property Rights: 1/3 better than treated
  - Netherlands 1983:
    - Growth effect placebos: 0/6 better than treated
    - Business Regulations: 1/4 better than treated
    - Legal System in Property Rights: 3/6 better than treated
  - Australia 1985:
    - Growth effect placebos: 4/20 better than treated
    - Business Regulations: 7/20 better than treated
    - Legal System in Property Rights: 8/20 better than treated
  - Australia 1998:
    - Growth effect placebos: 4/18 better than treated
    - Business Regulations: 1/18 better than treated
    - Legal System in Property Rights: 17/18 better than treated

### Inequality: methods and key findings
- Data and measures:
  - Use Solt (2014)’s Standardized World Income Inequality Database (SWIID).
  - Track market income Gini and net income Gini (net after fiscal redistribution).
- Scope:
  - Seven cases with available data (Chile 1974 and Netherlands 1983 dropped).
  - Use baseline SCM weights to construct synthetic unit Gini series (documentary, not causal).
- Findings relative to global trend:
  - Both market income Gini and net income Gini increased in five out of seven cases considering ten-year averages after reform started (exceptions: Chile 1983 and Ireland 1987).
  - Net income Gini increased less or decreased more than market income Gini in most cases (exceptions: Belgium 1992 and Ireland 1987), suggesting fiscal redistribution partially offset market-income inequality increases.
  - For Germany 2003 nine-year averages were used as the Gini series ends in 2012.
- Findings relative to synthetic units:
  - Market income Gini increased more in four out of seven reform cases relative to synthetic units: Australia 1985, Australia 1998, Germany 2003, New Zealand 1986.
  - Fiscal redistribution increased more in those four cases as well.
  - Cases where market income Gini increased less than synthetic unit and fiscal redistribution also increased less: Belgium 1992 and Ireland 1987.
  - Chile 1983: market Gini increased less than synthetic unit while fiscal redistribution increased more.
  - Overall: no consistent pattern that fiscal reforms led to higher market-income inequality or a systematic weakening of fiscal redistribution in reforming countries.
- Caveats:
  - Results are associations, not causal claims.
  - Data limitations prevented re-estimating SCM using both growth and inequality as outcome variables.

### Concluding remarks and policy-relevant implications
- Summary of evidence:
  - Countries identified as exposed to growth-friendly fiscal reforms experienced higher growth relative to their synthetic units, with larger impacts for countries at lower initial development levels.
  - Results broadly robust to placebo permutations and controlling for pre-reform non-fiscal structural reforms (business regulation, financial, labor market, legal, product market).
  - No definitive evidence that non-fiscal structural reforms were systematically stronger in reform countries relative to their synthetic units after reforms started.
  - No clear-cut evidence of a systematic trade-off between growth-friendly fiscal reforms and increases in inequality at the aggregate level; no evidence pointing to causality from structural fiscal reforms to more unequal income distribution.
- Suggested avenues for further research and policy analysis:
  - Identify whether broadening tax bases (rather than changes in standard rates) yields more desirable growth benefits.
  - Assess the impact of structural expenditure measures controlling for quality across spending categories.
  - Explore these questions using the synthetic control method where data availability permits.

*Source: Authors’ calculations and analysis in the provided content unit.*

### Appendix 1. Country Selection

### Appendix 1. Country Selection

### Quantitative Reform Indicators — European countries (EUR)
- Netherlands* — High Income: OECD — 8
- Ireland* — High Income: OECD — 8
- Germany* — High Income: OECD — 8
- Belgium* — High Income: OECD — 8
- Sweden — High Income: OECD — 7
- United Kingdom — High Income: OECD — 7
- Denmark — High Income: OECD — 7
- Finland — High Income: OECD — 7
- Hungary — Upper Middle Income: OECD — 7
- Greece — High Income: OECD — 7
- Portugal — High Income: OECD — 7
- Italy — High Income: OECD — 7
- Turkey — Upper Middle Income: OECD — 7
- Spain — High Income: OECD — 7
- Luxembourg — High Income: OECD — 7
- Austria — High Income: OECD — 6
- Slovenia — High Income: OECD — 6
- Israel — High Income: OECD — 6
- France — High Income: OECD — 5
- Cyprus — High Income: non-OECD — 5
- Poland — High Income: OECD — 4
- Switzerland — High Income: OECD — 4
- Iceland — High Income: OECD — 4
- Czech Republic — High Income: OECD — 4
- Bulgaria — Upper Middle Income: OECD — 4
- Slovak Republic — High Income: OECD — 2
- Estonia — High Income: OECD — 2
- Belarus — Upper Middle Income — 2
- Lithuania — High Income — 1
- Malta — High Income — 1

Note: (*) Selected countries.  
Source: Author's calculations.  
1 World Bank’s country classification by income.

### Quantitative Reform Indicators — Asian-Pacific countries (APD)
- Australia* — High Income: OECD — 6
- New Zealand* — High Income: OECD — 6
- Korea — High Income: OECD — 5
- Thailand — Upper Middle Income — 5
- Singapore — High Income — 5
- Japan — High Income: OECD — 4
- Malaysia — Upper Middle Income — 3
- China — Upper Middle Income — 2
- Hong Kong SAR — High Income — 1

Note: (*) Selected countries.  
Source: Author's calculations.  
2 World Bank’s country classification by income.

### Quantitative Reform Indicators — Western Hemisphere countries (WHD)
- Chile* — High Income: OECD — 6
- Canada — High Income: OECD — 5
- Uruguay — High Income — 5
- Panama — Upper Middle Income — 5
- Costa Rica — Upper Middle Income — 4
- Colombia — Upper Middle Income — 4
- United States — High Income: OECD — 4
- Dominican Republic — Upper Middle Income — 4
- Argentina — Upper Middle Income — 4
- Peru — Upper Middle Income — 3
- Jamaica — Upper Middle Income — 3
- Brazil — Upper Middle Income — 2

Note: (*) Selected countries.  
Source: Author's calculations.  
2 World Bank’s country classification by income.

### Comparator Countries — Appendix 2
- Type A: Countries with less than 5 reforms. Cases: Australia, Germany, and New Zealand
  - Albania, Argentina, Belarus, Bosnia and Herzegovina, Brazil, Bulgaria, Canada, China, Colombia, Costa Rica, Croatia, Cyprus, Czech Republic, Dominican Republic, Estonia, France, Germany, Hong Kong SAR, Iceland, Jamaica, Japan, Korea, Latvia, Lithuania, Malta, Malaysia, Malta, Panama, Peru, Poland, Serbia, Singapore, Slovak Republic, Switzerland, Thailand, United States, Uruguay
- Type B: Countries with less than 5 reforms and are in the same region. Cases: Chile
  - Argentina, Brazil, Canada, Colombia, Costa Rica, Dominican Republic, Jamaica, Panama, Peru, United States, Uruguay
- Type B: Cases: Belgium, Ireland, and Netherlands
  - Albania, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, Cyprus, Czech Republic, Estonia, France, Germany, Iceland, Latvia, Lithuania, Malta, Poland, Serbia, Slovak Republic, Switzerland

Source: Author's calculations.

*Source: Author's calculations.*

### Chapter 3.

### Chapter 3.

### References cited in Chapter 3
- Jaumotte, Lall, and Papageorgiou, 2013, “Rising Income Inequality: Technology, or Trade and Financial Globalization?”, IMF Economic Review, 61, pp. 271-309
- Klobner and Pfeifer, 2015, “Synthesizing Cash for Clunkers: Stabilizing the Car Market, Hurting the Environment”, Unpublished Manuscript.
- Kumhof, Ranciere, and Winant, 2015, “Inequality, Leverage, and Crises”, American Economic Review 105:3, pp. 1217-45.
- Martínez-Vazquez, Jorge, Violeta Vulovic, and Blanca Moreno Dodson, 2012, “The Impact of Tax and Expenditure Policies on Income Distribution: Evidence from a Large Panel of Countries.” Hacienda Pública Española 200.
- Muinelo-Gallo and Roca-Sagalés, 2013, “Joint determinants of fiscal policy, income inequality and economic growth”, Economic Modelling, 30, pp. 814-24.
- Newiak, Monique and Tim Willems (2016), “Evaluating the Impact of IMF Programs Using the Synthetic Control Method”, Unpublished Manuscript.
- OECD, 1999, “OECD Economic Surveys: New Zealand”.
- Ostry, Berg, and Tsangarides, 2014, “Redistribution, Inequality, and Growth”, IMF Staff Discussion Note 14/02.
- Prati, Onorato and Papageorgiou, 2013, “Which Reforms Work and Under What Institutional Environment? Evidence from a New Data Set on Structural Reforms”, The Review of Economics and Statistics, 95:3, pp. 946-968.
- Solt, 2014, “The Standardized World Income Inequality Database”.
- Trabandt and Uhlig, 2011, “The laffer curve revisited”, Journal of Monetary Economics, 58, pp. 305-327.
- Woo, Jaejoon, Elva Bova, Tidiane Kinda, and Y. Sophia Zhang (2013), “Distributional Consequences of Fiscal Consolidation and the Role of Fiscal Policy: What Do the Data Say?” IMF Working Paper WP/13/195.

*Source: Chapter 3, wp17145 - Chapter 3.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17145.pdf_