## Section V — Implications for policymakers in advanced economies (wpiea2021156-print-pdf)

## Source details

**Canonical URL:** [Section V — Implications for policymakers in advanced economies (wpiea2021156-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021156-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021156-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021156-print-pdf.pdf.json)

---

### Political-economy framing of climate change policies
- Short-term political constraints:
  - CCPs face resistance because short-term adjustment costs are significant while benefits materialize only in the very long run.
  - Benefits of CCPs are not directly observable because they materialize in the absence of environmental damage.
  - Costs are concentrated on a few stakeholders while benefits are widely distributed, generating strong concentrated opposition and diffuse support.
  - Regressivity: CCPs (especially market-based instruments such as emission taxes) can heighten inequality, tightening budgets of the poor and fueling political opposition.
  - International free-riding: stricter CCPs can create relocation of production and competitive disadvantage, producing incentives for a race-to-the-bottom.
- Mitigation of distributional consequences and political opposition:
  - Design progressive instruments where possible.
  - Implement targeted redistributive policies or time CCPs to periods when energy costs are low.
  - Provide targeted assistance programs for risk groups (the unemployed, the elderly) and invest in training and active labor market programs for workers in "dirty" sectors.

### Data, empirical strategy, and measurement
- Policy measure:
  - Environmental Policy Stringency (EPS) indicators (OECD, Botta and Kozluk, 2014), scaled from 0 (not stringent at all) to 6 (very stringent); breakdown by instrument type (market-based vs non-market-based).
- Political outcome:
  - Governmental popular support from the International Country Risk Guide (ICRG, 2020), indicator ranges between 0 (very high risk) and 4 (very low risk); annual coverage from 2001 onwards.
- Sample coverage:
  - 31 countries over the period 2001–15, covering 43 percent of total global carbon emissions (Muntean et al., 2018).
- Baseline estimating equation:
  - 푦_{i,t} = 훽 ΔEPS_{i,t} + 휃 X_{i,t} + α_i + γ_t + ε_{i,t}
- Main covariates controlled: deficit (as % of GDP), elderly (65+) as % of population, real GDP growth rate (gdpr), unemployment rate (unemprate), consumer price index (cpi), lagged government voteshare, taxrevue (% GDP), and Alesina et al. reform indicators.

### Main empirical findings — aggregate effects and instrument heterogeneity
- Aggregate political cost (baseline OLS):
  - A major change in CCPs—equivalent to an increase in the EPS indicator from the 25th to the 75th percentile of the EPS distribution—reduces popular support by about 10 percent (baseline OLS, Table 3).
  - Key D.EPS coefficients across specifications:
    - (I) -0.231*** (0.079)
    - (II) -0.246*** (0.079)
    - (III) -0.247*** (0.080)
    - (IV) -0.266*** (0.086)
    - (V) -0.295*** (0.084)
    - (VI) -0.284*** (0.088)
  - Interpretation note: 0.2 is equivalent to a 10 percent decline in popular support from an increase in EPS from the 1st to the 3rd quartile of the EPS distribution.
- Vote share outcome:
  - The same major change in EPS reduces incumbent vote share by more than 7 percentage points (based on a smaller sample of 108 observations, Table A2).
- Market-based vs non-market-based instruments:
  - Market-based measures produce negative and sizeable effects on governmental support:
    - D.EPS_MKT: -0.153** (0.074) (Table 4, column I)
    - Selected tax-related coefficients:
      - D.TAXES: -0.199** (0.092)
      - D.taxPCA: -0.202*** (0.062)
      - D.TAXCO2: -0.071** (0.035)
      - D.TAXNOX: -0.090** (0.042)
      - D.TAXSOX: -0.196*** (0.052)
  - Non-market-based measures are typically not statistically significant in reducing popular support:
    - D.EPS_NMKT: -0.066 (0.048) (Table 5)
    - Notable exception: D.ELV_DIESELSO: -0.206** (0.103)
  - Interpretation: price mark-ups (visible to consumers) are more traceable and politically costly than supply limits.

### Heterogeneity and mediating factors (fuel prices, dirty-energy share, inequality, social protection)
- Fuel prices:
  - Interaction estimates:
    - D.EPS x high_oilPrice: -0.320** (0.129) (Table 6, column II)
    - D.EPS x high_gasPrice: -0.493*** (0.170) (Table 6, column II)
  - IV-mediated estimates:
    - D.EPS_high_oilPrice: -1.241** (0.608) (Table 12-1, column II; 1st-stage F-Stat 12.161)
    - D.EPS_high_gasPrice: -1.184* (0.630) (Table 12-1, column IV; 1st-stage F-Stat 12.82)
  - When fuel prices are low, the effect of CCPs is not statistically different from zero.
- Dirty-energy sector share:
  - Interaction estimates:
    - D.EPS x high_energyShare: -0.264** (0.119) (Table 6, column IV)
  - IV-mediated estimate:
    - D.EPS_high_energyShare: -3.675** (1.769) (Table 12-2, column II; 1st-stage F-Stat 9.85)
  - In IV estimates, political costs are three times as large in countries with high dirty-energy input shares compared with countries with low shares.
- Initial EPS and policy salience:
  - D.EPS x high_initialEPS: -0.321*** (0.076) (Table 6, column IV)
  - IV: D.EPS_high_initialEPS: -1.343*** (0.513) (Table 12-2, column IV; 1st-stage F-Stat 25.166)
- Inequality and income shares:
  - Interaction estimates:
    - D.EPS x high_top1share: -0.300* (0.156) (Table 7, column II)
    - D.EPS x high_top10share: -0.320** (0.155) (Table 7, column IV)
    - D.EPS x high_top20share: -0.320** (0.146) (Table 7, column VI)
    - D.EPS x high_giniMRKT: -0.380** (0.170) (Table 7 cont., column II)
    - D.EPS x high_giniNET: -0.402** (0.162) (Table 7 cont., column IV)
  - IV heterogeneous results (Table 12-3 and Table 12-5):
    - D.EPS_high_top1share: -3.495*** (0.561)
    - D.EPS_high_top10share: -4.482*** (0.862)
    - D.EPS_high_top20share: -5.500*** (1.546)
    - D.EPS_high_giniMRKT: -4.569*** (1.757)
    - D.EPS_high_giniNET: -3.343*** (0.852)
    - 1st-stage F-Stats reported for these IV specs include 73.36, 30.61, 17.65, among others.
- Social protection and redistribution:
  - Low social expenditure amplifies negative ΔEPS effect:
    - D.EPS x low_socexp: -0.504*** (0.167) (Table 7 cont., column III)
    - D.EPS x low_almpExp: -0.483*** (0.158) (Table 7 cont., column V)
    - D.EPS x low_unempExp: -0.544*** (0.170) (Table 7 cont., column V)
  - IV results (Table 12-6):
    - D.EPS_low_socexp: -1.492*** (0.523) (1st-stage F-Stat 44.32)
    - D.EPS_low_almpExp: -1.863* (1.037) (1st-stage F-Stat 76.26)
    - D.EPS_low_unempExp: -1.860** (0.834) (1st-stage F-Stat 36.28)
  - Generous social expenditure (overall, ALMP-related, unemployment benefits) mitigates political costs.

### Identification strategy, IV evidence, and robustness
- Main endogeneity concerns:
  - Reverse causality: adoption of CCPs may require political support, biasing OLS.
  - Alternative bias: unpopular governments may implement CCPs because "it has nothing to lose," or natural disasters may simultaneously affect policy and preferences.
- Instrument strategy:
  - Interact time-varying global "policy pressure" variables (eventsFlood, majorHurric, peopleAffectedEQ, eventsWildfire) with country-specific vulnerability terms (coastLengthWRI, cenDist, urbanCoastLOG, agriLandPC) to form five instruments; baseline instrument is eventsFlood × coastLengthWRI.
- IV estimates and magnitude:
  - IV D.EPS coefficients (Table 8) range from -0.635* to -1.047** across IV specifications:
    - (I) -0.994*** (0.386)
    - (II) -0.741** (0.361)
    - (III) -0.635* (0.341)
    - (IV) -1.047** (0.533)
    - (V) -0.879** (0.377)
  - IV coefficient estimates are approximately three times larger than the OLS estimates, suggesting OLS is biased towards zero.
- Instrument validity and diagnostics:
  - First-stage F-statistics indicate instruments are sufficiently strong; Kleibergen‒Paap rk Wald F statistic exceeds Stock-Yogo critical values in reported specifications (examples: 1st-stage F-Stat 30.79; 1st-stage F-Stat 42.05).
  - Direct-instrument tests: including instruments as controls yields insignificant direct coefficients (Table 9); residuals are not associated with instruments (Table 10).
  - Robustness: adding controls for global trends (globalization, conflict/riot indices) and interactions with country size does not alter negative D.EPS IV coefficients (Table 11).

### Policy implications and recommendations
- Instrument choice matters:
  - Market-based instruments (emission taxes) are efficient but politically costly; non-market-based instruments (emission limits, R&D subsidies, regulation) typically do not carry significant political costs and remain viable second-best options.
  - Policy advice should balance economic efficiency with political feasibility to avoid the perfect becoming the enemy of the good.
- Timing matters:
  - Adopt CCPs when world energy prices are low to reduce visible impacts on household energy bills and to mitigate political backlash.
  - The paper notes the "current environment of very low world prices presents an opportunistic window to advance greener policies."
- Distributional mitigation:
  - High inequality magnifies political costs; targeted redistributive measures and trampoline policies (unemployment benefits, active labor market policies, retraining) aimed at losers from CCPs reduce political fallout.
  - Social protection generosity offsets regressivity and can render CCPs politically feasible.
- Structural diversification:
  - Economies reliant on dirty industries face greater political risk from CCPs; ex-ante policies to enhance sectoral and regional diversification will reduce political vulnerability and potential ballot-box punishment.
- Overarching conclusion:
  - CCPs often reduce government popularity, but effective strategies—choice of instrument, opportunistic timing, targeted redistribution, and structural diversification—can limit or remove political costs and help overcome the equilibrium of inaction.

*Source: wpiea2021156-print-pdf*

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

### References

### I. Introduction — context and motivation
- Climate change is framed as causing an "environmental and human catastrophe" with heightened global attention following youth-led advocacy (a "school strike by a sixteen-year old in Sweden") and the 2015 Paris Agreement.
- Economic evidence cited:
  - Industrial countries are estimated to suffer losses of "½-2 percent of GDP" if global temperatures rise "2-4 degrees Celsius by 2100" (Hsiang et al., 2017).
  - Kahn et al. (2019) project that unmitigated climate change will reduce global real GDP per capita by "more than 7 percent by the end of the century."
  - Burke et al. (2015) indicates that unmitigated global warming would likely lead to global GDP declining "by up to one quarter by 2100" (IMF, 2019).

### II. Research focus and contribution
- Objective: estimate the average effect of climate change policies (CCPs) on governmental support ("Index of Popular Support" from ICRG) using OECD’s Environmental Policy Stringency (EPS) indicators.
- Data sample: "34 advanced and emerging market economies" for the period "2001–2015".
- Key baseline finding:
  - CCPs reduce popular support on average, but this effect "mainly reflects the impact of market-based policy measures, such as emission taxes, rather than non-market-based measures, such as emission limits."
  - Non-market-based measures are described as "still efficient tools to combat climate change (IMF, 2019a) while being more politically viable (IMF, 2019)."

### III. Heterogeneity and mediating factors
- Political costs of CCPs are larger under specific conditions:
  - "in times of high fuel prices;"
  - "for countries with a high input share of dirty energy in mining and production;"
  - "when CCPs are adopted when inequality is high;"
  - "when social expenditures do not compensate the losers from CCPs."
- Counterfactual statement: "in the counterfactual to all of these situations, the political cost of CCPs is not statistically different from zero."

### IV. Identification strategy and endogeneity concerns
- Main methodological challenges:
  - Reverse causality: adoption of CCPs may require political support, implying potential bias in OLS estimates.
  - Alternative bias: unpopular governments may implement CCPs because "it has nothing to lose," or natural disasters may simultaneously affect policy and preferences.
- Instrumental Variable (IV) approach:
  - Instruments exploit an interaction between a time-varying global term and country-based geographical vulnerability indicators.
  - Global term: occurrence and damage of global extreme weather events (used as a proxy for "policy pressure" induced by climate change), assumed exogenous to government support.
  - Country-specific term: geographical indicators such as "the length of the coastline" and "the urban extent in coastal zones" as proxies for vulnerability.
  - The analysis controls for country and time fixed effects and is characterized as a differences-in-differences approach.
- Result from IV analysis:
  - Endogeneity in the OLS estimator is "likely to result in underestimation of the true causal effect." IV results are said to be robust to various instrument and validity tests.

### V. Framing within the literature and policy implications
- Political supply-side factors:
  - The rise of green parties and growing salience of environmental concerns increase incentives for mainstream parties to adopt green issues.
- Policy implication emphasized:
  - Market-based instruments (e.g., emission taxes) appear to carry political costs not observed for non-market-based instruments; this is notable given economists’ preference for Pigouvian taxation as "first-best corrective tool for carbon emissions."
- Distributional considerations:
  - Literature suggests revenue recycling and targeted compensation mechanisms can reduce efficiency-equity trade-offs associated with emission taxation.

*Source: wpiea2021156-print-pdf - References*

### Section V concludes with some implications for policymakers in advanced economies.

### Section V — Implications for policymakers in advanced economies (wpiea2021156-print-pdf)

### Political-economy framing of climate change policies
- Parallel to structural reforms: CCPs face resistance and lack of political will despite long-run welfare gains due to veto players required to distribute benefits and costs to satisfy all.
- Key political constraints:
  - Short-term adjustment costs are significant while benefits materialize only in the very long run.
  - Benefits of CCPs are not directly observable because they materialize in the absence of environmental damage.
  - Costs are concentrated on a few stakeholders while benefits are widely distributed, generating strong concentrated opposition and diffuse support.
  - Regressivity: CCPs (especially market-based instruments such as emission taxes) can heighten inequality, tightening budgets of the poor and fueling political opposition.
  - International free-riding: Countries that adopt stricter CCPs may face relocation of production and competitive disadvantage, creating incentives for a race-to-the-bottom.
- Distributional consequences and political opposition can be mitigated by:
  - Designing progressive instruments where possible.
  - Targeted redistributive policies or timing CCPs to periods when energy costs are low.
  - Targeted assistance programs for risk groups (the unemployed, the elderly) and investment in training and active labor market programs for workers in "dirty" sectors.

### Data and empirical strategy
- Policy measure: Environmental Policy Stringency (EPS) indicators (OECD, Botta and Kozluk, 2014), scaled from 0 (not stringent at all) to 6 (very stringent); breakdown by instrument type (market-based vs non-market-based).
- Political outcome: governmental popular support from the International Country Risk Guide (ICRG, 2020), indicator ranges between 0 (very high risk) and 4 (very low risk); annual coverage from 2001 onwards.
- Sample coverage: 31 countries over the period 2001–15, covering 43 percent of total global carbon emissions (Muntean et al., 2018).
- Baseline regression relates change in EPS (ΔEPS) to change in incumbent popular support, controlling for country and year fixed effects and economic, political and demographic covariates (including fiscal deficit as percentage of GDP, share of elderly (65+) as percentage of population, real GDP growth rate, unemployment rate, consumer price index, lagged government vote share, tax revenue as percentage of GDP, and Alesina et al. (2020a) reform indicators).

### Main empirical findings
- Aggregate political cost:
  - A major change in CCPs—equivalent to an increase in the EPS indicator from the 25th to the 75th percentile of the EPS distribution—reduces popular support by about 10 percent (baseline OLS, Table 3).
  - Using vote share of the incumbent government coalition as outcome: the same major change in EPS reduces the vote share by more than 7 percentage points (based on a smaller sample of 108 observations, Table A2).
- Instrument heterogeneity:
  - Market-based measures, especially taxes on emissions, produce negative and sizeable effects on governmental support (Table 4).
  - Non-market-based measures are typically not statistically significant in reducing popular support (Table 5), with the notable exception of nonmarket-based EPS on diesel which is politically salient.
  - Interpretation: price mark-ups (visible to consumers) are more traceable and politically costly than supply limits.
- Mediating factors and conditional effects:
  - Global fuel prices: in times of high oil and gasoline prices, political damage from CCPs is statistically significantly negative with coefficients 1.5 to 2 times as large as the baseline direct effect (Table 6, column 1-2). When fuel prices are low, the effect is not statistically different from zero.
  - Dirty-energy sector share: countries with a large input share of dirty-energy mining industries face larger political obstacles; in IV estimates, political costs are three times as large as in countries with low dirty-energy input shares (Table 12-2).
  - Inequality: CCPs implemented in times of high income inequality (market- and net-based Gini; high top-income shares; low bottom income shares) produce significantly larger political costs; effects are insignificant when inequality is low (Table 7; Table 12-3/12-4/12-5).
  - Social protection: generous social expenditure (overall, ALMP-related, unemployment benefits) mitigates political costs; lower social expenditure makes EPS rises particularly damaging (Table 7; Table 12-6).
  - Non-significant mediators: aggregate growth regimes, redistribution measured by difference between market and net Gini, vote share of green parties, opposition fractionalization, and democratic accountability did not systematically mediate CCP political costs.

### Identification and robustness (IV evidence)
- Motivation: address omitted variable bias, reverse causality, and measurement error concerns.
- Instrument strategy: interact time-varying global "policy pressure" variables (number of flood events, number of major hurricanes in the North Atlantic, number of people affected by earthquakes, number of wildfires) with constant country-specific vulnerability terms (length of coastline, minimum distance of centroid to coast, share of urban population in coastal area, agricultural land per capita). Five instruments constructed as interactions; baseline instrument is global number of flood events × country coastline length.
- Instrument validity:
  - First-stage F-statistics indicate instruments are sufficiently strong; Kleibergen‒Paap rk Wald F statistic exceeds Stock-Yogo critical values.
  - Direct-instrument tests: including instruments as controls in baseline model yields insignificant coefficients (Table 9); residuals are not associated with instruments (Table 10).
  - Controls for alternative global trends (globalization index, global conflict/riot indices) and interactions with country size do not alter results (Table 11).
- IV estimates:
  - IV results support OLS findings: ΔEPS changes have a significant negative effect on popular support for all five instruments with similar-sized coefficients (Table 8).
  - Magnitude: IV coefficient estimates are approximately three times larger than the OLS estimates, suggesting OLS is biased towards zero.
  - IV extended models confirm mediating effects: higher oil and gas prices, higher dirty-energy input shares, and higher inequality amplify political costs; generous social expenditure mitigates them (Table 12).

### Policy implications and recommendations
- Instrument choice matters:
  - Market-based instruments (emission taxes) are efficient but politically costly; non-market-based instruments (emission limits, R&D subsidies, regulation) typically do not carry significant political costs and remain viable second-best options.
  - Policy advice should balance economic efficiency with political feasibility to avoid the perfect becoming the enemy of the good.
- Timing matters:
  - Adopt CCPs when world energy prices are low to reduce visible impacts on household energy bills and to mitigate political backlash.
  - Current environment of very low world prices presents an opportunistic window to advance greener policies.
- Distributional mitigation:
  - High inequality magnifies political costs; targeted redistributive measures and trampoline policies (unemployment benefits, active labor market policies, retraining) aimed at losers from CCPs reduce political fallout.
  - Social protection generosity offsets regressivity and can render CCPs politically feasible.
- Structural diversification:
  - Economies reliant on dirty industries face greater political risk from CCPs; ex-ante policies to enhance sectoral and regional diversification will reduce political vulnerability and potential ballot-box punishment.
- Overarching conclusion:
  - CCPs often reduce government popularity, but effective strategies—choice of instrument, opportunistic timing, targeted redistribution, and structural diversification—can limit or remove political costs and help overcome the equilibrium of inaction.

*Source: wpiea2021156-print-pdf — Section V concludes with some implications for policymakers in advanced economies.*

### REFERENCES

### wpiea2021156-print-pdf - REFERENCES

### References (bibliography highlights)
- Extensive bibliography covering political economy, climate policy, environmental taxation, carbon pricing, public opinion on climate and nuclear power, measurement of environmental policy stringency, macroeconomic impacts of climate change, and related empirical methods.
- Key recurring themes in the citations: carbon taxation and revenue recycling; environmental policy stringency measurement; political drivers of renewable energy and climate policy; distributional impacts and inequality; climate-related economic damages and adaptation; methodology for policy position measurement and model uncertainty.

### Table 1. Variable description (selected variables and exact descriptions)
- D_Popularsupport: ICRG level of support for the government
- D.EPS: OECD Environmental Policy Stringency Index (aggregated) (first diff)
- D.EPS_MKT: OECD marked-based environmental policy stringency index (first diff)
- D.EPS_NMKT: OECD non-marked-based environmental policy stringency index (first diff)
- D.TAXES: OECD aggregated environmental taxation index (first diff)
- D.taxPCA: OECD PCA (first) of taxation sub-indices (first diff)
- D.TAXCO2: OECD taxation on CO2 index (first diff)
- D.TAXDIESEL: OECD taxation on Diesel index (first diff)
- D.TAXNOX: OECD taxation on NOx index (first diff)
- D.TAXSOX: OECD taxation on SOx index (first diff)
- D.STD: OECD aggregated limits index (first diff)
- D.limitPCA: OECD first principale component of all limits sub-indices (first diff)
- D.ELV_DIESELSO: OECD emission limits on Diesel index (first diff)
- D.ELV_NOX: OECD emission limits on NOx index (first diff)
- D.ELV_PM: OECD emission limits on PM index (first diff)
- D.ELV_SOX: OECD emission limits on SOx index (first diff)
- D.FIT: OECD aggregated feed-in-tariffs index (first diff)
- D.FIT_SOLAR: OECD feed-in-tariffs for solar index (first diff)
- D.FIT_WIND: OECD feed-in-tariffs for wind index (first diff)
- D.TRADESCH: OECD aggregated trading scheme index (first diff)
- D.TRADESCH_CO2: OECD trading scheme for CO2 index (first diff)
- D.TRADESCH_EEF: OECD trading scheme for Energy Efficiency Certificate index (first diff)
- D.TRADESCH_REC: OECD trading scheme for Renewable Energy Certificate index (first diff)
- D.RD_RE: OECD R&D subsidies for renewable energy index (first diff)
- D.RD_SUB: OECD aggregated R&D subsidies index (first diff)
- deficit: OECD general government deficit (as % of GDP)
- elderly: OECD share of the elderly (65+) (as % of total population)
- gdpr: WB growth rate of GDP at market prices (LCU)
- unemprate: WB unemployment, total (% of total labour force)
- cpi: OECD consumer price index
- voteshare: Alesina vote share of the main governing party (or coalition)
- taxrevue: WB tax revenue % GDP
- finref_index: Alesina domestic financial liberalization indicator
- product_index: Alesina product market liberalization indicator
- current_index: Alesina current account reform indicator
- capital_index: Alesina capital account reform indicator
- labor_index: Alesina labor market liberalization indicator
- oilPrice: FRED global price for brent crude
- gasPrice: FRED global price of natural gas
- top1share: WID share of pre-tax national income of top 1%
- top10share: WID share of pre-tax national income of top 10%
- top20share: WID share of pre-tax national income of top 20%
- bot1share: WID share of pre-tax national income of bottom 1%
- bot10share: WID share of pre-tax national income of bottom 10%
- bot20share: WID share of pre-tax national income of bottom 20%
- giniMRKT: OECD market-based Gini coefficient
- giniNET: OECD net Gini coefficient
- socexp: OECD total public social expenditure as percentage of GDP
- almpExp: OECD public social expenditure for ALPM as % of GDP
- unempExp: OECD public social expenditure for unemployment as % of GDP
- famExp: OECD public social expenditure for family as %of GDP
- healthExp: OECD public social expenditure for health as % of GDP
- houseExp: OECD public social expenditure for housing as % of GDP
- incapExp: OECD public social expenditure for incapacity as % of GDP
- oldExp: OECD public social expenditure for old age as % of GDP
- energyShare: OECD Share of value added of dirty energy producing/mining industries ("mining and extraction of energy producing products" (D05T06) and "Coke and refined petroleum products" (D19))
- cenDist: PSU distance from centroid of country to nearest coast (km)
- coastLengthWRI: WRI length of coastline
- coastLengthFB: Factbook length of coastline
- eventsFlood: EMDAT/OWID number of floods (global)
- majorHurric: NOAA/OWID frequency of major hurricanes in the North Atlantic
- peopleAffectedEQ: EMDAT/OWID number of people affected by earthquakes (global)
- urbanCoastLOG: CIESIN (log) share of urban extent in coastal zone (km2) affected by sea level elevation
- eventsWildfire: EMDAT/OWID number of wildfires (global)
- agriLandPC: Nationmaster agricultural land (km2) per capita
- globalization: Gygli et al. index of globalization (total) (global average)
- riot: ICRG index of riot (global average)
- conflict: ICRG index of total conflict (global average)
- logpop: WB (log) population (in million)
- left: Cruz et al. Dummy variable for left-wing chief of executive
- center: Cruz et al. Dummy variable for center-wing chief of executive
- right: Cruz et al. Dummy variable for right-wing chief of executive

### Table 2. Summary statistics (selected variables with exact values)
- D_Popularsupport — Mean: 2.223; Min: 0.500; Max: 3.792; SD: 0.579
- D.EPS — Mean: 0.106; Min: -1.400; Max: 1.112; SD: 0.317
- D.EPS_MKT — Mean: 0.076; Min: -2.000; Max: 2.083; SD: 0.425
- D.EPS_NMKT — Mean: 0.136; Min: -2.000; Max: 2.000; SD: 0.449
- D.TAXES — Mean: 0.013; Min: -1.000; Max: 1.750; SD: 0.280
- D.taxPCA — Mean: 0.054; Min: -1.006; Max: 2.722; SD: 0.335
- D.TAXCO2 — Mean: 0.031; Min: -1.000; Max: 6.000; SD: 0.383
- D.TAXDIESEL — Mean: -0.071; Min: -4.000; Max: 3.000; SD: 0.666
- D.TAXNOX — Mean: 0.058; Min: -1.000; Max: 6.000; SD: 0.513
- D.TAXSOX — Mean: 0.031; Min: -2.000; Max: 6.000; SD: 0.413
- D.STD — Mean: 0.190; Min: 0.000; Max: 3.500; SD: 0.569
- D.limitPCA — Mean: 0.200; Min: 0.000; Max: 3.689; SD: 0.601
- D.ELV_NOX — Mean: 0.220; Min: 0.000; Max: 5.000; SD: 0.906
- D.FIT — Mean: 0.105; Min: -6.000; Max: 6.000; SD: 1.057
- D.TRADESCH_CO2 — Mean: 0.137; Min: -6.000; Max: 6.000; SD: 1.878
- deficit — Mean: -2.582; Min: -32.063; Max: 18.633; SD: 4.613
- elderly — Mean: 14.367; Min: 4.099; Max: 26.648; SD: 4.715
- gdpr — Mean: 2.230; Min: -9.132; Max: 25.163; SD: 2.977
- unemprate — Mean: 8.181; Min: 2.119; Max: 33.473; SD: 5.311
- cpi — Mean: 0.996; Min: 0.476; Max: 2.616; SD: 0.227
- voteshare — Mean: -1.854; Min: -29.860; Max: 27.230; SD: 8.839
- taxrevue — Mean: 18.946; Min: 7.919; Max: 36.500; SD: 6.403
- oilPrice — Mean: 68.758; Min: 24.412; Max: 111.960; SD: 30.665
- gasPrice — Mean: 5.067; Min: 2.614; Max: 8.895; SD: 1.952
- top1share — Mean: 0.117; Min: 0.045; Max: 0.296; SD: 0.049
- top10share — Mean: 0.354; Min: 0.229; Max: 0.651; SD: 0.084
- giniMRKT — Mean: 47.337; Min: 32.917; Max: 69.400; SD: 5.679
- socexp — Mean: 20.742; Min: 4.793; Max: 32.213; SD: 5.698
- coastLengthWRI — Mean: 28511.000; Min: 0.000; Max: 265523.000; SD: 53217.010
- peopleAffectedEQ — Mean: 6370539; Min: 786413; Max: 46724390; SD: 11012820
- eventsWildfire — Mean: 10.600; Min: 4.000; Max: 22.000; SD: 4.791
- agriLandPC — Mean: 12.561; Min: 0.365; Max: 208.211; SD: 36.766
- globaliz — Mean: 58.290; Min: 53.319; Max: 61.753; SD: 2.899
- conflict — Mean: 834.641; Min: 179.419; Max: 2891.194; SD: 822.479
- logpop — Mean: 3.268; Min: 0.688; Max: 7.152; SD: 1.470
- left — Mean: 0.431; Min: 0; Max: 1; SD: 0.496
- center — Mean: 0.093; Min: 0; Max: 1; SD: 0.29
- right — Mean: 0.476; Min: 0; Max: 1; SD: 0.5

### Table 3. The effect of EPS changes on popular support — regression coefficients and statistics (exact values)
- Dependent variable: popular support
- Key coefficient on D.EPS across specifications (standard errors in parentheses):
  - (I) D.EPS: -0.231*** (0.079)
  - (II) D.EPS: -0.246*** (0.079)
  - (III) D.EPS: -0.247*** (0.080)
  - (IV) D.EPS: -0.266*** (0.086)
  - (V) D.EPS: -0.295*** (0.084)
  - (VI) D.EPS: -0.284*** (0.088)
- Other covariates (coefficients with standard errors in parentheses, by specification where shown):
  - deficit:
    - (I) 0.039*** (0.014)
    - (II) 0.028* (0.015)
    - (III) 0.027* (0.015)
    - (IV) 0.025 (0.017)
    - (V) 0.028* (0.014)
    - (VI) 0.029** (0.015)
  - elderly:
    - (I) 0.163*** (0.051)
    - (II) 0.177*** (0.055)
    - (III) 0.194*** (0.063)
    - (IV) 0.172** (0.069)
    - (V) 0.196*** (0.074)
    - (VI) 0.149** (0.066)
  - gdpr:
    - (I) 0.063** (0.029)
    - (II) 0.047 (0.030)
    - (III) 0.051 (0.031)
    - (IV) 0.048 (0.033)
    - (V) 0.041 (0.036)
    - (VI) 0.031 (0.030)
  - unemprate:
    - included in specifications with coefficients:
      - -0.035** (0.017)
      - -0.033* (0.018)
      - -0.028 (0.022)
      - -0.032 (0.022)
      - -0.034 (0.025)
  - cpi:
    - coefficients reported: 0.299 (0.309); 0.233 (0.314); 0.084 (0.299); -0.807* (0.457)
  - voteshare:
    - coefficients reported: 0.010* (0.005); 0.009* (0.005); 0.007 (0.004)
  - taxrevenue:
    - coefficients reported: -0.006 (0.047); -0.007 (0.045)
  - finref_index: 0.153 (1.620)
  - product_index: -0.680 (0.710)
  - current_index: 1.987 (1.316)
  - capital_index: 0.706 (1.170)
  - labor_index: 2.260 (1.998)
- Constants by specification:
  - (I) 0.659 (0.624)
  - (II) 0.724 (0.673)
  - (III) 0.281 (0.924)
  - (IV) 0.628 (1.010)
  - (V) 0.628 (1.508)
  - (VI) -1.750 (1.731)
- Fixed effects:
  - Country FE: Yes (all specs)
  - Time FE: Yes (all specs)
- Model fit and sample size:
  - R2:
    - (I) 0.455
    - (II) 0.466
    - (III) 0.468
    - (IV) 0.464
    - (V) 0.493
    - (VI) 0.512
  - Observations:
    - (I) 370
    - (II) 370
    - (III) 370
    - (IV) 326
    - (V) 321
    - (VI) 321
- Note (as in source): The outcome variable is popular support. Estimated based on 푦푖,푡 = 훽Δ퐸푃푆푖,푡 + 휃푋푖,푡 + 훼푖 + 훾푡 + 휖푖,푡. A coefficient of -

*Italic — Source: wpiea2021156-print-pdf - REFERENCES*

### 0.2 is equivalent to a 10 percent decline in popular support from an increase in EPS from the 1st to the 3rd quartile of

### wpiea2021156-print-pdf - 0.2 is equivalent to a 10 percent decline in popular support from an increase in EPS from the 1st to the 3rd quartile of

### Key finding: EPS changes reduce popular support
- Baseline and multiple specifications report negative and often significant effects of ΔEPS on popular support.
- A coefficient of -0.2 is equivalent to a 10 percent decline in popular support from an increase in EPS from the 1st to the 3rd quartile of the EPS distribution.
- Standard deviations are based on robust standard errors clustered at the country level. ***,**,* denote significance at 1, 5 and 10 percent, respectively.

### Market-based EPS changes (Table 4)
- Market-based EPS (D.EPS_MKT) effect: -0.153** (0.074) in one specification (Table 4, column I).
- Other specific policy-related coefficients reported (selected):
  - D.TAXES: -0.199** (0.092)
  - D.taxPCA: -0.202*** (0.062)
  - D.TAXCO2: -0.071** (0.035)
  - D.TAXDIESEL: 0.019 (0.066)
  - D.TAXNOX: -0.090** (0.042)
  - D.TAXSOX: -0.196*** (0.052)
- Control variables (consistent positive associations with popular support):
  - deficit ~ 0.038–0.040*** (0.013–0.014)
  - elderly ~ 0.150–0.166*** (0.050–0.053)
  - gdpr ~ 0.058–0.064** (0.027–0.028)
- R^2 values across Table 4 specifications: 0.439–0.452
- Observations: typically 370–378 across specifications.

### Non-market-based EPS changes (Table 5)
- D.EPS_NMKT: -0.066 (0.048) (not always significant)
- Specific non-market policy coefficients (selected):
  - D.ELV_DIESELSO: -0.206** (0.103)
  - D.RD_RE: -0.050* (0.027) (appears twice)
- Controls similar to market-based specifications:
  - deficit ~ 0.039*** (0.013–0.014)
  - elderly ~ 0.145–0.150*** (0.050–0.052)
  - gdpr ~ 0.057–0.062** (0.025–0.027)
- R^2 values: 0.441–0.450
- Observations: 378 in all columns.

### Mediation by fuel prices, dirty energy share, and initial EPS (Table 6, Table 12 series)
- Interaction effects highlight heterogeneity in ΔEPS impacts:
  - D.EPS x high_oilPrice: -0.320** (0.129) (Table 6, column II)
  - D.EPS x high_gasPrice: -0.493*** (0.170) (Table 6, column II)
  - D.EPS x high_energyShare: -0.264** (0.119) (Table 6, column IV)
  - D.EPS x high_initialEPS: -0.321*** (0.076) (Table 6, column IV)
- IV (peopleAffectEQ_urbanCoastLOG) mediated IV estimates (Table 12-1, 12-2):
  - D.EPS_high_oilPrice: -1.241** (0.608) (Table 12-1, column II; 1st-stage F-Stat 12.161)
  - D.EPS_high_gasPrice: -1.184* (0.630) (Table 12-1, column IV; 1st-stage F-Stat 12.82)
  - D.EPS_high_energyShare: -3.675** (1.769) (Table 12-2, column II; 1st-stage F-Stat 9.85)
  - D.EPS_high_initialEPS: -1.343*** (0.513) (Table 12-2, column IV; 1st-stage F-Stat 25.166)
- Observations for mediated IV specs: typically 366–370.

### Mediation by income shares and inequality (Table 7, Table 12-3 to 12-5)
- ΔEPS effects larger (more negative) when top income shares are high:
  - D.EPS x high_top1share: -0.300* (0.156) (Table 7, column II)
  - D.EPS x high_top10share: -0.320** (0.155) (Table 7, column IV)
  - D.EPS x high_top20share: -0.320** (0.146) (Table 7, column VI)
- IV heterogeneous results (Table 12-3):
  - D.EPS_high_top1share: -3.495*** (0.561)
  - D.EPS_high_top10share: -4.482*** (0.862)
  - D.EPS_high_top20share: -5.500*** (1.546)
  - 1st-stage F-Stats reported range widely; e.g., 73.36, 30.61, 17.65 in some columns.
- Inequality interactions (Table 7 cont. and Table 12-5):
  - D.EPS x high_giniMRKT: -0.380** (0.170) (Table 7 cont., column II)
  - IV: D.EPS_high_giniMRKT: -4.569*** (1.757) (Table 12-5, column II)
  - D.EPS x high_giniNET: -0.402** (0.162) (Table 7 cont., column IV)
  - IV: D.EPS_high_giniNET: -3.343*** (0.852) (Table 12-5, column IV)

### Mediation by social expenditure types (Table A3, Table 12-6)
- Low social expenditure amplifies negative ΔEPS effect:
  - D.EPS x low_socExp: -0.504*** (0.167) (Table 7 cont., column III)
  - D.EPS x low_almpExp: -0.483*** (0.158) (Table 7 cont., column V)
  - D.EPS x low_unempExp: -0.544*** (0.170) (Table 7 cont., column V)
- IV results (Table 12-6):
  - D.EPS_low_socexp: -1.492*** (0.523) (1st-stage F-Stat 44.32)
  - D.EPS_low_almpExp: -1.863* (1.037) (1st-stage F-Stat 76.26)
  - D.EPS_low_unempExp: -1.860** (0.834) (1st-stage F-Stat 36.28)

### Instrumental variable (IV) estimation and validity (Tables 8–11, 12 series)
- IV estimates of D.EPS (Table 8):
  - D.EPS ranges from -0.635* to -1.047** across IV specifications:
    - (I) -0.994*** (0.386)
    - (II) -0.741** (0.361)
    - (III) -0.635* (0.341)
    - (IV) -1.047** (0.533)
    - (V) -0.879** (0.377)
  - 1st-stage coefficients and F-Stats reported; examples:
    - 1st-stage coef. 0.003*** (0.001), 1st-stage F-Stat 30.79 (Table 8, column I)
    - 1st-stage coef. -7.552*** (1.165), 1st-stage F-Stat 42.05 (Table 8, column III)
- Validity checks (Table 9, Table 10):
  - Direct effect of instruments on popular support: D.EPS estimated -0.207 to -0.240*** (0.078–0.081) when instruments are included as controls (Table 9).
  - Correlation of instruments with baseline residuals is effectively zero (Table 10): coefficients reported as 0.000 or -0.000 with small standard errors.
- Robustness of IV with additional controls (Table 11):
  - D.EPS IV coefficients remain negative and significant under added controls:
    - Example: -1.555*** (0.433) in one specification.
  - 1st-stage F-Stats reported: 18.38–29.45 across specifications.

### Effect on incumbent vote share (Tables A2, A4)
- OLS and panel estimates (Table A2):
  - D.EPS on vote share: negative and significant in several specifications:
    - Examples: -6.908* (3.556), -9.042*** (2.350), -12.341*** (2.833)
  - L.voteshare highly negative and significant: e.g., -0.391*** (0.086) to -0.669*** (0.082)
  - R^2 range: 0.421–0.701; Observations vary from 80 to 109.
- IV estimates on vote share (Table A4) are less precise:
  - D.EPS IV coefficients range (not always significant): e.g., -6.944 (11.215), -15.049* (8.404)
  - 1st-stage statistics: examples include 1st-stage coef. 0.0027** (0.001), 1st-stage F-Stat 5.416 (column I); other first-stage F-Stats vary (e.g., 32.865).

### Heterogeneity by ideology (Table A5)
- Interaction of ΔEPS with party ideology:
  - D.EPS*left: -0.420** (0.185) to -0.498** (0.210) across specs
  - D.EPS*center: -0.190 to -0.265** (0.119–0.121) with several specifications significant
  - D.EPS*right: positive but not significant in reported columns (e.g., 0.175 (0.190))
- Controls and model fit:
  - deficit coefficients vary (e.g., 0.036** (0.016) to 0.027** (0.013))
  - R^2 range: 0.466–0.535
  - Observations: 289–305 depending on specification.

### Model notes and estimation details
- Main estimating equation for popular support:
  - 푦_{i,t} = 훽 ΔEPS_{i,t} + 휃 X_{i,t} + α_i + γ_t + ε_{i,t}
  - For heterogeneous specifications a switching function is used: F(z_{i,t}) = e^{−1.5*z_{i,t}} / (1 + e^{−1.5*z_{i,t}}) with separate β^L and β^H regimes.
- Standard errors: robust and clustered at the country level unless otherwise noted (e.g., Driscoll-Kraay standard errors with two lags for vote share).
- Sample sizes vary by specification and available covariates: common observation counts include 370, 378, 334, 301, 326, 378, 301, 259, 109, 80, 378.

*Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021156-print-pdf.pdf*

### 0.2 is equivalent to a 10 percent decline in popular support from an increase in EPS from the 1st to the 3rd quartile of

### 0.2 is equivalent to a 10 percent decline in popular support from an increase in EPS from the 1st to the 3rd quartile of

### Main interpretation
- A coefficient value of 0.2 is interpreted as equivalent to a 10 percent decline in popular support resulting from an increase in EPS from the 1st to the 3rd quartile of the EPS distribution.

### Statistical details
- Standard deviations are based on robust standard errors clustered at the country level (values shown in parentheses).
- Significance indicators in the table/text are denoted as ***, **, * corresponding to significance at 1, 5 and 10 percent, respectively.

*Source: wpiea2021156-print-pdf*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021156-print-pdf.pdf_