## wpiea2022009-print-pdf

## Source details

**Canonical URL:** [wpiea2022009-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022009-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022009-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022009-print-pdf.pdf.json)

---

### I. Introduction — context and objectives
- Climate danger recognized globally; gap remains between actions and what is needed to reach net-zero emissions by 2050 and to limit global temperature increase to 1.5 °C.
- Key premise: gross carbon emissions must fall substantially and rapidly to meet net-zero goals.
- Objective: identify and analyze structural breaks in carbon emission patterns for the top 20 carbon emitters using machine-learning indicator saturation methods, and interpret drivers via the Kaya Identity.
- Top 20 emitters (in descending cumulative CO2 up to 2019): USA, CHN, RUS, DEU, GBR, JPN, IND, FRA, CAN, UKR, POL, ITA, ZAF, MEX, IRN, AUS, KOR, BRA, SAU, ESP — these account for 80 percent of cumulative global CO2 emissions as of 2019.
- Note on sinks and scale: natural sinks remove between 9.5 and 11 Gt of CO2 per year; annual global CO2 emissions reached 38.0 Gt in 2019; no artificial sinks currently able to remove carbon at the necessary scale.

### II. Contribution and analytical framework
- Empirical contribution:
  - Show that policies indirectly reshaping economic structure can induce long-lasting downward trend shifts in emissions.
  - Existing climate policies appear far from sufficient for reducing absolute emissions (panel analysis in appendix supports this).
- Methodological contribution:
  - Apply Oxford University’s machine-learning indicator saturation approach (implemented via an R package) — step indicator saturation (SIS) and trend indicator saturation (TIS) with general-to-specific (GETS) block search — to identify trend changes and structural breaks across twenty countries.
  - Interpret changes using the Kaya Identity (population, GDP per capita, energy intensity, carbon intensity).
- Analytical aim: detect long-lasting (rather than transitory) downward trend shifts and relate timing to economic, political, or environmental events.

### III. Data and methodology
- Data sources and sample:
  - Gross CO2 emissions from the Global Carbon Project.
  - PPP GDP per capita from the World Bank.
  - Coal and oil consumption from BP’s Statistical Review of World Energy (2020) compiled by Our World in Data.
  - Annual data from 1965 to 2019 for the top 20 cumulative emitters.
- Variable treatment:
  - Logs of level variables used.
  - Focus on CO2 (highly correlated with GHG) and gross emissions (net emissions not used due to lack of accurate estimates).
- Indicator-saturation approach:
  - SIS and TIS used; impulse indicators dropped to focus on structural changes in level and trend.
  - Model forms:
    - Unconditional: yt = α + sum(ψi D_Si,t + ωi D_Ti,t) + εt
    - Conditional: yt = α + β zt + sum(ψi D_Si,t + ωi D_Ti,t) + εt, where zt may include GDP (main specification controls for GDP only).
  - Statistical significance threshold: α = 0.05 (main results); lower levels experimented with (1 percent) produce fewer break selections.

### IV. Interpretation of indicators and residuals
- Indicator roles:
  - Negative trend indicator (TIS) → decreasing pace of emissions or accelerating carbon reductions.
  - Negative step indicator (SIS) → downward level shift in emissions (permanent lower level).
- “Indicator coefficient path” referred to as “residual carbon emissions” (emissions over and above those induced by controls, e.g., GDP).

### V. Key empirical findings (unconditional analysis)
- Frequency of downward trend shifts:
  - Since the 1990s, downward trend shifts in carbon emissions have been rare: only 16 downward trend shifts in total among the top 20 emitters.
- Emerging markets (examples):
  - China, India, South Africa: none witnessed structural breaks in trends in recent decades; all three experienced positive level shifts in emissions.
  - China’s latest structural break in carbon-emission trend occurred in 1972 (a notable downward shift then).
- Advanced economies (examples):
  - Rapid declines mainly reflect downward shifts in levels rather than accelerated downward trends, except:
    - United Kingdom: trend break in 2008.
    - United States: trend break in 2008.
    - Germany: trend break in 2018.
  - Most other structural breaks for AEs are level shifts rather than trend accelerations.
- Model fit:
  - Figures for six selected countries and appendix figures for remaining 14 show the model fits the data well.

### VI. Interpretation and broader implications of structural breaks
- Many structural breaks associated with non-climate structural factors or pollution responses:
  - Example: German reunification (October 1990) produced non-climate structural changes affecting emissions.
  - Example: China’s evaluation-system change for government officials after a severe air pollution crisis in January 2013.
- Implication: addressing climate change likely requires policies that directly mitigate emissions and policies that reshape economic structure (indirect mitigation).
- Carbon taxation:
  - Highlighted as a promising transformative policy in the literature.
  - Current carbon tax rates are low among most of the top 20 emitters and politically constrained.
  - Historical structural breaks often coincide with crises or major political events rather than steady policy rollouts.
- Green finance and ESG:
  - Literature finds limited impact of certain green finance instruments on emissions; economy-wide policies remain essential.

### VII. Kaya Identity decomposition insights
- Kaya Identity expression used: CO2 = Population × (GDP/Population) × (Energy/GDP) × (CO2/Energy).
- Logarithmic form used in analysis: Ln(CO2) = Ln(GDP) + Ln(Energy/GDP) + Ln(CO2/Energy).
- Observations from decomposition (six-country illustrations):
  - US: CO2 sudden drop in 2008 coincides with contraction of GDP per capita and a shift in carbon intensity starting around the same time.
  - US policy link: following the Energy Independence and Security Act (December 2007), natural gas production from shale increased and natural gas price fell from $16.75 per gallon in June 2008 to $3.12 in August 2009, contributing to a fuel switch from coal/oil to gas and CO2 reductions.
  - Energy intensity trends:
    - For UK, US, and Germany, the energy-intensity contribution to CO2 has become increasingly negative over time since 1965.
    - China followed a similar pattern after the 1980s.
    - India and South Africa: energy-intensity contributions have remained broadly positive in recent decades.

### VIII. Conditional analysis (controlling for GDP) — patterns and interpretation
- Conditional model: Ln(CO2) = α + β Ln(GDP)t + ∑(ψi Di,tS + ωi Di,tT)Ti=1 + εt, allowing β ≠ 1 so structural breaks can capture policy effects on emissions independent of GDP effects.
- Key conditional findings:
  - Conditional emission patterns often differ from unconditional patterns:
    - China and the United States exhibited broadly increasing unconditional emissions before 2008, while emissions conditional on GDP have been declining for both countries — signaling mitigating forces outside the GDP channel.
  - Some unconditional breaks disappear and new conditional breaks appear:
    - The 2008 unconditional trend breaks for the UK and the US are not present when conditioning on GDP, indicating those breaks were driven by GDP contraction during the Global Financial Crisis.
    - Germany’s conditional emissions display a downward trend shift in 1991 (likely reflecting German reunification in October 1990).
    - China’s conditional emissions display a downward trend shift in 1978 (coinciding with economic reforms and opening-up) and another downward shift in 2012 (linked to pollution crisis responses in 2013).
- Statistical significance:
  - Main results reported at 5 percent significance; 1 percent results in appendices are similar but show fewer structural breaks.

### IX. Panel data analysis: Climate Policy Intensity Index (CPII) and carbon-growth regressions
- Policy data and CPII construction:
  - Climate Policy Database raw coverage: 1927 to 2021 with 4,980 policy measure occurrences; 1,458 distinct policy types.
  - Jurisdiction breakdown (full sample): Country = 4,383 (92.90 percent); Subnational region = 266 (5.64 percent); City = 44 (0.93 percent); Supranational region = 25 (0.53 percent); Total = 4,718 (100.00 percent).
  - CPII scoring rules (preserved exactly):
    - Initial score for each policy occurrence: 1.
    - If jurisdiction = “Subnational region” or “City” and sector = “General”, set score to 0.5.
    - If jurisdiction = “Subnational region” or “City” and sector is a specific sector (e.g., energy), set score to 0.25.
    - If Impact indicator = “High”, double the score.
    - Supranational measures receive same weight as country-level measures.
    - Take cumulative sum of scores from 1965 to current year (assumes measures remain in place perpetually).
  - Key caveat: CPII mostly measures quantity rather than quality of policies; weights are judgment-based.
- Cleaned sample and dependent/independent variables:
  - Final cleaned sample: 18,335 observations.
  - CPII descriptive statistics (cleaned sample): mean = 3.2; standard deviation = 18.6.
  - Dependent variable: annualized three-year cumulative growth rate of carbon emissions (CO2 Growth).
  - Independent variables (annualized three-year cumulative growth rates/changes): CPII_3YrChange; GDP_3YrCumGrowth; Coal_3YrCumGrowth; Oil_3YrCumGrowth; kgov_3YrCumGrowth; kpriv_3YrCumGrowth.
- Summary statistics for top 20 emitters (annualized three-year cumulative):
  - year: Obs = 1,060; Min = 1965; Max = 2019
  - CO2 Growth: Obs = 1,000; Mean = 4.2; Std. Dev. = 3.1; Min = 0.5; Max = 24.3
  - CPII: Obs = 1,060; Mean = 35.4; Std. Dev. = 64.3; Min = 0.0; Max = 424.0
  - CPII Change: Obs = 1,000; Mean = 2.8; Std. Dev. = 4.5; Min = 0.0; Max = 41.2
  - GDP Growth: Obs = 961; Mean = 3.8; Std. Dev. = 4.4; Min = -14.8; Max = 45.0
  - Coal Growth: Obs = 964; Mean = 2.1; Std. Dev. = 8.3; Min = -23.3; Max = 97.9
  - Oil Growth: Obs = 1,000; Mean = 3.0; Std. Dev. = 7.2; Min = -21.9; Max = 94.6
  - Government Capital Growth: Obs = 910; Mean = 3.6; Std. Dev. = 3.3; Min = -2.1; Max = 27.8
  - Private Capital Growth: Obs = 910; Mean = 4.2; Std. Dev. = 4.2; Min = -3.2; Max = 28.6

### X. Panel regression results — main estimates and economic interpretation
- Main empirical result:
  - Three-year cumulative change in CPII is negatively correlated with three-year cumulative growth rate of carbon emissions; GDP, coal, oil, government capital, and private capital growth are positively correlated with carbon growth.
- Key coefficient estimates for CPII_3YrChange (Appendix Table 6; p-values in parentheses):
  - Column (1) FE: -0.134*** (0.000)
  - Column (2) RE: -0.135*** (0.000)
  - Column (3) FE with GDP: -0.106*** (0.000)
  - Column (4) RE with GDP: -0.107*** (0.000)
  - Column (5) FE with GDP Coal Oil: -0.088*** (0.000)
  - Column (6) RE with GDP Coal Oil: -0.087*** (0.000)
  - Column (7) FE with GDP Coal Oil Capital Capital: -0.080*** (0.000)
  - Column (8) RE with GDP Coal Oil Capital Capital: -0.079*** (0.000)
- Selected GDP and capital coefficients (examples preserved from tables):
  - GDP_3YrCumGrowth: positive and significant where included (e.g., 0.219***, 0.224***, 0.046***, 0.050***, 0.027*, 0.028** with respective p-values).
  - kgov_3YrCumGrowth: 0.115*** and 0.119*** where included.
  - kpriv_3YrCumGrowth: 0.062*** and 0.065*** where included.
- Constants (examples):
  - Column (1): Constant = 4.613*** (0.000)
  - Column (7): Constant = 3.198*** (0.000)
- Model diagnostics:
  - Observations range: 1,000 to 838 depending on specification.
  - Number of countries: 20 in most columns; Columns (7) and (8) have 19 countries (capital stock data for Australia unavailable).
  - Hausman Test P-values: 0.410, 0.000, 0.000, 0.005 — fixed-effect model preferred in three out of four cases at the 5 percent significance level.
- Economic importance (implied policy scale to induce negative carbon-growth):
  - Based on Column (1): required annualized CPII change = 34.4 (i.e., roughly 34.4 new climate policies per year).
    - Derivation: in absence of new policies (CPII change = 0) carbon emissions grow at 4.613 percent; given marginal impact of -0.134, required change = 4.613 / 0.134 = 34.4.
    - Comparison: maximum observed CPII change in top 20 emitters = 41.2; average observed CPII change = 2.8.
  - Based on Column (7) (preferred model): roughly 53.7 new climate policies per year required for a “typical” country in a “typical” year among the 19 emitters in that regression, much larger than average observed CPII change of 2.8.
- Overall interpretation:
  - Climate policies implemented to date in top 20 emitters have been effective in slowing carbon emission growth rates.
  - Existing policies appear insufficient to reduce the absolute level of carbon emissions across most top 20 emitters.
- Important caveat: CPII does not fully capture policy quality or cross-country comparability; panel regression results should be interpreted with caution and complemented by country-specific analysis.

### XI. Robustness checks and alternative specifications
- Alternative policy measure:
  - Unweighted/raw count of policy instruments (CumCount_3YrChange) yields very similar results (Appendix Table 7).
- Time horizons:
  - Current-year changes, three-year cumulative changes, and five-year cumulative changes tested; results consistent.
  - Magnitudes of policy impacts increase with horizon:
    - Current-year changes: 0.059 (absolute magnitude example from Appendix Table 8 Column (7))
    - Three-year cumulative changes: 0.080 (example from Appendix Table 6 Column (7))
    - Five-year cumulative changes: 0.092 (example from Appendix Table 9 Column (7))
  - Interpretation: climate policies take time to affect carbon emissions.
- Lag specification:
  - Three-year lagged policy-change specifications (only policy change variables lagged) produce similar signs and magnitudes (Appendix Table 10).

### XII. Appendix empirical observations and country illustrations
- Timing and drivers of structural breaks (Appendix Table 11 and 12):
  - Downward trend shifts around 1980 in some countries, especially AEs, likely reflect lower oil consumption amid the oil crisis during the late 1970s and early 1980s.
  - Major climate initiatives (Kyoto Protocol adopted December 1997, entered into force February 2005; Paris Agreement adopted December 2015, entered into force November 2016) were not followed by statistically significant downward trend shifts in carbon emissions among top emitters.
  - Major economic crises are strongly associated with structural breaks; the 2008 global financial crisis induced downward emission paths for top AE emitters but not for major EM emitters.
  - Carbon taxes: among six emitters examined, only the United Kingdom implemented nationwide carbon taxes during the sample period (carbon price support in 2013); Table 1 identifies two level shifts for the United Kingdom in 2014 and 2016. Of 14 other breaks among the top 20 emitters in Appendix Table 11, only Ukraine displayed a level shift in 2014 after introducing a carbon tax in 2011.
- Germany detailed example:
  - Gross CO2 emission level reduced from around 1.1 Gt (20.8 log tons) in 1980 to 0.7 Gt (20.4 log tons) in 2019, or a 33 percent reduction.
  - Per-year reduction from 1980 to 2019 is 10,207,804 tons; based on this rate:
    - It would take 34 years to reduce half (350,977,554 tons) of the 2019 emission level.
    - At that rate, Germany would reduce gross CO2 emission to half of its 2019 level only by 2053 and to zero only by 2088.
  - Note: reaching net-zero does not require gross emissions to be zero due to absorption by natural sinks, but natural sinks can re-release carbon (e.g., forest fires), so cutting gross emissions remains important.

### XIII. Conclusions and policy implications
- Major global initiatives (Kyoto Protocol in 2005, Paris Agreement in 2015-2016) and traditional climate policies implemented so far do not seem to have reduced trends in either unconditional or GDP-conditional carbon emissions among the top emitters.
- Case studies indicate economic structural changes and strong political responses to pollution crises have played significant roles in some downward structural breaks.
- Existing climate policies implemented so far appear insufficient to reduce absolute levels of carbon emissions; bolder, more transformative climate policies are urgently needed to:
  - arrest rising trends;
  - reduce absolute levels; and
  - aim for carbon neutrality by 2050.
- Policy design suggestions:
  - Consider broader economic transitions (changes in economic structure) alongside traditional climate and carbon policies.
  - Explore policy areas such as green agri-food systems and international carbon price floors as potential complements.
  - Recognize non-climate policies that reshape economic structure can be critical for emission outcomes.

*Source: wpiea2022009-print-pdf - Appendix and main paper materials*

### References.....................................................................................................44

### wpiea2022009-print-pdf - References.....................................................................................................44

### I. Introduction — context and objectives
- Climate danger recognized globally; gap remains between actions and what is needed to reach net-zero emissions by 2050 and to limit global temperature increase to 1.5 °C.
- Key premise: gross carbon emissions must fall substantially and rapidly to meet net-zero goals.
- Paper objective: identify and analyze structural breaks in carbon emission patterns for the top 20 carbon emitters using machine-learning indicator saturation methods, and interpret drivers via the Kaya Identity.
- Focus: top 20 emitters (in descending cumulative CO2 up to 2019): USA, CHN, RUS, DEU, GBR, JPN, IND, FRA, CAN, UKR, POL, ITA, ZAF, MEX, IRN, AUS, KOR, BRA, SAU, ESP. These account for 80 percent of cumulative global CO2 emissions as of 2019.
- Note on sinks: natural sinks remove between 9.5 and 11 Gt of CO2 per year; annual global CO2 emissions reached 38.0 Gt in 2019; no artificial sinks currently able to remove carbon at the necessary scale.

### II. Contribution and analytical framework
- Twofold contribution:
  - Empirical: show that policies indirectly reshaping economic structure can induce long-lasting downward trend shifts in emissions; existing climate policies appear far from sufficient for reducing absolute emissions (panel analysis in appendix).
  - Methodological: apply Oxford University’s machine-learning indicator saturation approach (implemented via an R package) to identify trend changes and structural breaks across twenty countries; interpret changes using the Kaya Identity (population, GDP per capita, energy intensity, carbon intensity).
- Analytical aim: detect long-lasting (rather than transitory) downward trend shifts and relate their timing to economic, political, or environmental events.

### III. Data and methodology
- Data sources: gross CO2 emissions from the Global Carbon Project; PPP GDP per capita from the World Bank; coal and oil consumption from BP’s Statistical Review of World Energy (2020) compiled by Our World in Data.
- Sample period: annual data from 1965 to 2019 for the top 20 cumulative emitters.
- Variable transformation: logs of level variables used.
- Emissions focus: CO2 (highly correlated with GHG) and gross emissions (net emissions not used due to lack of accurate estimates).
- Machine-learning approach: indicator saturation (step indicator saturation (SIS) and trend indicator saturation (TIS)) with general-to-specific (GETS) block search to select significant indicators; impulse indicators dropped to focus on structural changes in level and trend.
- Model forms:
  - Unconditional: yt = α + sum(ψi D_Si,t + ωi D_Ti,t) + εt
  - Conditional: yt = α + β zt + sum(ψi D_Si,t + ωi D_Ti,t) + εt, where zt may include GDP (main specification controls for GDP only).
- Statistical significance threshold: α = 0.05 (experimented with lower levels producing fewer break selections; main results err on generosity to examine associations with climate policies).

### IV. Interpretation of indicators
- Indicator roles:
  - Negative trend indicator (TIS) → decreasing pace of emissions or accelerating carbon reductions.
  - Negative step indicator (SIS) → downward level shift in emissions (permanent lower level).
- “Indicator coefficient path” referred to as “residual carbon emissions” (emissions over and above those induced by controls, e.g., GDP).

### V. Key empirical findings (unconditional analysis)
- Since the 1990s, downward trend shifts in carbon emissions have been rare: only 16 downward trend shifts in total among the top 20 emitters (consistent with panel analysis in Appendix 1).
- Emerging markets (examples presented): China, India, South Africa
  - None of these three EMs witnessed structural breaks in trends, implying the pace of emissions remained relatively unchanged in past decades.
  - All three EMs experienced positive level shifts in emissions (permanently higher after those shifts).
  - Example: latest structural break in China’s carbon emission trend occurred in 1972 (trend experienced a notable downward shift).
- Advanced economies (examples presented): United Kingdom, United States, Germany
  - Rapid declines mainly reflect downward shifts in levels rather than accelerated downward trends, except:
    - United Kingdom: trend break in 2008.
    - United States: trend break in 2008.
    - Germany: trend break in 2018.
  - Most other structural breaks for AEs are level shifts rather than trend accelerations.
- Model fit: Figures for six selected countries and Appendix figures for remaining 14 show the model fits the data well.

### VI. Interpretation and broader implications
- Many identified structural breaks are associated with non-climate structural factors (e.g., economic structural change after German reunification in October 1990) or policy responses to pollution shocks (e.g., China’s change in evaluation system for government officials after a severe air pollution crisis in January 2013).
- Implication: addressing climate change likely requires policies that directly mitigate emissions and policies that reshape economic structure (indirect mitigation).
- Carbon taxation highlighted as a promising transformative policy from the literature, but current carbon tax rates are low among most of the top 20 emitters and politically constrained; historical structural breaks often coincide with crises or major political events rather than steady policy rollouts.
- Green finance and ESG evidence: literature finds limited impact of certain green finance instruments on emissions; economy-wide policies remain essential.

### VII. Methodological remarks and extensions
- The machine-learning indicator saturation approach is agnostic about break dates and suitable for highly nonstationary climate data.
- Approach complements panel-data regressions; both are employed in the paper (panel results reported in appendices).
- The framework can be extended to a web-based toolbox to analyze emission patterns and simulate climate/economic policies for broader country sets (follow-up project underway).

*Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022009-print-pdf.pdf*

### Appendix Table 11. Note that the emissions in some countries, especially in AEs, experienced

### Appendix Table 11. Note that the emissions in some countries, especially in AEs, experienced

### Key empirical observations on structural breaks and trends
- Downward trend shifts around 1980 in some countries, especially in AEs, likely reflect lower oil consumption amid the oil crisis during the late 1970s and early 1980s.
- Downward trend shifts do not seem to occur following major climate initiatives: the Kyoto Protocol (adopted December 1997, entered into force February 2005) and the Paris Agreement (adopted December 2015, entered into force November 2016) were not followed by statistically significant downward trend shifts in carbon emissions among top emitters.
- Major economic crises are strongly associated with structural breaks in emissions; the effect can differ across countries (for example, the 2008 global financial crisis induced downward emission paths for top AE emitters but not for major EM emitters).
- The effect of carbon taxes has yet to show up as major structural breaks in large emitters: among six emitters examined, only the United Kingdom implemented nationwide carbon taxes during the sample period (carbon price support in 2013); Table 1 identifies two level shifts for the United Kingdom in 2014 and 2016. Of the 14 other breaks among the top 20 emitters in Appendix Table 11, only Ukraine displayed a level shift in 2014 after introducing a carbon tax in 2011.

### Germany: a detailed example of pace and implications of reductions
- Gross CO2 emission level reduced from around 1.1 Gt (20.8 log tons) in 1980 to 0.7 Gt (20.4 log tons) in 2019, or a 33 percent reduction.
- Per-year reduction from 1980 to 2019 is 10,207,804 tons; based on this rate:
  - It would take 34 years to reduce half (350,977,554 tons) of the 2019 emission level.
  - At that rate, Germany would reduce gross CO2 emission to half of its 2019 level only by 2053 and to zero only by 2088.
- Note: reaching net-zero does not require gross emissions to be zero due to absorption by natural sinks, but natural sinks can re-release carbon (e.g., through forest fires), so cutting gross emissions remains important.

### Kaya Identity decomposition insights
- Kaya Identity expressed as CO2 = Population × (GDP/Population) × (Energy/GDP) × (CO2/Energy).
- Logarithmic form used in analysis: Ln(CO2) = Ln(GDP) + Ln(Energy/GDP) + Ln(CO2/Energy).
- Observations from Kaya decomposition for six countries:
  - US CO2 sudden drop in 2008 coincides with contraction of GDP per capita and a shift in carbon intensity starting around the same time.
  - Following the Energy Independence and Security Act (December 2007), natural gas production from shale increased and natural gas price fell from $16.75 per gallon in June 2008 to $3.12 in August 2009, contributing to a fuel switch from coal/oil to gas and CO2 reductions.
  - For the UK, the US, and Germany, the energy intensity contribution to CO2 has become increasingly negative over time since 1965 (reflecting structural shifts away from energy-intensive industries). China followed a similar pattern after the 1980s. For India and South Africa, energy intensity contributions have remained broadly positive in recent decades.

### Conditional analysis (controlling for GDP): patterns and interpretation
- Equation used: Ln(CO2) = α + β Ln(GDP)t + ∑(ψi Di,tS + ωi Di,tT)Ti=1 + εt, allowing β ≠ 1 so structural breaks can capture policy effects on emissions independent of GDP effects.
- Main conditional-analysis findings:
  - Conditional emission patterns often differ from unconditional patterns: for example, China and the United States exhibited broadly increasing unconditional emissions before 2008, while emissions conditional on GDP have been declining for both countries—signaling mitigating forces outside the GDP channel.
  - Some structural breaks in unconditional analysis disappear and new breaks appear in conditional analysis, highlighting the role of economic development in driving emission changes and isolating non-GDP drivers (policies, political events).
  - Specific examples:
    - The 2008 unconditional trend breaks for the UK and the US are not present when conditioning on GDP, indicating those breaks were driven by GDP contraction during the Global Financial Crisis rather than other factors.
    - Germany’s conditional emissions display a downward trend shift in 1991 (likely reflecting German reunification in October 1990 and consequent structural changes).
    - China’s conditional emissions display a downward trend shift in 1978 (coinciding with economic reforms and opening-up that reshaped economic structure away from heavy industries).
    - China’s conditional emissions show another downward shift in 2012; following a severe air pollution crisis in January 2013, the central government announced the Air Pollution Prevention and Control Action Plan in September 2013 and later changed official evaluation incentives in December 2013—illustrating how pollution crises and consequent policy responses may drive emission shifts.
- Statistical significance:
  - Results reported are obtained using a 5 percent significance level.
  - Results under a 1 percent significance level are presented in appendices and are similar but show fewer structural breaks (more stringent criteria yield fewer breaks).

### Conclusions and policy implications
- Major global initiatives (Kyoto Protocol in 2005, Paris Agreement in 2015-2016) and “traditional” climate policies implemented so far do not seem to have reduced trends in either unconditional or GDP-conditional carbon emissions among the top emitters.
- Case studies indicate economic structural changes and strong political responses to pollution crises have played significant roles in some downward structural breaks.
- Existing climate policies implemented so far appear insufficient to reduce absolute levels of carbon emissions; bolder, more transformative climate policies are urgently needed to arrest rising trends, reduce absolute levels, and aim for carbon neutrality by 2050.
- Policy design suggestions:
  - Consider broader economic transitions (changes in economic structure) alongside traditional climate and carbon policies.
  - Explore policy areas such as green agri-food systems and international carbon price floors as potential complements to existing measures.
  - Recognize that non-climate policies that reshape the economic structure can be critical for emission outcomes.

*Source: wpiea2022009-print-pdf - Appendix Table 11. Note that the emissions in some countries, especially in AEs, experienced*

### Appendix 1. Climate Policies and Carbon Emissions: A Panel Data Analysis

### Appendix 1. Climate Policies and Carbon Emissions: A Panel Data Analysis

### A. Description of raw data
- Policy data source: Climate Policy Database (NewClimate Institute with support from PBL Netherlands Environmental Assessment Agency and Wageningen University and Research).
- Coverage: 1927 to 2021 with a total of 4,980 policy measure occurrences in the raw data.
- Distinctive policy types: 1,458.
- Jurisdiction breakdown (full sample):
  - Country: 4,383 (92.90 percent)
  - Subnational region: 266 (5.64 percent)
  - City: 44 (0.93 percent)
  - Supranational region: 25 (0.53 percent)
  - Total: 4,718 (100.00 percent)
- Sector top frequencies (partial):
  - General (economy-wide): 1,001 (21.3 percent)
  - Electricity and heat, Renewables: 680 (14.5 percent)
  - Electricity and heat: 301 (6.4 percent)
  - Transport: 261 (5.6 percent)
  - Buildings: 207 (4.4 percent)
- Recorded policy impact:
  - Unknown: 4,273 (92.9 percent)
  - High: 261 (5.7 percent)
  - Unclear: 65 (1.4 percent)
  - Total: 4,599 (100.0 percent)
- Top 10 most frequent policy instruments (sub-sample, Appendix Table 1; cum. 25.0 percent):
  - Strategic planning: 267 (5.7 percent)
  - Grants and subsidies: 194 (4.1 percent)
  - Tax relief: 132 (2.8 percent)
  - Policy support: 93 (2.0 percent)
  - Target, GHG reduction target, Political & non-binding GHG reduction target: 93 (2.0 percent)
  - Fiscal or financial incentives: 84 (1.8 percent)
  - Regulatory Instruments: 84 (1.8 percent)
  - Policy support, Strategic planning: 77 (1.6 percent)
  - Building codes and standards: 75 (1.6 percent)
  - Energy and other taxes: 75 (1.6 percent)
  - Sub-Total: 1,174 (25.0 percent)

### B. Climate Policy Intensity Index (CPII): construction and caveats
- Purpose: summary statistic capturing frequency/intensity of climate policy measures.
- Data used in CPII: policies from 1965 to 2019, covering 193 countries/regions.
- Scoring algorithm (exact rules preserved):
  - Initial score for each policy occurrence: 1.
  - If jurisdiction = “Subnational region” or “City” and sector = “General”, set score to 0.5.
  - If jurisdiction = “Subnational region” or “City” and sector is a specific sector (e.g., energy), set score to 0.25.
  - If Impact indicator = “High”, double the score.
  - Supranational measures receive same weight as country-level measures.
  - Take cumulative sum of scores from 1965 to current year (assumes measures remain in place perpetually).
- Key caveat: CPII mostly measures quantity rather than quality of policies; weights are judgment-based and index should be used as preliminary/illustrative only.

### C. Cleaned data and methodology
- Final cleaned sample: 18,335 observations.
- CPII descriptive statistics (cleaned sample): mean = 3.2; standard deviation = 18.6.
- Dependent variable: annualized three-year cumulative growth rate of carbon emissions (CO2 Growth).
- Independent variables (annualized three-year cumulative growth rates/changes):
  - CPII_3YrChange (annualized 3-year difference in the CPII)
  - GDP_3YrCumGrowth
  - Coal_3YrCumGrowth
  - Oil_3YrCumGrowth
  - kgov_3YrCumGrowth (general government capital stock growth)
  - kpriv_3YrCumGrowth (private capital stock growth)
- Summary statistics for top 20 emitters (Appendix Table 5; all growth rates/changes are annualized three-year cumulative):
  - year: Obs = 1,060; Min = 1965; Max = 2019
  - CO2 Growth: Obs = 1,000; Mean = 4.2; Std. Dev. = 3.1; Min = 0.5; Max = 24.3
  - CPII: Obs = 1,060; Mean = 35.4; Std. Dev. = 64.3; Min = 0.0; Max = 424.0
  - CPII Change: Obs = 1,000; Mean = 2.8; Std. Dev. = 4.5; Min = 0.0; Max = 41.2
  - Change of Unweighted Policy Count: Obs = 1,000; Mean = 2.8; Std. Dev. = 4.6; Min = 0.0; Max = 41.0
  - GDP Growth: Obs = 961; Mean = 3.8; Std. Dev. = 4.4; Min = -14.8; Max = 45.0
  - Coal Growth: Obs = 964; Mean = 2.1; Std. Dev. = 8.3; Min = -23.3; Max = 97.9
  - Oil Growth: Obs = 1,000; Mean = 3.0; Std. Dev. = 7.2; Min = -21.9; Max = 94.6
  - Government Capital Growth: Obs = 910; Mean = 3.6; Std. Dev. = 3.3; Min = -2.1; Max = 27.8
  - Private Capital Growth: Obs = 910; Mean = 4.2; Std. Dev. = 4.2; Min = -3.2; Max = 28.6
- Methodology: panel data regressions (fixed-effect and random-effect models); model selection via Hausman test. Time dummies not included for stated reasons.

### D. Results and interpretation
- Main result: three-year cumulative change in CPII is negatively correlated with three-year cumulative growth rate of carbon emissions; major non-policy regressors (GDP, coal, oil, government capital, private capital) are positively correlated with carbon growth.
- Key coefficient estimates (Appendix Table 6; p-values in parentheses; significance levels preserved):
  - CPII_3YrChange:
    - Column (1) FE: -0.134*** (0.000)
    - Column (2) RE: -0.135*** (0.000)
    - Column (3) FE with GDP: -0.106*** (0.000)
    - Column (4) RE with GDP: -0.107*** (0.000)
    - Column (5) FE with GDP Coal Oil: -0.088*** (0.000)
    - Column (6) RE with GDP Coal Oil: -0.087*** (0.000)
    - Column (7) FE with GDP Coal Oil Capital Capital: -0.080*** (0.000)
    - Column (8) RE with GDP Coal Oil Capital Capital: -0.079*** (0.000)
  - GDP_3YrCumGrowth:
    - Columns where included: positive and significant (e.g., 0.219***, 0.224***, 0.046***, 0.050***, 0.027*, 0.028** with respective p-values shown in table).
  - Coal_3YrCumGrowth and Oil_3YrCumGrowth: positive and significant where included (exact coefficients preserved in Table 6).
  - kgov_3YrCumGrowth: 0.115*** and 0.119*** where included.
  - kpriv_3YrCumGrowth: 0.062*** and 0.065*** where included.
- Constants (selected):
  - Column (1): Constant = 4.613*** (0.000)
  - Column (7): Constant = 3.198*** (0.000)
- Model diagnostics (Appendix Table 6):
  - Observations range: 1,000 to 838 depending on specification.
  - Number of countries: 20 in most columns; Columns (7) and (8) have 19 countries (capital stock data for Australia unavailable).
  - Hausman Test P-values (row): 0.410, 0.000, 0.000, 0.005 — fixed-effect model preferred in three out of four cases at the 5 percent significance level.
- Economic importance: implied minimum annualized CPII change required to induce negative carbon emission growth:
  - Based on Column (1): 34.4 (i.e., roughly 34.4 new climate policies per year).
    - Derivation: in absence of new policies (CPII change = 0) carbon emissions grow at 4.613 percent; given marginal impact of -0.134, required change = 4.613 / 0.134 = 34.4.
    - Comparison: maximum observed CPII change in top 20 emitters = 41.2; average observed CPII change = 2.8.
  - Based on Column (7) (preferred model): roughly 53.7 new climate policies per year required for a “typical” country in a “typical” year among the 19 emitters in that regression, much larger than average observed CPII change of 2.8.
- Overall interpretation:
  - Climate policies implemented to date in top 20 emitters have been effective in slowing carbon emission growth rates.
  - Existing policies appear insufficient to reduce the absolute level of carbon emissions across most top 20 emitters.
- Important caveat: CPII does not fully capture policy quality or cross-country comparability; panel regression results should be interpreted with caution and complemented by country-specific analysis.

### E. Robustness checks
- Alternative policy measure: unweighted/raw count of policy instruments used instead of weighted CPII — results very similar.
- Alternative time horizons:
  - Current-year changes and five-year cumulative changes were tested; results consistent.
  - Magnitudes of policy impacts (absolute value) increase with horizon:
    - Current-year changes: 0.059
    - Three-year cumulative changes: 0.080
    - Five-year cumulative changes: 0.092
  - Interpretation: verifies that climate policies take time to affect carbon emissions.
- Lag specification: three-year lagged year-on-year policy changes used (only policy change variables lagged; other variables contemporaneous) — results similar in signs and magnitudes.

*Italic sources: Climate Policy Database; Global Carbon Project; World Bank; IMF; Our World in Data; and Authors’ calculations.*

### Appendix Table 7. Panel Regressions of Carbon Emission Growth on Climate Policies: Raw

### Appendix Table 7. Panel Regressions of Carbon Emission Growth on Climate Policies: Raw

### Key regression estimates (coefficients; p-values in parentheses)
- CumCount_3YrChange:
  - Column (1): -0.125*** (0.000)
  - Column (2): -0.125*** (0.000)
  - Column (3): -0.100*** (0.000)
  - Column (4): -0.100*** (0.000)
  - Column (5): -0.082*** (0.000)
  - Column (6): -0.081*** (0.000)
  - Column (7): -0.073*** (0.000)
  - Column (8): -0.073*** (0.000)
- GDP_3YrCumGrowth:
  - Column (3): 0.221*** (0.000)
  - Column (4): 0.226*** (0.000)
  - Column (5): 0.048*** (0.001)
  - Column (6): 0.052*** (0.000)
  - Column (7): 0.028** (0.043)
  - Column (8): 0.029** (0.034)
- Coal_3YrCumGrowth:
  - Column (5): 0.017*** (0.001)
  - Column (6): 0.019*** (0.000)
  - Column (7): 0.011** (0.038)
  - Column (8): 0.012** (0.022)
- Oil_3YrCumGrowth:
  - Column (5): 0.099*** (0.000)
  - Column (6): 0.100*** (0.000)
  - Column (7): 0.098*** (0.000)
  - Column (8): 0.099*** (0.000)
- kgov_3YrCumGrowth:
  - Column (7): 0.118*** (0.000)
  - Column (8): 0.122*** (0.000)
- kpriv_3YrCumGrowth:
  - Column (7): 0.061*** (0.000)
  - Column (8): 0.064*** (0.000)
- Constant (Columns (1)-(8)):
  - 4.585*** (0.000); 4.494*** (0.000); 3.694*** (0.000); 3.533*** (0.000); 3.721*** (0.000); 3.584*** (0.000); 3.166*** (0.000); 3.013*** (0.000)

### Model and sample statistics
- Observations: 1,000; 1,000; 961; 961; 928; 928; 838; 838
- R-squared (reported): 0.075; 0.268; 0.384; 0.453
- Number of Countries: 20; 20; 20; 20; 20; 20; 19; 19
- Hausman Test P-value: 0.354; 0.000; 0.000; 0.006

Notes: CumCount_3YrChange is the annualized 3-year difference in the unweighted climate policy counts. 3YrCumGrowth is the annualized 3-year cumulative growth rate. kgov = government capital; kpri = private capital. FE = Fixed Effect, RE = Random Effect. pval in parentheses, *** p<0.01, ** p<0.05, * p<0.1.

---

### Appendix Table 8. Panel Regressions of Carbon Emission Growth on Climate Policies: Current-Year

### Key regression estimates (coefficients; p-values in parentheses)
- CPII_Change:
  - Column (1): -0.089*** (0.000)
  - Column (2): -0.089*** (0.000)
  - Column (3): -0.077*** (0.000)
  - Column (4): -0.077*** (0.000)
  - Column (5): -0.065*** (0.000)
  - Column (6): -0.065*** (0.000)
  - Column (7): -0.059*** (0.000)
  - Column (8): -0.059*** (0.000)
- GDP_Growth:
  - Column (3): 0.119*** (0.000)
  - Column (4): 0.122*** (0.000)
  - Column (5): 0.022** (0.037)
  - Column (6): 0.024** (0.024)
  - Column (7): 0.014 (0.150)
  - Column (8): 0.015 (0.127)
- Coal_Growth:
  - Column (5): 0.006** (0.038)
  - Column (6): 0.006** (0.023)
  - Column (7): 0.003 (0.240)
  - Column (8): 0.003 (0.188)
- Oil_Growth:
  - Column (5): 0.077*** (0.000)
  - Column (6): 0.079*** (0.000)
  - Column (7): 0.074*** (0.000)
  - Column (8): 0.075*** (0.000)
- kgov_Growth:
  - Column (7): 0.119*** (0.000)
  - Column (8): 0.123*** (0.000)
- kpriv_Growth:
  - Column (7): 0.077*** (0.000)
  - Column (8): 0.080*** (0.000)
- Constant (Columns (1)-(8)):
  - 4.232*** (0.000); 4.155*** (0.000); 3.794*** (0.000); 3.648*** (0.000); 3.670*** (0.000); 3.561*** (0.000); 3.056*** (0.000); 2.929*** (0.000)

### Model and sample statistics
- Observations: 1,040; 1,040; 1,001; 1,001; 968; 968; 876; 876
- R-squared (reported): 0.067; 0.178; 0.290; 0.387
- Number of Countries: 20; 20; 20; 20; 20; 20; 19; 19
- Hausman Test P-value: 0.375; 0.000; 0.000; 0.003

Notes: CPII_Change is the 1-year difference in the CPII. Growth is year-on-year. kgov = government capital; kpri = private capital. FE = Fixed Effect, RE = Random Effect. pval in parentheses, *** p<0.01, ** p<0.05, * p<0.1.

---

### Appendix Table 9. Panel Regressions of Carbon Emission Growth on Climate Policies: Five-Year

### Key regression estimates (coefficients; p-values in parentheses)
- CPII_5YrChange:
  - Column (1): -0.163*** (0.000)
  - Column (2): -0.164*** (0.000)
  - Column (3): -0.123*** (0.000)
  - Column (4): -0.124*** (0.000)
  - Column (5): -0.102*** (0.000)
  - Column (6): -0.100*** (0.000)
  - Column (7): -0.092*** (0.000)
  - Column (8): -0.092*** (0.000)
- GDP_5YrCumGrowth:
  - Column (3): 0.295*** (0.000)
  - Column (4): 0.300*** (0.000)
  - Column (5): 0.097*** (0.000)
  - Column (6): 0.103*** (0.000)
  - Column (7): 0.062*** (0.000)
  - Column (8): 0.064*** (0.000)
- Coal_5YrCumGrowth:
  - Column (5): 0.015** (0.013)
  - Column (6): 0.018*** (0.004)
  - Column (7): 0.010 (0.112)
  - Column (8): 0.012* (0.067)
- Oil_5YrCumGrowth:
  - Column (5): 0.106*** (0.000)
  - Column (6): 0.107*** (0.000)
  - Column (7): 0.108*** (0.000)
  - Column (8): 0.110*** (0.000)
- kgov_5YrCumGrowth:
  - Column (7): 0.116*** (0.000)
  - Column (8): 0.121*** (0.000)
- kpriv_5YrCumGrowth:
  - Column (7): 0.044*** (0.002)
  - Column (8): 0.047*** (0.001)
- Constant (Columns (1)-(8)):
  - 4.965*** (0.000); 4.858*** (0.000); 3.697*** (0.000); 3.515*** (0.000); 3.771*** (0.000); 3.607*** (0.000); 3.294*** (0.000); 3.105*** (0.000)

### Model and sample statistics
- Observations: 960; 960; 921; 921; 888; 888; 800; 800
- R-squared (reported): 0.083; 0.320; 0.438; 0.480
- Number of Countries: 20; 20; 20; 20; 20; 20; 19; 19
- Hausman Test P-value: 0.419; 0.003; 0.000; 0.004

Notes: CPII_5YrChange is the annualized 5-year difference in the CPII. 5YrCumGrowth is the annualized 5-year cumulative growth rate. kgov = government capital; kpri = private capital. FE = Fixed Effect, RE = Random Effect. pval in parentheses, *** p<0.01, ** p<0.05, * p<0.1.

---

### Appendix Table 10. Panel Regressions of Carbon Emission Growth on Climate Policies: Three-Year Lagged Policy Changes

### Key regression estimates (coefficients; p-values in parentheses)
- L3.CPII_Change:
  - Column (1): -0.089*** (0.000)
  - Column (2): -0.090*** (0.000)
  - Column (3): -0.080*** (0.000)
  - Column (4): -0.081*** (0.000)
  - Column (5): -0.072*** (0.000)
  - Column (6): -0.072*** (0.000)
  - Column (7): -0.062*** (0.000)
  - Column (8): -0.062*** (0.000)
- GDP_Growth:
  - Column (3): 0.115*** (0.000)
  - Column (4): 0.118*** (0.000)
  - Column (5): 0.028** (0.011)
  - Column (6): 0.030*** (0.007)
  - Column (7): 0.028*** (0.006)
  - Column (8): 0.029*** (0.005)
- Coal_Growth:
  - Column (5): 0.005** (0.045)
  - Column (6): 0.006** (0.030)
  - Column (7): 0.003 (0.213)
  - Column (8): 0.003 (0.167)
- Oil_Growth:
  - Column (5): 0.062*** (0.000)
  - Column (6): 0.063*** (0.000)
  - Column (7): 0.054*** (0.000)
  - Column (8): 0.055*** (0.000)
- kgov_Growth:
  - Column (7): 0.138*** (0.000)
  - Column (8): 0.142*** (0.000)
- kpriv_Growth:
  - Column (7): 0.068*** (0.000)
  - Column (8): 0.071*** (0.000)
- Constant (Columns (1)-(8)):
  - 4.147*** (0.000); 4.060*** (0.000); 3.759*** (0.000); 3.617*** (0.000); 3.663*** (0.000); 3.566*** (0.000); 2.995*** (0.000); 2.878*** (0.000)

### Model and sample statistics
- Observations: 980; 980; 955; 955; 922; 922; 833; 833
- R-squared (reported): 0.071; 0.177; 0.238; 0.348
- Number of Countries: 20; 20; 20; 20; 20; 20; 19; 19
- Hausman Test P-value: 0.344; 0.000; 0.000; 0.009

Notes: L3.CPII_Change is the 3-year lagged change in the CPII. Growth is year-on-year. kgov = government capital; kpri = private capital. FE = Fixed Effect, RE = Random Effect. pval in parentheses, *** p<0.01, ** p<0.05, * p<0.1.

---

### Structural breaks, decomposition, and figures (Appendices 3–7, 11–12)
- Appendix Figure 3 and Appendix Figures 6–7 present structural breaks (unconditional and conditional analyses) for groups of countries: "14 Countries", "Top 20 Emitters".
- Appendix Table 11 (Timing of Structural Breaks for 14 Countries: Unconditional Analysis) and Appendix Table 12 (Timing of Structural Breaks for 14 Countries: Conditional Analysis) report year and sign for trend breaks and level shifts for individual countries (EMs and AEs). Example entries include specific trend-break years and level-shift years for countries such as BRA, IRN, MEX, POL, RUS, SAU, UKR, AUS, CAN, ESP, FRA, ITA, JPN, KOR as shown in the tables.
- Appendix Figure 4 provides a Kaya Decomposition for 14 Countries (Panels (a) EMs and (b) AEs).
- Appendix Figure 5 presents Structural Breaks for 14 Countries: Conditional Analysis (Panels (a) EMs and (b) AEs).

Sources for figures and structural-break tables: Global Carbon Project; World Bank; Our World in Data; IMF; and Authors' calculations.

*Source: wpiea2022009-print-pdf - Appendix Table 7. Panel Regressions of Carbon Emission Growth on Climate Policies: Raw*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022009-print-pdf.pdf_