## wpiea2021140-print-pdf - 1.3 percentage point increase in the overall unemployment rate in the province, with the increase being disproportionately higher among low- and medium- skilled workers.

## Source details

**Canonical URL:** [wpiea2021140-print-pdf - 1.3 percentage point increase in the overall unemployment rate in the province, with the increase being disproportionately higher among low- and medium- skilled workers.](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021140-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021140-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021140-print-pdf.pdf.json)

---

### Major empirical findings
- Stricter environmental policies lead to reallocation of labor across firms by emission intensity:
  - Reduced labor demand among high-emission firms (examples: carbon emission-intensive manufacturing including chemicals, metals and minerals, paper and packaging, and food and beverages, and construction sector firms).
  - Increased labor demand among low-emission firms (services show positive effects).
- Tightening environmental policies amid contractions appears to have a small positive employment effect, other things equal.
- Net effect of tighter environmental policies is modest:
  - A one-standard deviation tightening in aggregate EPS would result in employment losses of about 1 percent of aggregate employment.
  - Tighter market EPS results in a modest net employment gain.
  - Tighter non-market EPS results in a modest net employment loss.
- Heterogeneity across policy subcomponents and sectors:
  - For market-based trading schemes, the positive employment effects for low-emission intensity firms are relatively large and the negative effects for high-emission intensity firms are relatively small.
  - Estimates for taxes on emissions of nitrogen oxide show the opposite pattern (larger negative effects on high-emission firms).
- Medium-term dynamics:
  - The impact of environmental policies appears to reverse and fade away over the medium term, with no lasting effects beyond a 2-year time horizon.

### Quantification and magnitudes
- One-standard deviation tightening in aggregate EPS → employment losses of about 1 percent of aggregate employment.
- Precise near-term quantified effects (Specification 1, Table 2B and Figure 2):
  - A 1-standard deviation tightening in the aggregate EPS indicator would lower employment in high-emission intensity firms by 5 percent, and raise employment in low-emission intensity firms by 3.3 percent.
  - A 1-standard deviation tightening in the market EPS indicator would lower employment in high-emission intensity firms by 3.1 percent, while raising it in low-emission intensity firms by 3.5 percent.
  - A 1-standard deviation tightening in the non-market EPS indicator would lower employment in high-emission firms by 6.1 percent and raise it in low-emission firms by 3.5 percent.
- Net employment impacts (sample-level):
  - A 1-standard deviation increase in aggregate EPS leads to a net loss of between 500-600 thousand jobs in the entire sample of firms (about 1 percent of the total employment in the sample).
  - A 1-standard deviation tightening in market EPS results in a small net job increase between 13-34 thousand jobs.
  - Tightening all jointly significant market-based policies by 1 unit would increase employment by 1.8 percent in the sample distribution.
  - Tightening all jointly significant non-market policies by 1 unit would reduce employment by 0.8 percent in the sample distribution.
- Emissions and sample statistics:
  - The emissions-to-employment ratio of median high-emission intensity firm is more than 12 times as large as that of the median low-emission intensity firm.
  - Sample sizes: 670 firms when using firm-level CO2 emission data; 5305 firms when using sectoral dummies to proxy emission-intensity.
  - Country and time coverage: Data span 31 countries over 2000-2015.
  - EPS index scaled from 0 – 6 (from low to high stringency). EPS contains market and non-market sub-indices.

### Data and empirical specification
- Primary firm-level data source: Worldscope (balance sheets of listed firms) providing annual firm-level data on total number of employees, total staff costs, and value of physical capital in machinery and buildings.
- Firm-level CO2-equivalent emissions: total CO2-equivalent emissions in tons per year, including direct and indirect emissions; emission intensity constructed as emissions scaled by total employment.
- Sample construction and emission-intensity coding:
  - Only longest unbroken spell of continuous data per firm selected.
  - Firms included only if reporting at least 3 instance of CO2 emissions.
  - A firm is coded as high-emission intensity (d_high = 1) if its emissions-to-employment ratio exceed the median of the country-year distribution in any year for which it reports data (time-invariant coding).
- Econometric approach:
  - Estimating framework based on Van Reenen (1997).
  - Identification strategy: difference-in-difference via interaction of country-year EPS with firm emission-intensity indicator.
  - Estimation methodology: panel GMM to address endogeneity from lagged employment and fixed effects (Nickell bias).
  - All variables expressed in logs except EPS which enters in levels.
  - Controls include EPS, firm-level average annual employee wages (logs), real capital stock (logs), rental rate of capital (logs), and in additional specifications output gap controls.
- Additional data and methods (Chapter 3 October 2020):
  - Capital stocks deflated using annual series for the price of machinery and the price of building stock from Penn World Tables (PWT, version 9.1).
  - Rental rate of capital proxy from PWT used to proxy firm level R.
  - Output gap estimates taken from the IMF WEO database.
  - Panel GMM labor demand framework with controls for wages, rental rate of capital, capital stock, and lags of employment; GMM instruments include up to 9 lags of levels.
  - Medium-term dynamics estimated via Jorda local projections embedded with panel GMM; projection horizon h = 0...5 (5-year horizon).
  - Broader medium-term sample includes firms with at least 10 continuous years of data on employment, wages, and capital stock, giving a sample of about 2200 firms.
  - Underlying full dataset contains more than 42,000 firms (inactive/delisted firms are around 1 percent of the original sample).

### Specification results and sectoral heterogeneity
- Specification 1 — emission-intensity high vs low (near-term):
  - Interaction of EPS indicator with emission-intensity dummy is negative in all specifications, statistically significant for aggregate EPS and non-market EPS.
  - Lagging policy indicators by one period yields broadly similar results; market EPS and its interaction with emission-intensity are statistically significant in lagged specifications.
  - Carbon tax coefficient (and its interaction with emission-intensity) follows the same pattern as aggregate, market and non-market EPS but is insignificant in many specifications.
  - Business-cycle interaction:
    - Interaction of output gap with market EPS is negative and significant, implying tightening market EPS can have a positive employment effect in periods of economic contraction.
    - Estimates suggest in deep recessions a 1 unit tightening in market EPS could increase employment by about 2 percent, other things equal.
    - Thresholds: output gap below the 25th percentile and the 1st percentile correspond to about -1.5 percent and -6 percent of potential GDP respectively.
    - Positive and significant effect of tightening EPS during a recession observed for output gap values below the 1st quartile for aggregate and market EPS (lagged 1 period). At the 1st percentile the effect size is sizable but not statistically significant.
    - Auxiliary regressions show tighter EPS has a positive effect on inflation (statistically significant for PPI inflation and 1-year ahead inflation expectations), supporting an inflation channel that could lower real rates and stimulate demand under contractionary conditions.
- Specification 2 — sector proxy for emission-intensity (near-term):
  - Sector groups: fossil fuels, high emission manufacturing, low emission manufacturing, services, construction, transportation.
  - Interaction coefficients with aggregate EPS: fossil-fuel industries, high-emission manufacturing, construction, and transport have negative signs; construction interaction is significant.
  - Market EPS shows similar pattern; construction and services sector interactions significant in market EPS specification.
  - Non-market EPS interaction coefficients similar to aggregate and market EPS.
  - Carbon tax interactions enter with negative signs but are not significant.
  - Including output gap controls renders high-emission manufacturing interaction statistically significant in some specifications.
  - Utilities sector interaction negative but not statistically significant.
- Disaggregated policy effects (Table 7):
  - Market policies: more stringent trading schemes (green and white certificates) lead to relatively large positive effects on low-emission intensity firms and small negative effects on high-emission intensity firms.
  - Tighter taxes (e.g., NOX taxation) have sizable negative impact on employment in high-emission firms.
  - Representative quantified direction: tightening all jointly significant market-based policies by 1 unit would increase employment by 1.8 percent; tightening all jointly significant non-market policies by 1 unit would reduce employment by 0.8 percent.
  - Caveat: policy indices are not expressed in a common emission-reduction metric, so cross-policy comparisons have limitations.

### Medium-term dynamics (local projections)
- Impulse-responses over a 5-year horizon in response to a 1-unit change in EPS:
  - Aggregate EPS: small negative impact in the short term; over medium term initial negative impact tends to reverse and fade.
  - Market EPS: small positive near-term effect that reverses and fades over the medium term.
  - Non-market EPS: noticeable decline on impact; becomes positive by year 2; reverses by year 3 and fades over the medium term.
- Interpretation: initial effects (both negative and positive) are transitory and tend to dissipate over a 3–5 year horizon; no strong evidence of permanent effects.

### Robustness checks and data concerns
- Robustness:
  - Results robust to varying emission intensity thresholds (except at the 75th percentile threshold for high-emission definition).
  - Including leads of policy (lead of EPS generally not significant).
  - Weighted regressions accounting for unbalanced panels, robust standard errors, winsorizing, real wages instead of nominal wages.
  - Controlling for country-specific trends (country-year dummies) and alternative numbers of GMM instrument lags; Hansen J-test cannot reject instrument validity in alternative lag schemes.
  - Specification 2 (market EPS focus): high emissions industries and construction show negative employment effects in most robustness variants; services and low-emission industries interactions remain positive (services significant under winsorized data and with real wages).
  - Quantified net employment under robustness exercises remains modest (between +/- 0.4 percent of total employment in the sample).
- Data limitations:
  - Sparse firm-level CO2 reporting raises potential selection bias; overall 21 percent reporting (at least one year) in high-emission sectors versus sample average of 21 percent across sectors.
  - Sector reporting rates: Fossil fuel industries 26 percent, transport 24 percent, utilities 34 percent, high-emission manufacturing 15 percent.
  - Country differences in reporting: China 3 percent, India 4 percent, Indonesia 5 percent reporting at least one year versus sample average 32 percent — implying Specification 1 results are mainly reflective of advanced OECD countries.
  - Survivorship bias: delisted/inactive firms are around 1 percent of the original dataset; potential bias from delisting remains a caveat.

### Selected empirical coefficients (highlights)
- Representative labor-demand coefficients (Table 2A / 2B):
  - Log N(t-1) — 0.573***.
  - Log N(t-2) — -0.142***.
  - log capital stock — 0.276**.
  - log wages — -0.260**.
  - log r — 0.308**.
  - EPS (Aggregate) — 0.0270 (standard error (0.0273), Table 2A).
  - EPS X (High CO2=1) — -0.0785* (standard error (0.0477), Table 2A).
  - Lagged policies (Table 2B) show EPS X (High CO2=1) — -0.0858** in representative column.
- Sector interactions (Table 4 and Appendix):
  - Construction X EPS — -0.0836***.
  - Services X EPS — 0.0174 and 0.0225* in alternate columns.
  - High CO2 Industries X EPS — negative and sometimes significant (e.g., -0.0346*).
- Disaggregated policy examples (Table 7):
  - EPS (Green certificate) — 0.0399**.
  - EPS (Wind) — 0.0479**.
  - EPS X (High CO2=1) with NOX — -0.0746 (example).
  - Observations in policy-panel regressions: typically about 6,138; number of firms about 671.

### Summary of main findings
- Near-term: tighter EPS reduces labor demand in high-emission firms/sectors and increases it in low-emission ones (reallocation effect).
- Market vs non-market:
  - Non-market EPS has a nearly twice larger negative effect on high-emission intensity firms compared to market EPS.
  - Market-based policies tend to have more positive employment effects in low-emission sectors; net effects of market EPS can be positive depending on sector composition.
- Net employment impact of aggregate EPS is small but typically negative (sample: net loss 500-600 thousand jobs, about 1 percent of sample employment); market EPS net effect small and could be positive (13-34 thousand jobs in sample).
- Business cycle interaction: tightening EPS during recessions may not harm and could increase employment (possible inflation channel lowering real interest rates).
- Medium-term: initial employment effects of EPS tend to reverse and fade over a 3–5 year horizon; no strong evidence of permanent effects.

### Policy implications and recommendations
- Near-term transitional costs should be addressed to facilitate labor reallocation:
  - Reskilling the workforce to ease movement from high-carbon to low-carbon sectors.
  - Creating job opportunities in green sectors, for example through investment in green infrastructure.
- Regional policies may be required:
  - Target assistance to laid-off workers in communities and areas predominantly fossil-fuel dependent.
  - Redevelop affected regions with emphasis on low-carbon activities.
- Timing considerations:
  - Tightening environmental policies during downturns may have limited adverse employment effects and could even raise labor demand via cost-push inflation lowering real interest rates (given price rigidities and low nominal rates).
- Fiscal design:
  - Provide income support for workers vulnerable in high-emission sectors (e.g., recycling carbon tax revenues to affected households).
  - Accompany national-level increases in policy stringency with labor transition support: reskilling and public expenditure in green sectors to generate green jobs.

### Areas for further research
- Develop a more comprehensive firm-level emissions dataset to strengthen Specification 1.
- Extend analysis to unlisted and informal firms to assess differential labor demand responses.
- Incorporate richer measures of market competition to test sensitivity of firm responses.
- Express disaggregated policies in comparable emission-related units to enable cross-policy employment effect comparisons and to link policy stringency to emission reduction targets.

*Source: wpiea2021140-print-pdf (IMF).*

### 1.3 percentage point increase in the overall unemployment rate in the province, with the increase being disproportionate

### wpiea2021140-print-pdf - 1.3 percentage point increase in the overall unemployment rate in the province, with the increase being disproportionately higher among low- and medium- skilled workers.

### Major empirical findings
- Stricter environmental policies lead to reallocation of labor across firms by emission intensity:
  - Reduced labor demand among high-emission firms (examples: carbon emission-intensive manufacturing including chemicals, metals and minerals, paper and packaging, and food and beverages, and construction sector firms).
  - Increased labor demand among low-emission firms (services show positive effects).
- Tightening environmental policies amid contractions appears to have a small positive employment effect, other things equal.
- The net effect of tighter environmental policies is modest:
  - A one-standard deviation tightening in aggregate EPS would result in employment losses of about 1 percent of aggregate employment.
  - Tighter market EPS results in a modest net employment gain.
  - Tighter non-market EPS results in a modest net employment loss.
- Heterogeneity across policy subcomponents and sectors:
  - For market-based trading schemes, the positive employment effects for low-emission intensity firms are relatively large and the negative effects for high-emission intensity firms are relatively small.
  - Estimates for taxes on emissions of nitrogen oxide show the opposite pattern (larger negative effects on high-emission firms).
- Medium-term dynamics:
  - The impact of environmental policies appears to reverse and fade away over the medium term, with no lasting effects beyond a 2-year time horizon.

### Quantification and magnitudes
- One-standard deviation tightening in aggregate EPS → employment losses of about 1 percent of aggregate employment.
- The emissions-to-employment ratio of median high-emission intensity firm is more than 12 times as large as that of the median low-emission intensity firm.
- Sample sizes:
  - 670 firms when using firm-level CO2 emission data.
  - 5305 firms when using sectoral dummies to proxy emission-intensity.
- Country and time coverage:
  - Data span 31 countries over 2000-2015.
- EPS index properties:
  - EPS index scaled from 0 – 6 (from low to high stringency).
  - EPS contains market and non-market sub-indices.

### Data and empirical specification
- Primary firm-level data source: Worldscope (balance sheets of listed firms) providing:
  - Annual firm-level data on total number of employees, total staff costs, and value of physical capital in machinery and buildings.
  - Firm-level CO2-equivalent emissions (total CO2-equivalent emissions in tons per year, including direct and indirect emissions) used to construct emission intensity (emissions scaled by total employment).
- Sample construction and emission-intensity coding:
  - Only longest unbroken spell of continuous data per firm selected.
  - Firms included only if reporting at least 3 instance of CO2 emissions.
  - A firm is coded as high-emission intensity (d_high = 1) if its emissions-to-employment ratio exceed the median of the country-year distribution in any year for which it reports data (time-invariant coding).
- Econometric approach:
  - Estimating framework based on Van Reenen (1997).
  - Identification strategy: difference-in-difference via interaction of country-year EPS with firm emission-intensity indicator.
  - Estimation methodology: panel GMM to address endogeneity from lagged employment and fixed effects (Nickell bias).
  - All variables expressed in logs except EPS which enters in levels.
  - Controls include EPS, firm-level average annual employee wages (logs), real capital stock (logs), rental rate of capital (logs), and in additional specifications output gap controls.

### Policy implications and recommendations
- Near-term transitional costs should be addressed to facilitate labor reallocation:
  - Reskilling the workforce to ease movement from high-carbon to low-carbon sectors.
  - Creating job opportunities in green sectors, for example through investment in green infrastructure.
- Regional policies may be required:
  - Target assistance to laid-off workers in communities and areas predominantly fossil-fuel dependent.
  - Redevelop affected regions with emphasis on low-carbon activities.
- Timing considerations:
  - Tightening environmental policies during downturns may have limited adverse employment effects and could even raise labor demand via cost-push inflation lowering real interest rates (given price rigidities and low nominal rates).

*Source: wpiea2021140-print-pdf (IMF).*

### Chapter 3 October 2020).

### Chapter 3 October 2020

### Data and methodology
- Capital stocks deflated using annual series for the price of machinery and the price of building stock from Penn World Tables (PWT, version 9.1).
- Rental rate of capital proxy from PWT used to proxy firm level R (common price of capital within a country).
- Output gap estimates taken from the IMF WEO database.
- Panel GMM labor demand framework with controls for wages, rental rate of capital, capital stock, and lags of employment; GMM instruments include up to 9 lags of levels.
- Medium-term dynamics estimated via Jorda local projections embedded with panel GMM; projection horizon h = 0...5 (5-year horizon).
- Broader medium-term sample includes firms with at least 10 continuous years of data on employment, wages, and capital stock, giving a sample of about 2200 firms.
- Underlying full dataset contains more than 42,000 firms (inactive/delisted firms are around 1 percent of the original sample).

### Specification 1 — emission-intensity high vs low (near-term)
Findings:
- Tighter environmental policy stringency (EPS) leads to a decrease in employment in high emission firms and an increase in low-emission firms.
- Interaction of EPS indicator with emission-intensity dummy is negative in all specifications, statistically significant for aggregate EPS and non-market EPS.
- Lagging policy indicators by one period yields broadly similar results; market EPS and its interaction with emission-intensity are statistically significant in lagged specifications.
- Carbon tax coefficient (and its interaction with emission-intensity) follows the same pattern as aggregate, market and non-market EPS but is insignificant in many specifications.

Precise quantified effects (from Table 2B and Figure 2):
- A 1-standard deviation tightening in the aggregate EPS indicator would lower employment in high-emission intensity firms by 5 percent, and raise employment in low-emission intensity firms by 3.3 percent.
- A 1-standard deviation tightening in the market EPS indicator would lower employment in high-emission intensity firms by 3.1 percent, while raising it in low-emission intensity firms by 3.5 percent.
- A 1-standard deviation tightening in the non-market EPS indicator would lower employment in high-emission firms by 6.1 percent and raise it in low-emission firms by 3.5 percent.
- t-statistics noted: interaction with market EPS t-statistic = 1.5; in some lagged cases t-statistic falls to 1.6.

Business-cycle interaction:
- Interaction of output gap with market EPS is negative and significant, implying tightening market EPS can have a positive employment effect in periods of economic contraction.
- Estimates suggest in deep recessions a 1 unit tightening in market EPS could increase employment by about 2 percent, other things equal.
- Thresholds examined: output gap below the 25th percentile and the 1st percentile correspond to about -1.5 percent and -6 percent of potential GDP respectively.
- Positive and significant effect of tightening EPS during a recession observed for output gap values below the 1st quartile for aggregate and market EPS (lagged 1 period). At the 1st percentile the effect size is sizable but not statistically significant.
- Auxiliary regressions show tighter EPS has a positive effect on inflation (statistically significant for PPI inflation and 1-year ahead inflation expectations), supporting an inflation channel that could lower real rates and stimulate demand under contractionary conditions.

### Specification 2 — sector proxy for emission-intensity (near-term)
Design:
- Firms proxied by sector: fossil fuels, high emission manufacturing, low emission manufacturing, services, construction, transportation (Table 3 describes composition).
- Median firm emission-intensity by sector: services, low-emission manufacturing, and construction are lowest; high-emission industries and fossil fuels highest.

Findings:
- Interaction coefficients with aggregate EPS: fossil-fuel industries, high-emission manufacturing, construction, and transport have negative signs; construction interaction is significant.
- Market EPS shows similar pattern; construction and services sector interactions significant in market EPS specification.
- Non-market EPS interaction coefficients similar to aggregate and market EPS.
- Carbon tax interactions enter with negative signs but are not significant.
- Including output gap controls renders high-emission manufacturing interaction statistically significant in some specifications.
- Utilities sector (broader definition) interaction has expected negative sign but is not statistically significant.

### Net employment impacts (quantification)
- A 1-standard deviation increase in aggregate EPS leads to a net loss of between 500-600 thousand jobs in the entire sample of firms (about 1 percent of the total employment in the sample).
- A 1-standard deviation tightening in market EPS results in a small net job increase between 13-34 thousand jobs.
- Country-level: tighter aggregate EPS would in almost all cases lead to lower net employment, barring in the US.
- Average decline across all countries in the sample is about 1 percent of total employment.
- In the US, net increase due to tighter market EPS is 1.2 percent of total sample employment in 2015.
- Net effects are sensitive to sectoral composition and sample representativeness; where Worldscope and national accounts service shares are similar (Canada, Czech Republic, Denmark, Greece, Portugal, Turkey, UK, US), tighter market EPS yields net gains in most cases.
- Quantification sensitivity: alternative robustness exercises yield net effects typically modest, between +/- 0.4 percent of total employment in the sample.

Distributional notes:
- Median wages lowest among high-emission and low-emission manufacturing sectors; highest among fossil fuel industries, followed by services, construction, and transport.
- Reallocation toward services could offset distributional effects from the loss of higher-wage fossil fuels/construction/transport jobs in the listed-firm sample.

### Medium-term dynamics (local projections)
- Impulse-responses over a 5-year horizon in response to a 1-unit change in EPS:
  - Aggregate EPS: small negative impact in the short term; over medium term initial negative impact tends to reverse and fade.
  - Market EPS: small positive near-term effect that reverses and fades over the medium term.
  - Non-market EPS: noticeable decline on impact; becomes positive by year 2; reverses by year 3 and fades over the medium term.
- Interpretation: initial effects (both negative and positive) are transitory and tend to dissipate over a 3–5 year horizon.

### Robustness checks
- Results robust to:
  - Varying emission intensity thresholds (except at the 75th percentile threshold for high-emission definition).
  - Including leads of policy (lead of EPS generally not significant).
  - Weighted regressions accounting for unbalanced panels, robust standard errors, winsorizing, real wages instead of nominal wages.
  - Controlling for country-specific trends (country-year dummies) and alternative numbers of GMM instrument lags; Hansen J-test cannot reject instrument validity in alternative lag schemes.
- Specification 2 (market EPS focus): high emissions industries and construction show negative employment effects in most robustness variants; services and low-emission industries interactions remain positive (services significant under winsorized data and with real wages).
- Quantified net employment under robustness exercises remains modest (between +/- 0.4 percent of total employment in the sample).

### Disaggregated policy effects
- Each component of market and non-market EPS indices examined with a one-period lag (Table 7).
- Patterns consistent with reallocation effect: market and non-market subcomponents generally show negative interaction coefficients for high-emission firms and positive for low-emission firms when jointly significant.
- Among market policies:
  - More stringent trading schemes (green and white certificates) lead to relatively large positive effects on low-emission intensity firms and small negative effects on high-emission intensity firms.
  - Tighter taxes (e.g., NOX taxation) have sizable negative impact on employment in high-emission firms.
- Quantified direction of net effects using sample distribution:
  - Tightening all jointly significant market-based policies by 1 unit would increase employment by 1.8 percent in the sample distribution.
  - Tightening all jointly significant non-market policies by 1 unit would reduce employment by 0.8 percent in the sample distribution.
- Caveat: comparisons across indices or policy types are not directly meaningful because policy indices are not expressed in a common emission-reduction metric.

### Other estimation and data concerns
- Sparse firm-level CO2 reporting raises potential selection bias; however reporting rates by sector show:
  - Overall 21 percent reporting (at least one year) in high-emission sectors versus sample average of 21 percent across sectors.
  - Fossil fuel industries 26 percent, transport 24 percent, utilities 34 percent, high-emission manufacturing 15 percent.
- Country differences in reporting: China 3 percent, India 4 percent, Indonesia 5 percent reporting at least one year versus sample average 32 percent — implying Specification 1 results are mainly reflective of advanced OECD countries.
- Survivorship bias: delisted/inactive firms are a small fraction (around 1 percent) of the original dataset; potential bias from delisting remains a caveat.

### Summary of main findings
- Near-term: tighter EPS reduces labor demand in high-emission firms/sectors and increases it in low-emission ones (reallocation effect).
- Market vs non-market:
  - Non-market EPS has a nearly twice larger negative effect on high-emission intensity firms compared to market EPS.
  - Market-based policies tend to have more positive employment effects in low-emission sectors; net effects of market EPS can be positive depending on sector composition.
- Net employment impact of aggregate EPS is small but typically negative (sample: net loss 500-600 thousand jobs, about 1 percent of sample employment); market EPS net effect small and could be positive (13-34 thousand jobs in sample).
- Business cycle interaction: tightening EPS during recessions may not harm and could increase employment (possible inflation channel lowering real interest rates).
- Medium-term: initial employment effects of EPS tend to reverse and fade over a 3–5 year horizon; no strong evidence of permanent effects.

### Policy implications and further research
- Policy recommendations:
  - Accompany national-level increases in policy stringency with labor transition support: reskilling and public expenditure in green sectors to generate green jobs.
  - Provide income support for workers vulnerable in high-emission sectors (e.g., recycling carbon tax revenues to affected households).
- Areas for further work:
  - Develop a more comprehensive firm-level emissions dataset to strengthen Specification 1.
  - Extend analysis to unlisted and informal firms to assess differential labor demand responses.
  - Incorporate richer measures of market competition to test sensitivity of firm responses.
  - Express disaggregated policies in comparable emission-related units to enable cross-policy employment effect comparisons and to link policy stringency to emission reduction targets.

*Source: wpiea2021140-print-pdf - Chapter 3 October 2020).*

### References

### References

### Bibliographic entries
- Ben-David, I., S. Kleimeier, and M. Viehs, 2020, “Exporting pollution: where do multinational firms emit CO2?”, NBER Working Paper 25063.
- Deschenes, Olivier, 2010, “Climate policy and labor markets,” NBER Working Paper 16111.
- Deschenes, Olivier, 2018, “Environmental regulations and labor markets,” IZA World of Labor 2018.
- Eggertson, G., 2012, “Was the New Deal contractionary?”, American Economic Review 102(1).
- Greenstone, M., 2002, “The Impacts of Environmental Regulations on Industrial Activity: Evidence from the 1970 and 1977 Clean Air Act Amendments and the Census of Manufactures,” Journal of Political Economy 110.
- IMF, 2019, “How to Mitigate Climate Change,” Chapter 3 in Fiscal Monitor, October 2019, Washington DC.
- IMF, 2020, “Mitigating Climate Change – Growth and Distribution Friendly Strategies,” Chapter 3 in World Economic Outlook, October 2020, Washington D.C.
- IMF, 2020, “Sectoral Policies for Climate Change Mitigation in the EU,” IMF Departmental Paper (European Department), Washington D.C.
- Hepburn, C., 2006, “Regulation by prices, quantities or both: a review of instrument choices,” Oxford Review of Economic Policy 22(2).
- Huang, L., G. Krigsvoll, F. Johansen, Y. Liu, X. Zhang. 2018, “Carbon emissions of global construction sector,” Renewable and Sustainable Energy Reviews 81.
- Kahn, M.E., 1997, “Particulate pollution trends in the United States,” Regional Science and Urban Economics 27.
- Liu, M., R. Shadbegian, and B. Zhang, 2017, “Does environmental regulation affect labor demand in China? Evidence from the textile printing and dyeing industry,” Journal of Environmental Economics and Management 86.
- Metcalf, G.E., and J.H. Stock, 2020, “Measuring the Macroeconomic Impact of Carbon Taxes,” AEA papers and Proceedings 110.
- OECD, 2014, “Measuring environmental policy stringency in OECD countries – a composite index approach,” Economic Department Working Papers No. 1177, Paris.
- OECD, 2020, “Measuring and assessing the effects of environmental policy uncertainty,” ECO/CPE/WP1/2020(22), Paris.
- Probst, M., and C. Sauter, 2015, “CO2 Emissions and Greenhouse Gas Policy Stringency – An Empirical Assessment,” Working Paper 15-03, IRENE.
- Van Reenen, J., 1997, “Employment and Technological Innovation: Evidence from U.K. Manufacturing Firms,” Journal of Labor Economics, 15(2).
- Yip, Chi Man, 2018, “On the labor market consequences of environmental taxes,” Journal of Environmental Economics and Management 89.
- Yamazaki, A., 2017, “Jobs and climate policy: Evidence from British Columbia’s revenue-neutral carbon tax,” Journal of Environmental Economics and Management 83.
- Weitzman, M., 1974, “Prices vs Quantities,” Review of Economic Studies, October, 41(4).
- Wolde-Rufael, Y., and E. Mulat-Weldemeskel, 2021, “Do environmental taxes and environmental stringency policies reduce CO2 emissions? Evidence from 7 emerging economies,” Environmental Science and Pollution Research (2021).

### Tables and empirical material (references to empirical tables in the chapter)
- Table 1: Summary statistics (Estimation sample 1 includes 670 firms, from 30 countries over 2000-2015. Sample 2 consists of 5305 firms, covering 31 countries over 2000-2015.)
  - Sample 1: interaction with high/low CO2 (Variable — Mean — Std. Dev.)
    - Aggregate EPS — 2.5 — 0.9
    - Market EPS — 1.9 — 1.0
    - EPS: CO2 Tax — 0.1 — 0.9
    - Non-Market EPS — 3.1 — 1.2
    - log employees — 9.6 — 1.5
    - log capital stock — 14.5 — 1.8
    - log wage — 3.9 — 0.8
    - log r — 0.0 — 0.1
    - Output gap — 0.0 — 2.2
    - log CO2 emissions — 8.9 — 10.0
  - Sample 2: interaction with sector dummies (Variable — Mean — Std. Dev.)
    - Aggregate EPS — 2.2 — 0.9
    - Market EPS — 1.7 — 0.9
    - EPS: CO2 Tax — 0.1 — 0.8
    - Non-Market EPS — 2.7 — 1.3
    - log employees — 7.8 — 1.9
    - log capital stock — 12.2 — 2.2
    - log wage — 3.2 — 1.2
    - log r — 0.0 — 0.1
    - Output gap — -0.2 — 2.0

- Table 3: Sector descriptions and codes (Source: Worldscope Database - Data Definitions Guide (Issue 15). 1/ Median CO2 emissions/employee (in thousand tons), 2015.)
  - Fossil fuels — Industries related to coal, oil and gas exploration, mining, refining, and marketing. — Worldscope code: 5010 — Emissions: 1281
  - High emission manufacturing — Manufacturing in chemicals, metals and minerals, paper and packaging, and food and beverages industries — Worldscope codes: 5110, 5120, 5130, 5410 — Emissions: 1101
  - Other (low emission) manufacturing — Manufacturing in renewable energy, aerospace and defense, industrial machinery and equipment, ship-building, consumer durables and non-durables, IT equipment — Worldscope codes: 5020, 5030, 5210, 5230, 5310, 5320 (excl. home building activities), 5420 (excl. personal services) — Emissions: 189
  - Services — Industrial and commercial services, consumer cyclical services (hotels and entertainment, media and publishing), retail services, personal services, financial, insurance, and real estate services, IT services. — Worldscope codes: 5220 (excl. commercial and engineering construction), 5330, 5340, 5420 (excl. personal products), 5430, 5510, 5560, 5530, 5540, 5550, 5610 — Emissions: 126
  - Construction — Commercial and industrial construction and engineering, home building — Worldscope codes: 5220 10, 5320 30 10212 — Emissions: 12
  - Transport — Transport by air, sea, and land; transport infrastructure operators — Worldscope code: 5240 — Emissions: 973
  - Utilities — Electrical fossil fuel-based utilities and electric utilities not elsewhere classified, including independent power producers — Worldscope codes: 5910101010, 5910101012, 5910102010, 5910102011 — Emissions: 13300

### Empirical results: key coefficients and robustness (selected highlights)

### Table 2A / 2B — CO2 emission intensity and labor demand (representative coefficients)
- Dependent variable: Log N
  - Log N(t-1) — 0.573*** (column 1, Table 2A)
  - Log N(t-2) — -0.142*** (column 1, Table 2A)
  - log capital stock — 0.276** (column 1, Table 2A)
  - log wages — -0.260** (column 1, Table 2A)
  - log r — 0.308** (column 1, Table 2A)
  - EPS (Aggregate) — 0.0270 (standard error reported (0.0273) in Table 2A)
  - EPS X (High CO2=1) — -0.0785* (standard error (0.0477) in Table 2A)
  - Observations: 6,415 (column 1, Table 2A)
  - Number of panel id: 674 (column 1, Table 2A)
  - Note: *** p<0.01, ** p<0.05, * p<0.1. N = employees.

- Table 2B (lagged policies, column 1 representative)
  - Log N(t-1) — 0.573***
  - Log N(t-2) — -0.142***
  - log capital stock — 0.276**
  - log wages — -0.260**
  - log r — 0.308**
  - EPS (Aggregate) — 0.0341 (standard error reported)
  - EPS X (High CO2=1) — -0.0858** (standard error reported)
  - Observations: 6,415
  - Number of panel id: 674

### Table 4 — Impact of tighter EPS on labor demand across sectors (selected significant interactions)
- Construction X EPS — -0.0836*** (column 1)
- High CO2 Industries X EPS — -0.0228 (column 2) and in other specifications negative and sometimes significant (e.g., -0.0346* in one column)
- Services X EPS — 0.0174 (column 1) and 0.0225* (column 2)
- Joint significance (p-val) reported across specifications (examples): 0.05, 0.06, 0.65, 0.05, 0.05, 0.07, 0.03, 0.07, 0.10
- Observations: ranges reported (e.g., 25,631; 25,637; 26,072)
- Number of firms: typically 5,305 (varies by column; one column reports 3845)

### Table 5 — Robustness checks (Specification 1) (selected entries)
- Specification 1, Higher emission threshold:
  - EPS (Aggregate) — 0.00833 (standard error (0.0173))
  - EPS X High Emission — -0.0752* (standard error (0.0416))
- Including leads:
  - EPS (Aggregate) — 0.0502** (standard error (0.0253))
  - EPS X High Emission — -0.0991** (standard error (0.0395))
- Weighted regression:
  - EPS (Market) — 0.0389** (standard error (0.0171))
  - EPS X High Emission — -0.0980*** (standard error (0.0379))
- Notes: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Columns 1-3 show interactions with lagged EPS. Columns 4-6 show interactions with contemporaneous EPS.

### Table 7 — Impact of specific market and non-market policies (selected coefficients)
- Panel covers multiple EPS indicators (Diesel, NOX, SOX, Green certificate, Carbon, White certificate, Wind, Solar, NOX, SOX, PM, Sulphur in diesel, R&D)
- Representative coefficient highlights:
  - Log N(t-1) — 0.561*** across columns (standard errors around (0.105)-(0.109))
  - Log N(t-2) — -0.143*** across columns
  - log capital stock — 0.283*** (column 1)
  - log wages — -0.257** (column 1)
  - EPS coefficients vary by policy indicator (examples):
    - EPS (Green certificate) — 0.0399** (standard error (0.0188))
    - EPS (Carbon) — 0.00618 (standard error (0.00493))
    - EPS (Wind) — 0.0479** (standard error (0.0203))
  - EPS X (High CO2=1) shows a mix of positive and negative interactions; examples:
    - EPS X (High CO2=1) with Diesel — 0.0456* (standard error (0.0257))
    - EPS X (High CO2=1) with NOX — -0.0746 (standard error (0.0663))
    - EPS X (High CO2=1) with Wind — -0.0500* (standard error (0.0300))
- Observations: 6,138 in most columns
- Number of firms: 671
- Note: Policies lagged one period. N = employees. NOX = nitrous oxide, SOX = sulphur oxide, PM = particulate matter. Green and White certificates refer to trading schemes requiring firms to have a certain share of renewable energy share in total energy, and required energy savings respectively; FITs are feed-in tariffs which are subsidies to renewable energy producers.

### Appendix: robustness and additional tables (selected values)
- Table 6: Robustness checks (Specification 2) — interactions with Market EPS (example rows)
  - Fossil fuel industries (Lag policies) — 0.00448 (standard error (0.0392))
  - High emission manufacturing (Weighted) — -0.0405*** (standard error (0.0150))
  - Construction (Lag policies) — -0.0744*** (standard error (0.0265))
  - Services industries — 0.0156 (standard error (0.0121))
  - Observations: 26,132; 25,631; 25,631; 25,631; 25,637 (across columns)
  - Number of firms: 5,332; 5,305; 5,305; 5,305; 305 (across columns)
  - Joint significance (p-val): Yes across columns shown

- Appendix Table A2. EPS and Inflation (selected coefficients)
  - Dependent variable: CPI
    - EPS — 0.00793 (standard error (0.0133))
  - Dependent variable: PPI
    - EPS — 0.0177** (standard error (0.00782))
  - Dependent variable: Expected inflation 1/
    - EPS — 0.0248* (standard error (0.0130))
  - Observations: 525; 430; 525
  - Number of countries: 31; 28; 31
  - Note: 1/ 1-year ahead inflation expectations are proxied IMF WEO forecasts for one-year ahead CPI inflation published in the October WEO of each year. PPI includes manufacturing and wholesale price indices depending on data availability. Lagged inflation on the right-hand side is instrumented by further lags following a GMM methodology. Both EPS and output gap are lagged one period.

- Appendix Table A1: EPS and Output gap (selected)
  - Output gap 1/ — -0.104* (Aggregate EPS column)
  - Output gap X EPS 1/ — 0.0389* (Aggregate EPS column)
  - Observations: 6,138; Number of panel id: 671
  - Note: Output gap is a dummy variable = 1 if the real output gap is below the threshold (25th, and 1st percentile of the distribution) shown. "EPS" refers to Aggregate EPS in column 1; Market EPS in column 2; and Non-market EPS in column 3. Regressions include all independent variables and controls of Specification 1 (not shown). All policies are lagged 1 period. Panel and year fixed effects are included. Wages, capital, and rental rate are GMM-instrumented with lags. N = employees.

*Content unit: wpiea2021140-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021140-print-pdf.pdf_