## wpiea2022129-print-pdf

## Source details

**Canonical URL:** [wpiea2022129-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022129-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022129-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022129-print-pdf.pdf.json)

---

### Definitions and measurement
- Green-intensive Jobs
  - Based on O*NET’s Green Task Development Project taxonomy at Standard Occupational Classification (SOC) 8-digit level (Center,2010 and Center,2021).
  - O*NET distinguishes three types of green tasks: (i) existing occupations expected to be in high demand due to the greening of the economy; (ii) occupations expected to undergo significant changes in task content due to the greening of the economy (green-enhanced); and (iii) new occupations in the green economy.
  - Vona et al.,2018 compute a ratio of green tasks to total tasks as a measure of green intensity (the “greenness ratio” defined in their equation 1).
  - Employees are assumed uniformly distributed across 8-digit occupations within each 6-digit SOC occupation to obtain green intensity at the 6-digit level.
  - Green intensity at the 6-digit occupation level is a continuum between 0 and 1.

- Pollution-intensive Jobs
  - Polluting occupations are defined as those prevalent in polluting industries.
  - Polluting industries: a set of 62 4-digit NAICS industries that are in the 95th percentile of pollution intensity for at least 3 pollutants among CO2, CO, VOC, NOx, SO2, PM10, PM2.5, and Lead.
  - Pollution-intensive occupations are defined as those with a 7 times higher probability of working in polluting sectors than in any other job.
  - At the 6-digit occupation level pollution intensity is occupation-based and takes only the value of 0 or 1.

- Important distinction
  - Green intensity is a share of green tasks (continuous between 0 and 1 at 6-digit level); pollution intensity is binary at the 6-digit occupation level.
  - When collapsing occupations (e.g., to 5-digit level, mapping SOC vintages, or converting to alternative occupation codes), employment-weighted measures are generated and both green and pollution intensities become continua between 0 and 1.

### Aggregation and weighting conventions
- Uniform within-6-digit assumption: employees assumed uniformly distributed across 8-digit sub-occupations within each 6-digit occupation (Appendix B in Vona et al.,2018). Note: 83.3 percent of workers are in occupations without an 8-digit sub-category.
- Employment-weighted measures used when collapsing occupation-level data to coarser levels (e.g., 5-digit occupations, industry-level, or alternative occupation codes).
- Time-series comparisons treated cautiously due to SOC classification changes over time; 2002-2003 excluded from some charts because occupational codes for those years are not as detailed and seem to shift state-level employment-weighted green intensities (but not visibly for pollution-intensities).

### Data sources and estimation approach (overview across sections 2.2–2.6)
- State/National and Industry-level employment
  - Combine occupation definitions with U.S. Bureau of Labor Statistics Occupational Employment Wage Statistics (OEWS) state/national-industry-occupation employment.
  - OEWS provides occupations at SOC 6-digit and industries at NAICS 4- to 6-digit levels.
  - After 2012, state-industry-occupation data available from OEWS Research Estimates; prior to 2012, national-industry-occupation information is used.
  - State-occupation level data for 2004-2019 from OEWS used for time-series charts in Section 3. OEWS is repeated cross-sections not a panel.

- County/Commuting Zone and industry-level estimates
  - County employment from Eckert et al.,2020 (harmonized and censored observations filled from U.S. Census County Business Patterns).
  - Industries disaggregated at NAICS 6-digit and aggregated to 5-digit NAICS for this analysis.
  - Assumption: occupational breakdown of State-industry-year (post-2012) and National-industry-year (pre-2012) observations are representative for all counties in the same State (post-2012) or the whole country (pre-2012) in the same industry for the same year.
  - County-level data converted into 2000 census commuting zones.

- County Business Patterns (CBP)
  - Complement employment with CBP data for number of establishments, annual payrolls, and number of establishments divided by twelve employment size buckets.
  - CBP measures used are independent of green/polluting job definitions.

- Current Population Survey (CPS)
  - IPUMS U.S. CPS March Supplements between 1998 and 2019.
  - CPS March Supplements survey around 190,000 individuals on average each year and contain demographic, labor market, and earnings information.
  - Hourly wages inferred from annual earnings and usual hours worked per week and deflated by U.S. Consumer Price Index for All Urban Consumers (CPI-U) to 2015 constant U.S. dollars.
  - Merge procedure: 6-digit SOC 2010 occupation code to CPS unharmonized occupation code using BLS concordance; then merge unharmonized 2010 CPS codes to harmonized IPUMS occ2010 codes.
  - CPS allows study of demographics of holders of green- and pollution-intensive jobs, prevalence across income and age groups, wage premium comparisons, and transitions between pollution-intensive and green-intensive jobs (CPS re-samples workers after eight months).

- Clean Air Act (CAA) and regulatory data
  - CAA enforces national ambient air quality standards (NAAQS) for six criteria pollutants: sulphur dioxide (SO2), particulates (TSP, PM2.5, and PM10), nitrogen dioxide (NO2), carbon monoxide (CO), ozone (O3), and lead (Pb).
  - EPA labels a county “out of attainment” if one or more pollutants exceed thresholds; states must adopt state implementation plans (SIPs); EPA can impose sanctions such as withholding federal grants.
  - Data combined:
    - Integrated Compliance Information System (ICIS) database for plant-level regulatory program and pollutant permit information (plants labeled regulated if they record Title V Permit, SIP Source, SIP Source under federal jurisdiction, PSD permit, NSR permit, or NSPS permit in “Air Program Code” per Walker (2013)).
    - EPA panel data recording non-attainment designation over time at the county level.
  - Combining ICIS sector information with EPA county non-attainment panels produces a county-industry panel indicating which industry in which county is not meeting NAAQS at each point in time.

### Key empirical findings from Section 3 (Anatomy of Green- and Pollution-intensive Jobs)
- National and cross-state trends (2004-2019)
  - The share of green jobs has not materially grown over time at the national level, though cross-state dispersion has risen following the global financial crisis and remained fairly steady thereafter.
  - Pollution-intensive jobs show a slight decline as a share of total national employment; the national share is on a slight downward trend, while the median across states is more stable.
  - Bars in Figure 1 show the range (min-max) of state shares has varied over time.

- Industry versus within-industry variation
  - Polluting jobs vary more between industries than within industries across states.
  - Green jobs show significant within-industry variation across states; the variation across states of green job intensity for the same industry is almost as high as the between-industry variation for each state.
  - Implication: (1) green jobs are relatively more dispersed across industries than polluting jobs; (2) there may be opportunities to increase share of green jobs without large sectoral shifts within a state, unlike for polluting jobs where particular industries dominate.

- Geographic concentration and overlap
  - Regional patterns (2016):
    - West, Southwest, and parts of the Midwest have important concentrations of green-intensive jobs (notable industries: research and development, engineering services, aerospace manufacturing).
    - Southeast and Southwest have prevalence of higher pollution-intensive jobs (including extractive industries, electric power generation/transmission/distribution, wood, and textile industries).
  - At the commuting zone level, green and pollution-intensive jobs tend to be located close together.
    - For mapping in Section 3, “green-rich” and “polluting-rich” regions defined as top 25th percentile in their respective intensities.
    - 72 percent of commuting zones that are in the top 25th percentile for pollution-intensive jobs are either also in the top 25th percentile for green jobs or border a commuting zone that is.

- Interpretation for worker mobility and transitions
  - Spatial overlap of green- and pollution-intensive jobs suggests geographic mobility may not be the principal friction preventing workers in pollution-intensive industries from reallocating to greener jobs.
  - Household and CPS data are used later to examine whether worker characteristics (skills, wages, demographics) create mismatches that impede transitions.

### Distributional facts on CAA shocks
- Shock aggregation uses employment in t−2 to avoid mechanical relationships when relating shocks in t to employment changes between t−1 and t+h.
- Out of 624 NAICS 5-digit industries in the dataset, 426 (close to 70%) were affected in at least one commuting zone at least at one point in time.
- Out of the 50 states plus District of Columbia, 46 (over 90%) were affected in at least one industry at least at one point in time.

### Household-level Evidence (Section 3.2)
- Demographic patterns of green, polluting, and neutral jobs
  - High-skilled workers (defined as educational attainment level with some college or above) have, on average, more green intensive jobs than low-skilled workers.
  - Routine jobs tend to have both higher green- and higher pollution-intensity than non-routine jobs; however, workers with non-routine occupations work proportionally more in greener jobs.
  - Urban workers tend to have higher green intensity and lower pollution intensity than rural workers.
  - Neutral jobs are well distributed across skill and urban/rural status.
  - Regression controlling for all demographic characteristics finds: high-skilled, urban, and less automation-vulnerable workers tend to have greener jobs.

- Age and life-cycle patterns
  - Share of workers holding neutral jobs exhibits a slight U-shape over the life cycle, with the highest concentration of neutral jobs among the 16-19 years old group.
  - Green and pollution intensities display an inverted U-shape over the life cycle.

- Income patterns
  - Green intensity of jobs increases across income deciles.
  - Pollution intensity has an inverted-U relationship with income; the average pollution intensity score declines at the top few deciles of the income distribution.

- Wage differentials: green vs. polluting jobs
  - Mincer-type regression (annual, 1998–2019) finds:
    - The wage premium of green vs. polluting jobs is estimated to be around 2 percent and has trended modestly upward over time.

- Transition probabilities and wage implications of occupational switches
  - Workers who previously held a green-intensive job and change jobs have a more than 40 percent chance of moving to a new green-intensive job.
  - Workers who previously held pollution-intensive jobs have around a 15 percent chance of moving to a green job when they switch jobs.
  - Workers who previously held neutral jobs have around a 10 percent chance of moving to a green job when they switch jobs.
  - Workers who held pollution-intensive jobs can transition to neutral jobs with a probability of around 60 percent.

- Earnings dynamics for job-switchers (employed two consecutive years, EE) — key coefficients from Table 1 (changes in log real hourly wage ∆lnW_t)
  - OccSwitch × Green_{t−1}: -0.0528 ∗∗∗ (col 1); -0.0429 ∗∗∗ (col 2); -0.0524 ∗∗∗ (col 3); -0.0427 ∗∗∗ (col 4).
  - OccSwitch × Polluting_{t−1}: -0.00637 (col 1); 0.00464 (col 2); -0.00749 (col 3); 0.00395 (col 4).
  - OccSwitch: 0.0100 ∗∗∗ (col 1); 0.000446 (col 2); 0.0102 ∗∗∗ (col 3); 0.000602 (col 4).
  - Constant: 0.0580 ∗∗∗ (col 1); 0.453 ∗∗∗ (col 2); 0.101 ∗∗∗ (col 3); 0.490 ∗∗∗ (col 4).
  - Observations: 349,128 (cols 1 and 3) and 346,865 (cols 2 and 4).
  - Interpretation:
    - There is a wage penalty for workers who previously held green jobs when they switch occupations (“OccSwitch × Green_{t−1}”), robust across specifications.
    - No statistically significant wage penalty is found for those who previously held polluting jobs (“OccSwitch × Polluting_{t−1}”).
    - For those who held neutral jobs in the previous year, an occupational switch is associated with a real hourly wage increase in some specifications.

- Earnings dynamics by direction of environmental change (EE) — key coefficients from Table 2 (changes in log real hourly wage ∆lnW_t)
  - Greener: 0.0256 ∗∗∗ (col 1); 0.0299 ∗∗∗ (col 2); 0.0262 ∗∗∗ (col 3); 0.0305 ∗∗∗ (col 4).
  - LessGreen: -0.0542 ∗∗∗ (col 1); -0.0496 ∗∗∗ (col 2); -0.0538 ∗∗∗ (col 3); -0.0493 ∗∗∗ (col 4).
  - MorePolluting: 0.0507 ∗∗∗ (col 1); 0.0561 ∗∗∗ (col 2); 0.0497 ∗∗∗ (col 3); 0.0555 ∗∗∗ (col 4).
  - LessPolluting: -0.00888 (col 1); -0.00296 (col 2); -0.0100 (col 3); -0.00367 (col 4).
  - Constant: 0.0604 ∗∗∗ (col 1); 0.454 ∗∗∗ (col 2); 0.103 ∗∗∗ (col 3); 0.491 ∗∗∗ (col 4).
  - Interpretation:
    - Workers who move to less green jobs experience a real wage loss relative to those who did not switch occupations or whose jobs did not change in green/pollution intensity.
    - Workers who switch to less polluting jobs do not experience a similarly significant wage loss.
    - Workers who move to greener jobs experience a real wage gain; those who move to more polluting jobs also experience real wage gains.

- Earnings dynamics for workers re-employed after unemployment (UE) — key coefficients from Table 3 (ln(W_t / W_hat_{t−1}))
  - OccSwitch × Green_{t−1}: -0.0850 ∗∗∗ (col 1); -0.0831 ∗∗∗ (col 2); -0.0816 ∗∗∗ (col 3); -0.0808 ∗∗∗ (col 4).
  - OccSwitch × Polluting_{t−1}: -0.0361 (col 1); -0.0630 (col 2); -0.0323 (col 3); -0.0576 (col 4).
  - OccSwitch: -0.175 ∗∗∗ (col 1); -0.191 ∗∗∗ (col 2); -0.179 ∗∗∗ (col 3); -0.195 ∗∗∗ (col 4).
  - Constant: -0.375 ∗∗∗ (col 1); 0.0213 (col 2); -0.362 ∗∗∗ (col 3); -0.0168 (col 4).
  - Observations: 9,418 (all columns).
  - Interpretation:
    - Re-employed workers who previously held green jobs and switch occupations earn a lower wage when returning to work.
    - Workers who previously held polluting or neutral jobs do not experience significant wage losses upon switching after unemployment.
    - Occupational switches via unemployment are generally associated with additional wage losses (OccSwitch strongly negative).

- Summary of household-level evidence
  - On average, a worker who previously held a green job experiences a wage loss when changing occupations; this is mainly driven by moves into less-green jobs.
  - A worker who previously held a polluting job does not, on average, experience a wage loss upon switching occupations.
  - These patterns hold for both continuously employed workers (EE) and workers who found work after an unemployment spell (UE).

### Employment Effects of Environmental Regulation: Baseline (Section 4.2)
- Methodology
  - Local projection regression a la Jordà (2005):
    - Y_{l,i,t+h} = α_h + β_h * shock_{l,i,t} + δ^h_{l,t} + γ^h_{i,t} + ε_{l,i,t+h}
  - Y_{l,i,t+h} is change in (i) log employment or (ii) log number of establishments between t−1 and t+h.
  - Horizons h = 1,...,10 (up to 10 years ahead).
  - shock_{l,i,t} is regulatory affectedness of industry i in commuting zone l at time t.
  - Controls: commuting zone × time fixed effects and industry × time fixed effects.
  - Regressions run unweighted and weighted by average employment in the commuting zone × industry.
  - Standard errors clustered at the industry (NAICS 5-digit) level.

- Main employment findings (industry × commuting zone level)
  - Employment in the affected industry decreases significantly in response to environmental regulation.
  - Timing and magnitude:
    - The effect is most significant in the first years; confidence bands widen in later years due to declining observations.
    - After two years, the coefficient is at approximately -0.02%. Interpretation: if 100% of employment in a particular industry in a particular commuting zone were affected today, we would expect a decline in employment of 2% after two years in that industry and commuting zone.
  - Weighted vs. unweighted:
    - Effects are of similar magnitude for weighted regressions but with wider confidence bands.
    - Wider bands in weighted regressions could suggest the significant effect is driven by relatively smaller industry × commuting zone combinations.

- Findings on number of establishments
  - The number of establishments in the affected industry decreases significantly in response to environmental regulation.
  - Effect is only significant for the weighted regressions.
  - Suggests the negative effect on number of establishments is driven by industry-regions with a large number of employees.

- Interaction with Green and Polluting Job Intensity
  - Baseline regression augmented with interactions: shock × green_{l,i,t−1} and shock × poll_{l,i,t−1} where dummies indicate above-median green or pollution intensity in t−1.
  - Differential effects:
    - Industries with high green-job intensity show positive effects of environmental regulation on employment.
    - Industries with high polluting-job intensity show negative effects.
    - At the ten-year horizon, total employment increased in commuting zones and industries that already had a large share of green jobs before the shock, and decreased in areas with a large share of polluting jobs.

- Spillovers (commuting zone level)
  - Commuting zone regression: Y_{l,t+h} = α_h + β_h * shock_{l,t} + δ^h_t + γ^h_l + ε_{l,t+h}
  - Controls: commuting zone and time fixed effects; standard errors clustered at commuting zone level.
  - Findings:
    - Effect on total employment at the commuting zone level is insignificant.
    - Suggests employees who lost industry jobs due to the Clean Air Act were able to find work in other industries within their commuting zone.
    - Both unweighted and weighted regressions show mostly insignificant effects and an even positive effect in the last 2 years of the horizon.

- Effects on pay
  - Dependent variable: log of average payroll per employee (total payroll divided by number of employees).
  - Results:
    - Average payroll does not fall in response to environmental regulation at the industry-commuting zone level or at the commuting zone level over horizons h = 1,...,10.
    - Appendix note: total payroll decreases in affected industries, at least in the first three years, which is expected because employment decreases while average payroll is unchanged.

- Robustness and interpretation notes
  - Confidence bands widen in later years due to declining observations across horizons.
  - Weighted regressions often have wider confidence bands; some significant effects (e.g., number of establishments) are driven by industry-regions with large employment.
  - The observed labor market flexibility implies limited adverse effects on average pay and commuting-zone-level employment, but distributional and localized effects may still be significant and could require federal or state-level support.

*Source: Excerpted content from the provided IMF Working Paper sections (2.1, 3.2, and 4.2) in the supplied PDF.*

### 2.1  Defining Green- and Pollution-intensive Occupations or Jobs

### 2.1  Defining Green- and Pollution-intensive Occupations or Jobs

### Definitions and measurement
- Green-intensive Jobs
  - Based on O*NET’s Green Task Development Project taxonomy at Standard Occupational Classification (SOC) 8-digit level (Center,2010 and Center,2021).
  - O*NET distinguishes three types of green tasks: (i) existing occupations expected to be in high demand due to the greening of the economy; (ii) occupations expected to undergo significant changes in task content due to the greening of the economy (green-enhanced); and (iii) new occupations in the green economy.
  - Vona et al.,2018 compute a ratio of green tasks to total tasks as a measure of green intensity (the “greenness ratio” defined in their equation 1).
  - Employees are assumed uniformly distributed across 8-digit occupations within each 6-digit SOC occupation to obtain green intensity at the 6-digit level (because 6-digit is the maximum disaggregation available in publicly-available employment data from the Bureau of Labor Statistics).
  - Green intensity at the 6-digit occupation level is a continuum between 0 and 1.

- Pollution-intensive Jobs
  - Polluting occupations are defined as those prevalent in polluting industries.
  - Polluting industries: a set of 62 4-digit NAICS industries that are in the 95th percentile of pollution intensity for at least 3 pollutants among CO2, CO, VOC, NOx, SO2, PM10, PM2.5, and Lead.
  - Pollution-intensive occupations are defined as those with a 7 times higher probability of working in polluting sectors than in any other job.
  - At the 6-digit occupation level pollution intensity is occupation-based and takes only the value of 0 or 1.

- Important distinction
  - Green intensity is a share of green tasks (continuous between 0 and 1 at 6-digit level); pollution intensity is binary at the 6-digit occupation level.
  - When collapsing occupations (e.g., to 5-digit level, mapping SOC vintages, or converting to alternative occupation codes), employment-weighted measures are generated and both green and pollution intensities become continua between 0 and 1. Interpretations and magnitude comparisons between green and pollution intensities should be made with caution.

### Aggregation and weighting conventions
- Uniform within-6-digit assumption: employees assumed uniformly distributed across 8-digit sub-occupations within each 6-digit occupation (see Appendix B in Vona et al.,2018). Note: 83.3 percent of workers are in occupations without an 8-digit sub-category.
- Employment-weighted measures are used when collapsing occupation-level data to coarser levels (e.g., 5-digit occupations, industry-level, or alternative occupation codes).
- Time-series comparisons should be treated cautiously due to SOC classification changes over time; 2002-2003 are excluded from some charts because occupational codes for those years are not as detailed and seem to shift state-level employment-weighted green intensities (but not visibly for pollution-intensities).

### Data sources and estimation approach (overview across sections 2.2–2.6)
- State/National and Industry-level employment
  - Combine the occupation definitions (green and pollution) with U.S. Bureau of Labor Statistics Occupational Employment Wage Statistics (OEWS) state/national-industry-occupation employment.
  - OEWS provides occupations at SOC 6-digit and industries at NAICS 4- to 6-digit levels.
  - After 2012, state-industry-occupation data available from OEWS Research Estimates; prior to 2012, national-industry-occupation information is used.
  - State-occupation level data for 2004-2019 from OEWS used for time-series charts in Section 3. OEWS is repeated cross-sections not a panel.

- County/Commuting Zone and industry-level estimates
  - County employment from Eckert et al.,2020 (harmonized and censored observations filled from U.S. Census County Business Patterns).
  - Industries disaggregated at NAICS 6-digit and aggregated to 5-digit NAICS for this analysis.
  - Assumption: occupational breakdown of State-industry-year (post-2012) and National-industry-year (pre-2012) observations are representative for all counties in the same State (post-2012) or the whole country (pre-2012) in the same industry for the same year.
  - County-level data converted into 2000 census commuting zones.

- County Business Patterns (CBP)
  - Complement employment with CBP data for number of establishments, annual payrolls, and number of establishments divided by twelve employment size buckets.
  - CBP measures used are independent of green/polluting job definitions.

- Current Population Survey (CPS)
  - IPUMS U.S. CPS March Supplements between 1998 and 2019.
  - CPS March Supplements survey around 190,000 individuals on average each year and contain demographic, labor market, and earnings information.
  - Hourly wages inferred from annual earnings and usual hours worked per week and deflated by U.S. Consumer Price Index for All Urban Consumers (CPI-U) to 2015 constant U.S. dollars.
  - Merge procedure: 6-digit SOC 2010 occupation code to CPS unharmonized occupation code using BLS concordance; then merge unharmonized 2010 CPS codes to harmonized IPUMS occ2010 codes.
  - CPS allows study of demographics of holders of green- and pollution-intensive jobs, prevalence across income and age groups, wage premium comparisons, and transitions between pollution-intensive and green-intensive jobs (CPS re-samples workers after eight months).

- Clean Air Act (CAA) and regulatory data
  - CAA enforces national ambient air quality standards (NAAQS) for six criteria pollutants: sulphur dioxide (SO2), particulates (TSP, PM2.5, and PM10), nitrogen dioxide (NO2), carbon monoxide (CO), ozone (O3), and lead (Pb).
  - EPA labels a county “out of attainment” if one or more pollutants exceed thresholds; states must adopt state implementation plans (SIPs); EPA can impose sanctions such as withholding federal grants.
  - Data combined:
    - Integrated Compliance Information System (ICIS) database for plant-level regulatory program and pollutant permit information (label plants regulated if they record Title V Permit, SIP Source, SIP Source under federal jurisdiction, PSD permit, NSR permit, or NSPS permit in “Air Program Code” per Walker (2013)).
    - EPA panel data recording non-attainment designation over time at the county level.
  - Combining ICIS sector information with EPA county non-attainment panels produces a county-industry panel indicating which industry in which county is not meeting NAAQS at each point in time.

### Key empirical findings from Section 3 (Anatomy of Green- and Pollution-intensive Jobs)
- National and cross-state trends (2004-2019)
  - The share of green jobs has not materially grown over time at the national level, though cross-state dispersion has risen following the global financial crisis and remained fairly steady thereafter.
  - Pollution-intensive jobs show a slight decline as a share of total national employment; the national share is on a slight downward trend, while the median across states is more stable.
  - Bars in Figure 1 show the range (min-max) of state shares has varied over time.

- Industry versus within-industry variation
  - Polluting jobs vary more between industries than within industries across states.
  - Green jobs show significant within-industry variation across states; the variation across states of green job intensity for the same industry is almost as high as the between-industry variation for each state.
  - Implication: (1) green jobs are relatively more dispersed across industries than polluting jobs; (2) there may be opportunities to increase share of green jobs without large sectoral shifts within a state, unlike for polluting jobs where particular industries dominate.

- Geographic concentration and overlap
  - Regional patterns (2016):
    - West, Southwest, and parts of the Midwest have important concentrations of green-intensive jobs (notable industries: research and development, engineering services, aerospace manufacturing).
    - Southeast and Southwest have prevalence of higher pollution-intensive jobs (including extractive industries, electric power generation/transmission/distribution, wood, and textile industries).
  - At the commuting zone level, green and pollution-intensive jobs tend to be located close together.
    - For mapping in Section 3, “green-rich” and “polluting-rich” regions defined as top 25th percentile in their respective intensities.
    - 72 percent of commuting zones that are in the top 25th percentile for pollution-intensive jobs are either also in the top 25th percentile for green jobs or border a commuting zone that is.

- Interpretation for worker mobility and transitions
  - Spatial overlap of green- and pollution-intensive jobs suggests geographic mobility may not be the principal friction preventing workers in pollution-intensive industries from reallocating to greener jobs.
  - Household and CPS data are used later to examine whether worker characteristics (skills, wages, demographics) create mismatches that impede transitions.

*Source: Excerpt from “2.1  Defining Green- and Pollution-intensive Occupations or Jobs” and related sections in the provided content.*

### 3.2  Household-level Evidence

### 3.2 Household-level Evidence

### Demographic patterns of green, polluting, and neutral jobs
- High-skilled workers (defined as educational attainment level with some college or above) have, on average, more green intensive jobs than low-skilled workers.
- Routine jobs tend to have both higher green- and higher pollution-intensity than non-routine jobs; however, workers with non-routine occupations work proportionally more in greener jobs.
- Urban workers tend to have higher green intensity and lower pollution intensity than rural workers.
- Neutral jobs are well distributed across skill and urban/rural status.
- Regression controlling for all demographic characteristics finds: high-skilled, urban, and less automation-vulnerable workers tend to have greener jobs.

### Age and life-cycle patterns
- Share of workers holding neutral jobs exhibits a slight U-shape over the life cycle, with the highest concentration of neutral jobs among the 16-19 years old group.
- Green and pollution intensities display an inverted U-shape over the life cycle.
- These patterns contradict the prior that green jobs are predominantly held by young workers and that older-worker retirements will naturally accomplish the green transition.

### Income patterns
- Green intensity of jobs increases across income deciles: higher income groups tend to hold greener jobs.
- Pollution intensity has an inverted-U relationship with income; the average pollution intensity score declines at the top few deciles of the income distribution.

### Wage differentials: green vs. polluting jobs
- A Mincer-type regression (annual, 1998–2019) with dependent variable log real hourly wage and controls for high-skilled dummy, age, age squared, and urban dummy finds:
  - The wage premium of green vs. polluting jobs is estimated to be around 2 percent and has trended modestly upward over time.
  - The observed premium could reflect other unobserved characteristics of green jobs (e.g., specific skills) not fully controlled for in the Mincer regressions.

### Transition probabilities and wage implications of occupational switches
- Transition probabilities by original occupation type:
  - Workers who previously held a green-intensive job and change jobs have a more than 40 percent chance of moving to a new green-intensive job.
  - Workers who previously held pollution-intensive jobs have around a 15 percent chance of moving to a green job when they switch jobs.
  - Workers who previously held neutral jobs have around a 10 percent chance of moving to a green job when they switch jobs.
  - Workers who held pollution-intensive jobs can transition to neutral jobs with a probability of around 60 percent.
- The CPS limitation: CPS does not collect the last wage of those that reported being unemployed in the previous year and does not track workers beyond two consecutive years; CPS does ask about previous occupation held by unemployed persons.

### Earnings dynamics for job-switchers (employed two consecutive years, EE)
- Key estimates from Table 1 (changes in log real hourly wage ∆lnW_t for workers employed in two consecutive years):
  - OccSwitch × Green_{t−1}: -0.0528 ∗∗∗ (col 1); -0.0429 ∗∗∗ (col 2); -0.0524 ∗∗∗ (col 3); -0.0427 ∗∗∗ (col 4).
  - OccSwitch × Polluting_{t−1}: -0.00637 (col 1); 0.00464 (col 2); -0.00749 (col 3); 0.00395 (col 4).
  - OccSwitch: 0.0100 ∗∗∗ (col 1); 0.000446 (col 2); 0.0102 ∗∗∗ (col 3); 0.000602 (col 4).
  - Constant: 0.0580 ∗∗∗ (col 1); 0.453 ∗∗∗ (col 2); 0.101 ∗∗∗ (col 3); 0.490 ∗∗∗ (col 4).
  - Observations: 349,128 (cols 1 and 3) and 346,865 (cols 2 and 4).
- Interpretation:
  - There is a wage penalty for workers who previously held green jobs when they switch occupations (“OccSwitch × Green_{t−1}”), robust across specifications.
  - No statistically significant wage penalty is found for those who previously held polluting jobs (“OccSwitch × Polluting_{t−1}”).
  - For those who held neutral jobs in the previous year, an occupational switch is associated with a real hourly wage increase (OccSwitch positive in some columns).

### Earnings dynamics by direction of environmental change (EE)
- Key estimates from Table 2 (changes in log real hourly wage ∆lnW_t based on changes in environmental properties):
  - Greener: 0.0256 ∗∗∗ (col 1); 0.0299 ∗∗∗ (col 2); 0.0262 ∗∗∗ (col 3); 0.0305 ∗∗∗ (col 4).
  - LessGreen: -0.0542 ∗∗∗ (col 1); -0.0496 ∗∗∗ (col 2); -0.0538 ∗∗∗ (col 3); -0.0493 ∗∗∗ (col 4).
  - MorePolluting: 0.0507 ∗∗∗ (col 1); 0.0561 ∗∗∗ (col 2); 0.0497 ∗∗∗ (col 3); 0.0555 ∗∗∗ (col 4).
  - LessPolluting: -0.00888 (col 1); -0.00296 (col 2); -0.0100 (col 3); -0.00367 (col 4).
  - Constant: 0.0604 ∗∗∗ (col 1); 0.454 ∗∗∗ (col 2); 0.103 ∗∗∗ (col 3); 0.491 ∗∗∗ (col 4).
- Interpretation:
  - Workers who move to less green jobs experience a real wage loss relative to those who did not switch occupations or whose jobs did not change in green/pollution intensity.
  - Workers who switch to less polluting jobs do not experience a similarly significant wage loss.
  - Workers who move to greener jobs experience a real wage gain; those who move to more polluting jobs also experience real wage gains.

### Earnings dynamics for workers re-employed after unemployment (UE)
- Approach: For each occupation (IPUMS CPS occ2010 4-digit codes, >450 occupations), run Mincer-type wage regressions on employed sample; impute predicted wages for unemployed given past occupation and demographics; compare actual wage upon reemployment to predicted wage.
- Key estimates from Table 3 (ln(W_t / W_hat_{t−1}) for workers employed in t and unemployed in t−1):
  - OccSwitch × Green_{t−1}: -0.0850 ∗∗∗ (col 1); -0.0831 ∗∗∗ (col 2); -0.0816 ∗∗∗ (col 3); -0.0808 ∗∗∗ (col 4).
  - OccSwitch × Polluting_{t−1}: -0.0361 (col 1); -0.0630 (col 2); -0.0323 (col 3); -0.0576 (col 4).
  - OccSwitch: -0.175 ∗∗∗ (col 1); -0.191 ∗∗∗ (col 2); -0.179 ∗∗∗ (col 3); -0.195 ∗∗∗ (col 4).
  - Constant: -0.375 ∗∗∗ (col 1); 0.0213 (col 2); -0.362 ∗∗∗ (col 3); -0.0168 (col 4).
  - Observations: 9,418 (all columns).
- Interpretation:
  - Re-employed workers who previously held green jobs and switch occupations earn a lower wage when returning to work.
  - Workers who previously held polluting or neutral jobs do not experience significant wage losses upon switching after unemployment.
  - Occupational switches via unemployment are generally associated with additional wage losses (OccSwitch strongly negative).

### Summary of household-level evidence
- On average, a worker who previously held a green job experiences a wage loss when changing occupations; this is mainly driven by moves into less-green jobs.
- A worker who previously held a polluting job does not, on average, experience a wage loss upon switching occupations.
- These patterns hold for both continuously employed workers (EE) and workers who found work after an unemployment spell (UE).

### Defining Clean Air Act (CAA) shocks (for later local labor market analysis)
- Shock definition (equation as specified):
  shock_{l,i,t} = sum_{c ∈ l} NA_{c,i,t} * emp_{c,i,t−2} / emp_{l,i,t−2}
  - NA_{c,i,t} is a dummy for “non-attainment” in county c, industry i (NAICS 5-digit), and time t.
  - emp_{c,i,t−2} and emp_{l,i,t−2} are employment in industry i at t−2 in county c and in commuting zone l, respectively.
  - An industry is deemed affected if not attaining for at least one pollutant.
  - The shock is aggregated to commuting zone l; county-industry employment weights are used so county-industries that employ more workers carry greater weight.
  - Employment in t−2 is used to avoid mechanical relationships when relating shocks in t to employment changes between t−1 and t+h.
- Distributional facts:
  - Out of 624 NAICS 5-digit industries in the dataset, 426 (close to 70%) were affected in at least one commuting zone at least at one point in time.
  - Out of the 50 states plus District of Columbia, 46 (over 90%) were affected in at least one industry at least at one point in time.
  - Although only a few commuting zone-industry observations are affected at a given point in time, the degree to which local industries are affected is well distributed over time (share of commuting zone-industry affected shown for years 2000-2016).

*Source: U.S. CPS and Authors’ calculations (chapter 3.2).*

### 4.2  Employment Effects of Environmental Regulation: Baseline

### 4.2  Employment Effects of Environmental Regulation: Baseline

### Methodology
- Local projection regression a la Jordà (2005) estimated:
  - Y_{l,i,t+h} = α_h + β_h * shock_{l,i,t} + δ^h_{l,t} + γ^h_{i,t} + ε_{l,i,t+h}
  - Y_{l,i,t+h} is change in (i) log employment or (ii) log number of establishments between t−1 and t+h.
  - Horizons (h) explored up to 10 years ahead (h = 1,...,10).
  - shock_{l,i,t} is regulatory affectedness of industry i in commuting zone l at time t.
  - Controls: commuting zone × time fixed effects and industry × time fixed effects.
  - Regressions run unweighted and weighted by average employment in the commuting zone × industry.
  - Standard errors clustered at the industry (NAICS 5-digit) level.

### Main findings on employment (industry × commuting zone level)
- Employment in the affected industry decreases significantly in response to environmental regulation.
- Timing and magnitude:
  - The effect is most significant in the first years; confidence bands widen in later years due to declining observations.
  - After two years, the coefficient is at approximately -0.02%. Interpretation: if 100% of employment in a particular industry in a particular commuting zone were affected today, we would expect a decline in employment of 2% after two years in that industry and commuting zone.
- Weighted vs. unweighted:
  - Effects are of similar magnitude for weighted regressions but with wider confidence bands.
  - Wider bands in weighted regressions could suggest the significant effect is driven by relatively smaller industry × commuting zone combinations.
  - Possible interpretation: larger industry-region observations (with many employees) may be in larger commuting zones with more opportunities in other sectors and hence adjust employment more easily.

### Findings on number of establishments
- The number of establishments in the affected industry decreases significantly in response to environmental regulation.
- Results by weighting:
  - Effect is only significant for the weighted regressions.
  - Suggests the negative effect on number of establishments is driven by industry-regions with a large number of employees.
  - Intuition: larger industry-regions may include several establishments of the same firm, so firms can adjust by reducing the number of establishments.

### Interaction with Green and Polluting Job Intensity (overview)
- Baseline regression augmented to interact shock with green and pollution intensity dummies:
  - Y_{l,i,t+h} = α_h + β^1_h * shock_{l,i,t} + β^2_h * shock_{l,i,t} * green_{l,i,t−1} + β^3_h * shock_{l,i,t} * poll_{l,i,t−1} + δ^h_{l,t} + γ^h_{i,t} + χ_{l,i,t+h}
  - green_{l,i,t−1} and poll_{l,i,t−1} are dummies indicating whether an industry in a commuting zone was above the median of green or pollution intensity in t−1.
- Differential effects:
  - Industries with high green-job intensity show positive effects of environmental regulation on employment.
  - Industries with high polluting-job intensity show negative effects.
  - At the ten-year horizon, total employment increased in commuting zones and industries that already had a large share of green jobs before the shock, and decreased in areas with a large share of polluting jobs.
  - Plotted quantities: β̂_1 + β̂_2 (green) and β̂_1 + β̂_3 (polluting) over h = 1,...,10.

### Spillovers (commuting zone level)
- Regression at commuting zone level:
  - Y_{l,t+h} = α_h + β_h * shock_{l,t} + δ^h_t + γ^h_l + ε_{l,t+h}
  - Controls: commuting zone and time fixed effects; standard errors clustered at commuting zone level.
- Findings:
  - Effect on total employment at the commuting zone level is insignificant.
  - Suggests employees who lost industry jobs due to the Clean Air Act were able to find work in other industries within their commuting zone.
  - Both unweighted and weighted regressions show mostly insignificant effects and an even positive effect in the last 2 years of the horizon.
  - The small magnitude of the coefficient hints at economically negligible effects at the commuting zone level.

### Effects on pay
- Dependent variable: log of average payroll per employee (total payroll divided by number of employees) to capture wages plus non-wage benefits.
  - Total payroll includes salaries, wages, commissions, dismissal pay, bonuses, vacation allowances, sick-leave pay, employee contributions to qualified pension plans; includes deductions for social security, income tax, insurance, and union dues.
- Results:
  - Average payroll does not fall in response to environmental regulation at the industry-commuting zone level or at the commuting zone level over horizons h = 1,...,10.
  - Consistent with labor market reallocation: because workers find jobs in other industries, average payroll remains unaffected.
  - Appendix note: total payroll decreases in affected industries, at least in the first three years, which is expected because employment decreases while average payroll is unchanged.

### Robustness and interpretation notes
- Confidence bands widen in later years due to declining observations across horizons.
- Weighted regressions often have wider confidence bands, and some significant effects (e.g., number of establishments) are driven by industry-regions with large employment.
- The labor market flexibility observed implies limited adverse effects on average pay and commuting-zone-level employment, but distributional and localized effects may still be significant and could require federal or state-level support.

*Source: IMF Working Paper — section 4.2 and adjacent sections on employment effects, interactions with green/polluting intensity, spillovers, and pay.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022129-print-pdf.pdf_