Why is Pollution from U.S. Manufacturing Declining?

Transcription

1 Why is Pollution from U.S. Manufacturing Declining? The Roles of Trade, Regulation, Productivity, and Preferences Joseph S. Shapiro Reed Walker Yale University and NBER UC Berkeley and NBER CUNY Graduate Center February 2015

2 Why are Manufacturing Pollution Emissions Declining? 1990= Real Output CO NO x PM 2.5 PM 10 SO 2 VOCs Figure: Pollution Emissions from U.S. Manufacturing

3 Why are Manufacturing Pollution Emissions Declining? Potential explanations: Foreign competitiveness (Pierce and Schott 2012; Autor, Dorn, and Hanson 2013) Environmental regulation (Henderson 1996; Correia et al. 2013) Preferences (Levinson and O Brien 2013) Productivity (Bloom et al. 2010, Martin 2011) How distinguish empirically?

4 Why are Manufacturing Pollution Emissions Declining? Plant-Level Evidence for Productivity: Log CO Emissions Per Dollar Real Output CO Log Total Factor Productivity Log NOx Emissions Per Dollar Real Output NOx Log Total Factor Productivity Log PM2.5 Emissions Per Dollar Real Output PM Log Total Factor Productivity Log PM10 Emissions Per Dollar Real Output PM Log Total Factor Productivity Log SO2 Emissions Per Dollar Real Output SO Log Total Factor Productivity Log VOC Emissions Per Dollar Real Output VOC Log Total Factor Productivity

5 Why are Manufacturing Pollution Emissions Declining? This paper: Statistical decomposition Trade-environment model

6 Why are Manufacturing Pollution Emissions Declining? This paper: Statistical decomposition Trade-environment model Findings: Most pollution decrease is within narrowly-defined products Stringency of environmental regulation more than doubled Environmental regulation explains 75% or more of decline in pollution emissions Trade, productivity, preferences play limited role

7 Existing Research and Contributions What is new here? Trade & Environment (Grossman and Krueger 1995; Antweiler, Copeland, and Taylor 2001; Copeland and Taylor 2003; Levinson 2009; Forslid, Okubo, and Ultveit-Moe 2011) We structurally estimate a model of heterogeneous firms and endogenous pollution abatement Environmental regulation (Greenstone 2002; Ryan 2012; Walker 2013) We measure the change in all local and national environmental regulation (shadow price of pollution) Gravity models (Eaton and Kortum 2002; Melitz 2002; Dekle, Eaton, and Kortum 2007; Chaney 2008; Eaton, Kortum, Neiman, and Romalis 2011; Hsieh and Ossa 2011; Arkolakis, Costinot, and Rodriguez-Clare 2012; Eaton, Kortum, Neiman, and Romalis 2013; Shapiro 2013)

8 Overview Statistical Decomposition Trade-Environment Model Data Estimation and Results: Parameters and Shocks Counterfactuals Sensitivity Conclusion

9 Overview Statistical Decomposition Trade-Environment Model Data Estimation and Results: Parameters and Shocks Counterfactuals Sensitivity Conclusion

10 Statistical Decomposition: Background Builds on Levinson (2009) Standard decomposition: Scale: increase in real output Composition: shift in output from clean (e.g., furniture to steel) Technique: pollution per unit output Goals: Establish what fraction of pollution reductions come from scale, composition, and technique effects Clarify what we learn from model s stronger assumptions

11 Statistical Decomposition: Methodology Pollution summed across industries: Z = s z s = s x s e s = X s κ s e s In vector notation, Totally differentiating gives Z = Xκ e dz = κ } edx {{} Scale + Xe dκ }{{} Composition + Xκ de }{{} Technique

12 Statistical Decomposition: Data Data for statistical decomposition: National Emissions Inventory and Annual Survey of Manufactures (both 1990) Fuzzy string matching to create plant-level database Product-level information Apportion plant emissions to plant-product using product revenue shares

13 Statistical Decomposition: NO x 1990 = Scale (Census) Scale & Composition (NEI+Census) Scale, Composition, & Technique (NEI) Year Figure: Nitrogen Oxides Emissions from U.S. Manufacturing: Scale, Composition, and Technique Effects For ~1200 products defined in census microdata (e.g., carbon wire rods )

14 Statistical Decomposition: Criteria Pollutants 1990 = CO 1990 = NOx Year Year PM10 PM = = Year Year 1990 = SO = VOC Year Year Scale (Census) Scale & Composition (NEI+Census) Scale, Composition, & Technique (NEI)

15 Overview Statistical Decomposition Trade-Environment Model Data Estimation and Results: Parameters and Shocks Counterfactuals Sensitivity Conclusion

16 Trade-Environment Model Assumption 1: Consumers have CES Preferences Multiple sectors Assumption 2: Market structure is monopolistic competition Like Melitz (2003) but firms pay pollution taxes. Productivity distribution is Pareto Assumption 3: Pollution is a second output which is taxed Like Copeland and Taylor (2003) Equivalently, production is Cobb-Douglas in factors and in pollution Assumption 4: Competitive Equilibrium Lets us calculate counterfactual outcomes.

17 Trade-Environment Model: General Setup Representative agent One factor with inelastic supply ( labor )

18 Trade-Environment Model Assumption 1: Consumers have CES Preferences U d = s [ q od,s (ω) σs 1 σs dω o ω Ω o,s ] σs σs 1 β d,s Z δ d Multi-sector CES, pollution damages Z δ d Pollution a pure externality Assumption 2: Market structure is monopolistic competition Assumption 3: Production is Cobb-Douglas in pollution and factors Assumption 4: Competitive Equilibrium

19 Trade-Environment Model Assumption 1: Consumers have CES Preferences Assumption 2: Market structure is monopolistic competition π o,s (ϕ)= d π od,s (ϕ) w o f e o,s π od,s (ϕ)= p od,s (ϕ) q od,s (ϕ) w o l od,s (ϕ) τ od,s t o z od,s (ϕ) τ od,s w o f od,s G o,s (ϕ)= 1 (b o,s ) θs / (ϕ) θs Profits π od,s, pollution z od,s, pollution tax t o, Pareto productivity G o,s Assumption 3: Production is Cobb-Douglas in pollution and factors Assumption 4: Competitive Equilibrium

20 Trade-Environment Model Assumption 1: Consumers have CES Preferences Assumption 2: Market structure is monopolistic competition Assumption 3: Pollution z od,s = (1 ξ) 1/αs ϕl od,s All firms undertake some abatement. Equivalent: production is Cobb-Douglas in pollution and factors; abatement sector; potential output Assumption 4: Competitive Equilibrium

21 Trade-Environment Model Assumption 1: Consumers have CES Preferences Assumption 2: Market structure is monopolistic competition Assumption 3: Production is Cobb-Douglas in pollution and in factors Assumption 4: Competitive Equilibrium Labor market clearing: L o = L e o + L m o + L p o Utility maximization implies gravity λ od,s = Me o,s ( wo i Me i,s θs ) θs b o,s (τ od,s ) 1 αs (f od,s ) 1 (σs 1)(1 αs ) (t o ) ( wi θs ) θs b i,s (τ id,s ) 1 αs (f id,s ) 1 (1 αs )(σs 1) (t i ) θs θs αs θs 1 αs αs θs 1 αs

22 Trade-Environment Model: Equilibrium Conditions Labor market clearing Free entry condition + zero cutoff profit In changes Useful implication: change in pollution emissions

23 Trade-Environment Model: Equilibrium Conditions in Levels Labor market clearing: L d = s 1 (θ s+1 α s)(σ s 1) σ sθ s β d,s s M e d,sf e d,s (θ s + 1) Free entry condition + zero cutoff profit fo,s e σ s θ s (σ s 1) (1 α s ) = d (w o ) 1 (w o /b o,s ) θs (τ od,s ) θs 1 α (fod,s ) 1 (σs 1)(1 αs ) (t o ) i Me i,s (w i/b i,s ) θs (τ id,s ) θs 1 αs (f id,s ) 1 θs (σs 1)(1 αs ) (t i ) θs αs θs 1 αs E αs θs d,s 1 αs

24 Trade-Environment Model: Equilibrium Conditions in Changes Methodology (Dekle, Eaton, and Kortum 2007): ˆx x /x Labor market clearing: 1 = ψ o η s ˆM o,s e s Free entry condition + zero cutoff profit ŵ o = ( ) θs ) ŵ ζ o od,s (ˆτod,s ) θs 1 1 αs (ˆf ˆb od,s o,s ( ) d i λ id,s ˆM θs ) i,s e ŵ o (ˆτod,s ) θs 1 1 αs (ˆfod,s ˆb o,s θs (σs 1)(1 αs ) (ˆt o,s ) αs θs 1 αs θs (σs 1)(1 αs ) (ˆt o,s ) αs θs 1 αs ˆβ d,s ŵ d

25 Trade-Environment Model: Equilibrium Conditions in Changes Change in pollution emissions Ẑ o = s ˆM e o,s ŵ oˆt o,s Z o,s

26 Model Summary: Classes of Variables Data (X od,s, Z o,s ) Easy observed in year 1990 Parameters (σ s, θ s, α s ) Partial equilibrium relationships estimated from regressions Shocks (ˆτ od,s, ˆf od,s, ˆt o,s, ˆb o,s, ˆβ o,s ) Policies that we choose to define a counterfactual. Endogenous Variables (ŵ o, ˆM o,s ) Determined by interaction of supply and demand to achieve a competitive equilibrium

27 Trade-Environment Model: Comparative Statics Pollution per unit output ( pollution intensity ): z od,s = 1 ( wo q od,s ϕ 1 αs t o,s α s 1 α s ) 1 αs Plant-level comparative statics. Pollution per unit output lower for More productive plants (ϕ) More stringent environmental regulation (t o,s )

28 Overview Statistical Decomposition Trade-Environment Model Data Estimation and Results: Parameters and Shocks Counterfactuals Sensitivity Conclusion

29 Data Plant-level Microdata 1990 and 2005 Annual Survey of Manufactures Value of shipments, inventory-adjusted Payments for factors and intermediates Industry-year output and materials deflators 60,000 plants/year US National Emissions Inventory Plant-level pollution emissions from every US source Main pollutants: CO, PM10, PM2.5, NOx, SO2, VOCs Pollution Abatement Costs and Expenditures Survey (PACE) Reported expenditures on air pollution Capital expenditures

30 Data US industry-year aggregates National Emissions Inventory 1990, 1996, 1999, 2002, 2005, 2008 International country-industry-year aggregates: OECD STAN Gross output and international trade 26 countries, 17 industries (2-digit ISIC Rev. 3) Aggregate to 2 countries: US and Foreign

31 Data: Sectors Code Description ISIC Rev. 3 Codes 1 Food, beverages, tobacco Textiles, apparel, fur, leather Wood products 20 4 Paper and publishing Coke, refined petroleum, nuclear fuel 23 6 Chemicals 24 7 Rubber and plastics 25 8 Other non-metallic minerals 26 9 Basic metals Fabricated metals Machinery and equipment Office, accounting, computing, and electrical machinery Radio, television, communication equipment Medical, precision, and optical, watches, clocks Motor vehicles, trailers Other transport equipment Furniture, manufactures n.e.c., recycling 36-37

32 Overview Statistical Decomposition Trade-Environment Model Data Estimation and Results: Parameters and Shocks Counterfactuals Sensitivity Conclusion

33 Estimates and Results: Parameters and Shocks Parameters Pollution elasticity Elasticity of substitution Productivity dispersion Counterfactual shocks: Foreign competitiveness Domestic competitiveness Environmental regulation Consumer preferences

34 Estimates and Results: Pollution Elasticity Pollution elasticity α: Estimating equation: ( zi,t ) ln q i,t z q = (1 ξ)(1 α)/α = 1 α α ln(1 ξ i,t) + η t + ɛ i,t Instrument 1 ξ with nonattainment designations. Rationale: reverse causality.

35 Estimates and Results: Pollution Elasticity CO NO x (O 3 ) PM 10 PM 2.5 VOC (O 3 ) Total (Any) Panel A: First Stage Nonattain cp Polluter p (0.015) (0.011) (0.085) (0.068) (0.009) (0.009) Panel B: Reduced Form Nonattain cp Polluter p (5.244) (4.782) (6.860) (4.427) (1.214) (1.979) Panel C: Instrumental Variables Abatement Expenditure Ratio (64.278) (72.412) (78.483) (46.795) (36.827) (25.373) N First Stage F-Stat Panel D: Pollution Elasticity Parameter Pollution Elasticity (α) (0.004) (0.007) (0.009) (0.013) (0.002) (0.003) County-NAICS FE X X X X X X

36 Estimates and Results: Macro Parameters Elasticity of Substitution σ s : w o L p o,s = (1 α s ) σ s 1 X o,s σ s Pareto shape parameter θ s : ln(pr{x > X i,s }) = γ 0,s + γ 1,s ln(x i,s ) + ɛ i,s θ s = γ 1,s (1 σ s )

37 Estimates and Results: Macro Parameters Industry Elasticity of Pareto Shape Shape Parameter Substitution Parameter Standard Error Food, Beverages, Tobacco (0.13) Textiles, Apparel, Fur, Leather (0.10) Wood Products (0.17) Paper and Publishing (0.10) Coke, Refined Petroleum, Fuels (1.67) Chemicals (0.08) Rubber and Plastics (0.08) Other Non-metallic Minerals (0.11) Basic Metals (0.50) Fabricated Metals (0.06) Machinery and Equipment (0.14) Office, Computing, Electrical (0.15) Radio, Television, Communication (0.23) Medical, Precision, and Optical (0.06) Motor Vehicles, Trailers (0.18) Other Transport Equipment (0.13) Furniture, Other, Recycling (0.03) Mean Across Industries (0.23)

38 Estimates and Results: Shocks Need actual, historic values Parameters and data all we need to analyze counterfactuals But we want to analyze a specific counterfactual What if one shock followed its actual, historic path but other shocks stayed fixed at 1990 values? This requires knowing the actual, historic path of each shock How did trade costs, competitiveness, environmental regulation evolve ? In principle, could use data on tariffs, shipping costs, announcements of new environmental regulation, etc. to investigate this Instead, we use the model itself to infer historic values

39 Estimates and Results: Shocks Gravity equation in changes (1-sector version) ( ) θs ˆλ od,s = ˆM o,s e ŵo (ˆτ od,s) θs 1 αs ˆb o,s (ˆfod,s ) 1 θs (σs 1)(1 αs ) (ˆt o,s ) αs θs 1 αs Invert it to define a shock: (ˆτ o,d ) θ 1 α ) 1 θ (ˆf (ˆb ) (σ 1)(1 α) θ (ŵ o ) od o = ˆλ θ od ˆM o (ˆt o ) αθ 1 α

40 Estimates and Results: Shocks Definition of foreign competitiveness shock: ˆΓ od,s (1/ˆb o,s ) θs (ˆτ od,s ) θs/(1 αs) (ˆf od,s ) 1 θs/(σs 1)(1 αs) (ˆt ) αsθs/(1 α s) o,s for o U.S Measurement of foreign competitiveness shock: ˆΓ od,s = ˆλ od,s ˆM e o,s (ŵ o ) θs (ˆPd,s ) ( θs 1 αs ŵ d w d L d ˆβ NX ) 1 θs (σs 1)(1 αs ) d NX d d,s w d L d NX d

41 Estimates and Results: Shocks Definition of U.S. competitiveness shock: ˆΓ od,s ( 1/ˆb o,s ) θs (ˆτ od,s ) θs/(1 αs) (ˆf od,s ) 1 θs/(σs 1)(1 αs) for o = U.S. Measurement of U.S. competitiveness shock: ˆΓ od,s = (ˆt o,s ) αs θs 1 αs ˆλ od,s ˆM e o,s (ŵ o ) θs (ˆP d,s ) ( θs 1 αs ŵ d w d L d ˆβ NX ) 1 θs (σs 1)(1 αs ) d NX d d,s w d L d NX d

42 Estimates and Results: Shocks Preference Shock: ˆβ d,s = o X od,s / o,s X od,s o X od,s/ o,s X od,s Pollution regulation shock: ˆt o,s = ŵo ˆM e o,s Ẑ o,s

43 Estimates and Results: Historic Shocks, (a) Foreign Competitiveness (b) U.S. Competitiveness 400 Dirty Industries Clean Industries = = Year Year (c) Foreign Preferences (d) U.S. Preferences = = Year Year

44 Estimates and Results: Historic Shocks, Figure: U.S. Environmental Regulation = Year

45 Estimates and Results: Endogenous Variables (a) Foreign Wages (b) U.S. Wages = = Year (c) Foreign Firm Entry Year (d) U.S. Firm Entry = = Year Dirty Industries Clean Industries Year

46 Overview Statistical Decomposition Trade-Environment Model Data Estimation and Results: Parameters and Shocks Counterfactuals Sensitivity Conclusion

47 Counterfactuals: Algorithm Required data Data from 1990 (X od,s, Z o,s ), Parameter vectors (α s, σ s, θ s ) Three Step Algorithm 1 Define counterfactual: choose shocks {ˆΓ od,s, ˆt o,s, ˆβ o,s } 2 Find equilibrium: find changes to wages and firm entry (ŵ o, ˆM e o,s) which make equilibrium conditions hold with equality 3 Recover U.S. pollution emissions, given results of first two steps

48 Counterfactuals: Algorithm Counterfactuals we study One shock takes on historic values, others fixed at 1990 levels. Example counterfactual Foreign competitiveness follows its historical path, other shocks fixed at 1990: {ˆΓ od,s, ˆt o,s, ˆβ o,s } = { {ˆΓ od,s, 1, 1} if o U.S. {1, 1, 1} if o = U.S.

49 Counterfactuals: Results = Actual Data (All Shocks) Foreign Competitiveness Shocks Only Year Figure: Counterfactual U.S. Manufacturing NOx Emissions, Foreign Competitiveness Shocks Only

50 Counterfactuals: Results = Actual Data (All Shocks) U.S. Competitiveness Shocks Only Year Figure: Counterfactual U.S. Manufacturing NOx Emissions, U.S. Competitiveness Shocks Only

51 Counterfactuals: Results = Actual Data (All Shocks) U.S. Preference Shocks Only Year Figure: Counterfactual U.S. Manufacturing NOx Emissions, U.S. Preference Shocks Only

52 Counterfactuals: Results = Actual Data (All Shocks) U.S. Regulation Shocks Only Year Figure: Counterfactual U.S. Manufacturing NOx Emissions, U.S. Regulation Shocks Only

53 Counterfactuals: Results, by Pollutant 120 CO NO x PM = Actual Data (All Shocks) Foreign Competitiveness Shocks Only US Competitiveness Shocks Only 20 US Regulation Shocks Only US Preference Shocks Only Deficit Shocks Only Year 1990= Year 1990= Year 120 PM SO VOCs = = = Year Year Year Figure: Counterfactual U.S. Manufacturing Pollution Emissions Under Subsets of Shocks,

54 Overview Statistical Decomposition Trade-Environment Model Data Estimation and Results: Parameters and Shocks Counterfactuals Sensitivity Conclusion

55 Sensitivity 1: Role of Other Shocks 150 U.S. Pollution: 1990= Foreign Competitiveness: Dirty Industries Foreign Competitiveness: Clean Industries Proportional Change in Foreign Competitiveness

56 Sensitivity 1: Role of Other Shocks 150 U.S. Pollution: 1990= U.S. Competitiveness: Dirty Industries U.S. Competitiveness: Clean Industries Proportional Change in US Competitiveness

57 Sensitivity 1: Role of Other Shocks 250 U.S. Pollution: 1990= U.S. Regulation: Dirty Industries U.S. Regulation: Clean Industries Proportional Change in U.S. Environmental Regulation

58 Sensitivity 2: Pollution Taxes and NOx Budget Trading Program ln(ˆt jst ) = β 1 (1[NBP s ] 1[NBPIndustry j ] 1[Year > 2002]) + η st + γ jt + ψ js + ɛ jst Predicted Marginal Effect of NBP on Pollution Tax Year

59 Sensitivity 3: Implicit Pollution Taxes for Air Pollution and for CO2 1990= CO NO x PM 10 PM 2.5 SO 2 VOCs CO Year

60 Sensitivity 3: Historic Pollution Decomposition for CO = Actual Data (All Shocks) 40 Foreign Competitiveness Shocks Only US Competitiveness Shocks Only 20 US Regulation Shocks Only US Preference Shocks Only Deficit Shocks Only Year

61 Sensitivity 4: Share of Heat Input from Each Fuel Share of Total BTU Year Coal Distillate Oil LPG Natural Gas Residual Oil Other

62 Sensitivity 4: Potential SO2 Emissions per BTU, by Fuel Total SOx per BTU Year Coal Distillate Oil Natural Gas Residual Oil

63 Sensitivity 4: Potential SO2 Emissions per BTU, by Fuel 1990= Year NO x PM 10 SO 2 VOCs CO

64 Sensitivity: Other Considerations Other considerations: Detail of industry categories Constant v. increasing returns to scale in pollution abatement Induced innovation, improvements in abatement technology

65 Overview Statistical Decomposition Trade-Environment Model Data Estimation and Results: Parameters and Shocks Counterfactuals Conclusions

66 Conclusions Why are pollution emissions from manufacturing declining? Open and important question. Methods from trade, application to environmental economics Findings: Most of the decline is within narrowly-defined industries Pollution tax which rationalizes observed firm behavior has more than doubled since 1990 Environmental regulation explains 75 percent of more of observed reductions in pollution emissions Trade costs, productivity, preferences play little role Topics for future work: apply these methods to energy efficiency; market power.

67 Trade-Environment Model: Price Index in Changes Price index in changes: ˆP d = s (ˆP d,s ) β d,s ˆP d,s = [( ˆβ d,s ) θs (1 αs )(σs 1) 1 o λ od,s ˆM e o,s ( ŵo ˆb o,s ) θs (ˆτ od,s) θs 1 αs (ˆf od,s ) 1 θs (ˆt ) (1 αs )(σs 1) αs θs o 1 αs ] 1 αs θs Back to Slides