The Effect of Environmental Enforcement on Product Choice and Competition: Theory and evidence from India∗ Molly Lipscomb† University of Colorado, Boulder Department of Economics December 11, 2008 Abstract Sustainable development requires attaining a balance between maintaining environmental quality and controlling the impact of environmental regulations on productivity and competitiveness. This paper examines a margin of adjustment not commonly addressed in evaluating environmental regulation; the response by multi-product firms at the intensive (quantity per variety) and extensive (number of varieties) margins of their product portfolios. It parses the effect of changes in enforcement across the distribution of low to high productivity firms. I find (a) that firms react to changes in enforcement by increasing the share of their product portfolios allocated to clean products, (b) that consistent with theory the magnitude of this effect varies across the productivity distribution, and (c) that this in turn affects competition and quantities supplied in these markets. I develop a product selection model for heterogeneous multi-product firms and apply it to a setting in which large firms face changes in the enforcement of environmental regulations. The model shows that firms respond to changes in the fixed cost of producing the pollution intensive product by either shifting their portfolios toward the clean product or focusing their operations on a high margin pollution intensive product. The overall response depends on the ratio of profit margin to fixed cost of the pollution intensive product relative to the clean product and the intrinsic productivity of the firm. Portfolio reallocations by competitors allow highly productive firms specializing in the pollution intensive industry to gain market share. I test the predictions of the model using a firm and product-level data set for over 2300 large firms across all manufacturing industries in India for the period 1997-2005. This data is overlaid with state-level environmental enforcement data. High productivity firms in the pollution intensive industry invest in new, generally cleaner capital and gain in both sales and profitability following an increase in enforcement. Low productivity firms in both industries have reduced sales. The market consolidation toward cleaner high productivity firms in response to environmental regulation shows that the costs of regulation are not evenly distributed. Portfolio reallocations in response to changes in environmental regulations cause compositional changes in local manufacturing which increase the impact of the change in environmental enforcement beyond the benefits associated with the required abatement technologies. ∗

The author gratefully thanks Mushfiq Mobarak, Wolfgang Keller, Beata Javorcik, Aaditya Mattoo, Keith Maskus, Stephen Redding, Erin Mansur, Sheila Olmstead, Stephen Yeaple, Hale Utar, Maggie Chen, Randall Walsh, A.V. Chari, Leonardo Iacovone, Samuel Raisanen, Stephen Nicar, Scott Holladay, and conference participants at the Vail Environmental Economics Conference, the Yale Environmental Economics Seminar, and NEUDC for helpful comments. Any errors are my own. † Preliminary draft–comments appreciated, email: [email protected]. For the most recently updated version of this paper, please see http://molly.lipscomb.googlepages.com/mollylipscombresearch.

1

1

Introduction

The impetus to improve environmental protection in developing countries is often stymied by political pressure to avoid policies which risk impeding growth. As a result, environmental quality in developing countries has suffered. As of 2004, India had four major cities with particulate matter pollution levels over 100 µg/m3 ; levels in Delhi were measured at 150 µg/m3 (WDI (2004)).1 The OECD estimates that in India 13 percent of air pollution is emitted by industries and 50 percent of water pollution is industrial waste (OECD (2006)). Improving environmental regulation so that growth in developing economies is sustainable is an essential policy goal and requires an understanding of the effects of changes in regulations on local companies. There is, however, very little evidence in the literature as to how environmental regulation affects firms in developing countries. This paper analyzes the effect of environmental enforcement on companies from all manufacturing sectors across the states of India. It adds to the literature by analyzing an additional margin of adjustment by companies to changes in environmental regulations which has not previously been considered in the literature: shifts in the firms’ product portfolios. There are several reasons that systematic product portfolio changes in response to environmental policies are particularly important. Many large firms in developing countries produce a diverse basket of products. As firms respond to enforcement by increasing their production of the clean product and decreasing production of the pollution-intensive (“dirty”) product, the local product mix becomes cleaner. Environmental quality from increased enforcement of environmental regulations may therefore improve beyond the direct impact of firms acquiring the required abatement technologies. As these multi-product firms shift production away from the pollution-intensive market and toward cleaner products, the competitive structure of both markets is altered. Systematic changes in firms’ product mix may also impact measured productivity effects of a change in government policy. Multi-product firms produce a spectrum of products, each with an implicit level of productivity. Because a firm’s measured productivity is the weighted average of the productivity from each of the products in the firm’s portfolio, measured productivity effects from environmental enforcement are impacted by product portfolio changes. Bernard, Redding, and Schott (2005) show that productivity estimates in a trade liberalization environment can be 1

The World Health Organization guidelines warn against levels over 20 µg/m3 as particulate matter can cause significant health damage(WHO (2005))

2

biased as firms adapt to liberalization by changing their product mix; a similar effect applies to estimates of productivity effects of environmental regulations. The effect of environmental regulations on manufacturing competitiveness is a contested issue both in public policy circles and in the environmental economics literature. The early literature treating the effects of environmental regulations on productivity generally finds a negative effect (see Christainsen and Haveman (1981) for a survey of this literature), although there have also been suggestions that innovation resulting from increased regulation could increase productivity, see Maleug (1989) and Porter and der Linde (1995). Evidence from Smith and Walsh (2000) suggests that estimates which find support for the Porter hypothesis of increased productivity as a result of increased regulations suffer from measurement error as a result of the difficulty of disentangling the productivity effect of abatement technologies from the effects of changing factor prices. Gray (1987) finds that the average OSHA inspection rate decreases total factor productivity in an industry by 0.27 percentage points between the periods 1959-1969 and 1973-1978. Environmental regulation can also have large repercussions in terms of the overall scale and composition of local industrial activity. Recent literature on the effects of environmental regulation has focused on plant births, deaths and location. Becker and Henderson (2000) and List, Millimett, Fredriksson, and McHone (2003) find strong effects of local environmental regulations on plant location. Becker and Henderson (2000) show that differences in environmental enforcement between attainment and non-attainment areas can have important consequences on the structure of local industries as tougher enforcement leads to shifts in local industry toward less regulated small scale industry. Greenstone (2002) finds significant negative employment and capital effects for nonattainment relative to attainment counties. There is less available empirical evidence focusing on the effect of environmental regulations in developing countries where the industrial environment is significantly different from the developed country context. Murty and S.Kumar (2003) show that there may have been some productivity benefits from regulation to firms in the sugar industry using a distance function approach to estimate the production function for ninety-two sugar firms over three years (although this approach builds decreases in environmental damage directly into firm output). Murty and S.Kumar (2002) find increasing marginal costs for reducing pollution concentrations and decreasing marginal costs for decreasing pollution loads in India.

3

Empirical evidence suggests that there is variation in the cost of compliance across the productivity distribution of firms in pollution-intensive industries. Berman and Bui (2001) find that the productivity cost of environmental regulation for the petroleum refining industry is less than estimated in previous studies as abatement technologies may be more productive than older technologies. Gray and Shadbegian (2002) use data from the iron and steel and pulp and paper industry and Gray and Shadbegian (2003) use data from the pulp and paper industry to show that the overall costs of compliance are related to plant level productivity. Heterogeneity in productivity appears to be an important characteristic in guiding which plants adopt new abatement technologies. For this reason, in modeling the effects on product choice I explicitly maintain the assumption of heterogeneity in intrinsic productivity across firms. The trade literature has a key insight which informs my work in that it models variation in the effect of government policies across the productivity distribution of firms. Melitz (2003) extends the Hopenhayn (1992) model of firm entry and exit in a heterogeneous productivity context to a monopolistic competition framework to show that the effects of exposure to trade differ across the productivity distribution–the most productive firms export while the least productive firms are forced out of the market. I apply the concepts of the Melitz (2003) model to the question of how firms react to changes in environmental regulation. In addition to estimating the overall effect of changes in environmental regulations in a developing country, this paper adds to the literature by proposing a key margin of adjustment which should be analyzed when considering the productivity effect of environmental regulations: the product portfolio. To my knowledge, no current work in environmental economics examines firm adaptation to environmental regulations at the product level. Product level studies are rare as the data requirements for such projects are large and most of the literature on product level effects is new. The bulk of the existing products literature focuses on the effects of trade liberalization. My approach to analyzing product level changes by firms in response to changes in government environmental policies is therefore informed primarily by the trade literature. Bernard, Redding, and Schott (2007) and Bernard, Redding, and Schott (2006b) show that differences in product types coupled with heterogeneity among firms can lead to product switching in response to changes in environmental enforcement. Bernard, Redding, and Schott (2006a) and

4

Bernard, Redding, and Schott (2003) provide econometric results showing that the propensity to add and drop products varies across the productivity distribution. Eckel and Neary (2006) show that firms respond to trade liberalization by returning their product mix toward core competencies. Nocke and Yeaple (2006) develop an endogenous theory of firm scope which explains changes in firm size dispersion in response to changes in trade liberalization, and find that the firm size distribution flattens with multilateral trade liberalization. My results appear consistent with the main findings of the trade literature. Iacovone and Javorcik (2008) find that as firms choose products for export, they focus on core products. Similarly, in reaction to increased enforcement in a market, firms whose products in that market are not core products shift their product portfolio toward clean products. In structuring an approach to analyzing product level environmental regulation changes, I adapt a Cournot oligopoly model of multi-product firms which allocate their product portfolios between the production of varieties of a clean and a dirty product. I recognize the importance of firm level heterogeneity in intrinsic productivity using an approach related to Melitz (2003). My model is different from existing models in that it considers the firm’s choice in product portfolio selected from unlimited varieties of each of two products. This allows me to examine the product selection changes which occur in response to changes in environmental policies. India offers a particularly interesting natural experiment in which to test the effects of changing environmental regulations. Environmental enforcement underwent extensive changes in 1997; prior to 1997 there was little enforcement of regulations and prosecution could be carried out only through the judicial system which was extremely slow moving. In 1997, in response to pressure by the World Bank and other multinational organizations, major changes were made in the enforcement of environmental law, surprising local firms. Enforcement is now carried out by the states, and state environmental ministries have the authority to close offending plants. Local control over environmental enforcement affords nice variation across the states of India which enables improved identification of the effects of environmental regulations. I test the effects of changes in local environmental enforcement over time and across states on a panel of 2300 firms with annual financial and product level data over a period of 8 years, with a focus on firm response to changes in enforcement at the level of choice of product mix. This paper’s empirical contribution to the literature is in its analysis of how large multi-product

5

firms adapt to changes in environmental enforcement and how this adjustment varies across the productivity distribution. By analyzing firm response to changes in enforcement at the product level, I am able to distinguish the portfolio reallocation effect among producers of both dirty and clean products from the market consolidation effect in the pollution-intensive market which follows from reduced competition. I trace the response among firms to changes in enforcement by analyzing changes at the extensive and intensive product margins and analyze the difference in effects across the productivity distribution. I then examine the impact that these changes had on overall firm revenue and profit. On the extensive margin, firms drop pollution-intensive products more often when environmental enforcement increases. This effect varies across the productivity distribution in that high productivity firms are more likely to drop dirty products in response to changes in enforcement and less likely to drop clean products. The converse holds for low productivity firms. On the intensive margin, as predicted by the model, there is no overall effect of changes in enforcement, but firms adapt quantities produced to changes in competition as other firms shift their portfolios out of the dirty market and into the clean markets. This is as predicted by the theoretical model. The portfolio shifts resulting from the extensive margin and intensive margin response to changes in environmental enforcement lead to important changes in the structure of the industry. There is an overall reduction in the sales of dirty products across all firms. However, there is also an increase in the sales of dirty products at high productivity firms producing only pollution-intensive products. These firms respond to the decrease in competition in the pollution-intensive market. Overall, the local industry consolidates toward high productivity firms following an increase in environmental enforcement. High productivity firms producing the dirty product actually increase in profitability following an increase in enforcement. Analysis of investment among producers of only the dirty product suggests that following environmental enforcement firms remaining in the dirty industry invest in capital stock. New capital stock is less likely to be pollution-intensive than the older capital stock being retired, therefore as sales consolidate toward the high productivity firms, these firms are also using newer and cleaner technologies. This paper provides a novel approach to analyzing the impact of changes in environmental regulations in that it analyzes firm response at the product level. Product level data makes it possible

6

to separate the effect of portfolio shifts in product mix in response to changes in environmental enforcement from composition effects as the competition decreases in the pollution-intensive industry. Changes in local environmental enforcement have a strong impact on the local economy as firms shift out of producing the dirty product and into producing the clean product. Sales consolidate toward high productivity firms, and profitability increases at high productivity firms even in the dirty industry. In section 2, I suggest a modeling framework for a firm’s decision in determining the quantity and varieties of the clean and dirty product which they will produce. In section 3, I explain the policy context in India and summarize the characteristics of the data. In section 4, I test the predictions of the model and trace the response of firms to changes in environmental enforcement at the product level. In the conclusion of this paper, I suggest several avenues for future research which would improve the estimates and allow for increased confidence in the measurement of the local industry impact of changes in environmental enforcement.

2

Modeling Product Responses to Changes in Environmental Enforcement

Changes in environmental enforcement change a firm’s cost of non-compliance with environmental regulations. Prior to an increase in enforcement, a firm producing pollution-intensive products may fail to invest in the required abatement technology believing that the probability of having to pay a penalty is low. Following the increase in enforcement intensity, the manager’s expectation that the firm will face stiff penalties for failure to comply may increase. As a result, the manager will choose to either invest in the required technology or stop producing the pollution-intensive product. I show how changes in environmental enforcement can affect a firm’s selection of its optimal product portfolio. I present a multi-product Cournot competition model in which firms interact in a three stage game. This model allows us to investigate some of the industry composition effects as firms react to the portfolio reallocation of other firms. The interaction effects substantially complicate the model, so while I am not able to derive closed form solutions, I use the implicit function theorem to derive signs of the derivatives in which we are interested. The model shows that portfolio reallocation across the productivity distribution depends on the ratio of fixed cost to profit margin in the dirty product market relative to the clean product market

7

as well as the productivity distribution across firms. High productivity pollution-intensive firms can be beneficiaries of an increase in environmental enforcement under most parameter values, while clean product producers lose market share as a result of the increase in environmental enforcement.

2.1

A Cournot Model of Interaction among Heterogeneous Multi-product Firms Responding to Enforcement Changes

I present a model in which heterogeneous firms compete in a three stage Cournot game in which consumers have a love of variety across each of two products, a clean product X, and a dirty product Q. X and Q are composites of the production of each variety of the clean and pollution-intensive R NR δ R NR γ products. That is, they can also be written: X = 0 0 xdidj and Q = 0 0 qdidj. In the first stage of the game, firms decide whether or not to enter the market. Firms pay Fe to enter and receive a productivity draw ϕ on entering. Higher realizations of ϕ will mean that a firm faces lower marginal costs of producing either product as compared to firms which receive lower realizations of ϕ. In the second stage of the model, firms decide how many varieties (δ and γ) of each product to produce. In the final stage of the model, firms choose the quantity of each variety which they have chosen to produce–x and q. Demand for each product is: p = a − b(1 − e)x − beX for the clean product and ρ = a − b(1 − e)q − beQ for the dirty product. Consumer preference for variety is signified by e. Because (1 − e) is the preference for variety among consumers, e must be between 0 and 1. As e approaches 1, consumers see new varieties of the clean or dirty product as equivalent to all other varieties of that product. That is, there are singular products, X and Q, and the problem becomes a Cournot oligopoly problem in which firms compete over quantities of two products. As e approaches 0, new varieties of the product are each their own product in which the firm is a monopoly producer. In the monopoly case, the firm chooses to produce either varieties of X or Q depending on the ratio of profit margin to fixed cost, and produces only the product with the highest profit ratio. There is a fixed cost for each variety which the firm decides to produce, and this fixed cost increases in the total number of varieties which the firm produces. This assumption reflects the fact that as a firm invests in more products it will be more highly leveraged in order to cover its large capital investments, and as a result its cost of capital will be higher. In addition, managerial competence will be spread across more products as the firm diversifies product lines. Schoar (2002)

8

finds that conglomerates on average begin as higher productivity than stand-alone plants, but as they diversify, their productivity goes down. Similarly, we can expect that the fixed costs of entering new product lines will increase as the firm diversifies farther from its core products. Profit for a firm producing both the clean and the dirty product is therefore defined as:

πj = δj [pj −

d c ]xj − δj F (γj + δj ) + γj [ρj − ]qj − γj G(γj + δj ) − Fe ϕj ϕj

We assume that firms do not respond to decisions made by the other firms in terms of their product choice and quantities. We find Nash Equilibrium by using backward induction, so we begin by choosing quantities q and x to maximize profits. Taking the first order conditions and applying the Leibniz rule we find: dπi c = δj [a − b[(1 − e)xj + eX] − ] − b(1 − e)xj δj − beδj2 xj = 0 dxj ϕ

(1)

d dπi = γj [a − b[(1 − e)qj + eQ] − ] − b(1 − e)qj γj − beγj2 qj = 0 dqj ϕ

(2)

Solving, we find the reaction functions q =

d a−beQ−j − ϕ e 2b(1+ 2 (γ−1))

and x =

c a−beX−j − ϕ e 2b(1+ 2 (δ−1)) .

Because the reaction functions for x and q have γ and δ in the denominator, the problem becomes unwieldy, so I solve for the asymptotic cases of e = 0 and e = 1, and use the implicit function theorem to derive the expected signs of the comparative statics when 0 < e < 1.

2.2

Absolute Love of Variety: e = 0

Because there is absolute love of variety, producing more of any variety has no effect on the price of other varieties. This means that the firm simply produces the product which has the highest average profit until the marginal profit of producing another variety equals the marginal cost of increasing fixed costs for varieties. Firms produce only the product with the highest ratio of marginal profit to fixed cost. We solve the final stage of the game for producers of either product to find the reaction functions: x =

c a− ϕ 2b

, or q =

d a− ϕ 2b .

We substitute the reaction functions for x and q into the profit

function and take the first order conditions with respect to δ and γ in order to derive the solutions for the system. Assuming that the firm produces the pollution-intensive product, the solution is as

9

follows: π=

a − ϕd
d 
 γ a−b − − Gγ 2 2b 2b ϕ

a−

d ϕ

a − ϕd
d
a − ϕd
dπ a−b − − 2Gγ = 0 = dγ 2b 2b ϕ γ=

(a − ϕd )2 8bG

(3)

(4)

(5)

(a − ϕd )2 dγ <0 =− dG 8bG2

(6)

dq =0 dG

(7)

(a − ϕd )2 d2 γ <0 = −2 dGdϕ 8bG2 ϕ

(8)

Here the only cases in which changes in the fixed cost of a variety can have an effect on the firm’s choice of the clean product are cases in which increases in G cause the profit of producing a variety of the dirty product to be less than profit from producing a variety of the clean product, so the firm switches its entire line of production to the production of clean products.

2.3

Indifference to Variety: e = 1

When e = 1 the problem becomes similar to a Cournot oligopoly problem with asymmetric costs, with the caveat that some firms now produce both products. There are no varieties of either good since the consumer ascribes zero value to increased variety. The firm may produce one product or both products depending on the marginal profit it derives from production of the second product: the marginal profit that it derives from the addition of the second product line would have to be at least as large as the increase in total fixed costs, or F + 2G in order to add the clean product, G + 2F in order to add the dirty product. Clearly only the most productive firms will produce both products, as entering the second market raises the fixed cost of producing both products, and therefore would only be profitable for the highest productivity firms. We then solve the Cournot Oligopoly problem for segmented markets. Because e = 1 the firm’s problem simplifies to one of the following three cases:

π = (a − bX)x − F

10

(9)

π = (a − bQ)q − G

(10)

π = (a − bX)x + (a − bQ)q − 2(G + F )

(11)

We use backward induction to solve, with the problem now a two stage game since the firm produces at most one variety of each product. Firms that produce in both markets act in each market as a Cournot Oligopolist if they enter. Taking the first order conditions of equations 9 and 10 with respect to quantity in order to derive x and q, we find that for firms producing x:

a − bX−j − 2bx −

c =0 ϕ

(12)

a − bQ−j − 2bq −

d =0 ϕ

(13)

We can also write this as: a − bX −

c c = p − = bx ϕ ϕ

(14)

d = bq ϕ

(15)

and a − bQ − Summing across firms, we find: X=

Q=

Na −

RN 0

1 + bN RM Ma − 0

c ϕ

(16)

d ϕ

(17)

1 + bM

where N, M are the measures of firms choosing to produce in each market. We can substitute these into the price equations, and use price in the reaction functions for x and q found above, in order to find that the quantities produced by each firm are:

x=a−

q =a−

RN

Na −

0

c ϕ dj

1 + bN Ma −

RM 0

1 + bM

11

d ϕ dj

−

c ϕ

(18)

−

d ϕ

(19)

2.3.1

Entry and Exit

The market will be segmented as some firms choose to produce only the dirty good, some firms choose to produce only the clean good, and some firms choose to produce both products. We assume here that the fixed cost of producing the dirty good is higher than the fixed cost of producing the clean good. This assumption can be justified in that many dirty industries are heavy industries (iron and steel, refineries, sugar) which require large capital investments in order to enter. It can be shown that the segmentation of the market will occur by productivity, with the lowest productivity firms which remain in the market entering the clean (low fixed cost) product market until some cutoff productivity ϕ ∗ ∗. At ϕ ∗ ∗ firms will switch into producing only the dirty (high fixed cost) product. Finally, at some productivity level ϕ ∗ ∗∗, firms will produce both products. We solve for the equilibrium number of firms by solving the first stage of the model using the zero profit condition. We know that the lowest productivity firm in the clean market earns zero expected profits. Substituting in the solution found above for x, we have: 

a−

Na −

RN 0

1 + bN

c ϕ dj

−

c 2 (1 − ϕ∗) = F + Fe ϕ∗

(20)

This means that the following equality must hold: Z b 0

N

p p c c c dj + bN [ b(F + Fe ) − ] = a − b(F + Fe ) − ϕ ϕ∗ ϕ∗

(21)

Producers in the dirty product market enter the dirty market rather than the clean market if they believe that their profits will be at least as high as they would be in the clean product market. That is; RN RM N a − 0 ϕc dj
c
2 M a − 0 ϕd dj
d
2 1 1 a−b − − F − Fe ≤ a − b − − G − Fe b 1 + bN ϕ b 1 + bM ϕ

(22)

For the marginal firm producing in the clean product market (the firm with productivity level ϕ ∗ ∗, equality will hold. Finally, firms decide to produce in both markets if: RN RM N a − 0 ϕc dj
c
2 1 M a − 0 ϕd dj
d
2 1 a−b − + a−b − − 2(G + F ) − Fe ≥ b 1 + bN ϕ b 1 + bM ϕ

12

RM M a − 0 ϕd dj
d
2 1 a−b − − G − Fe b 1 + bM ϕ Equality must again hold at ϕ ∗ ∗∗. Setting C =

RN 0

c ϕ dj

and D =

(23) RM 0

d ϕ dj,

this gives us the

following equations (after simplifying): 

a−b

Na − C c
2 (1 − ϕ∗) = F + Fe − 1 + bN N

(24)

1 Na − C
c
2 Ma − D
d
2 1 a−b − − −F = a−b −G b 1 + bN ϕ∗∗ b 1 + bM ϕ∗∗  c 2 Na − C
− a−b = 2G + 2F 1 + bN ϕ ∗ ∗∗

(25) (26)

In addition, we know that M is the measure of firms producing the dirty product, so it must be the interval M = (ϕ ∗ ∗, 1). N is the measure of firms producing the clean product, so it is the union S of two intervals: N = (ϕ∗, ϕ ∗ ∗) (ϕ ∗ ∗∗, 1). The effect of a change in the fixed cost of the dirty product will depend on the distribution of productivity across the firms. In order to solve, I assume a uniform distribution. This means R ϕ∗∗ R1 R1 that M = 1 − ϕ ∗ ∗, N = 1 − ϕ ∗ ∗ ∗ +ϕ ∗ ∗ − ϕ∗, C = ϕ∗∗∗ ϕc dj + ϕ∗ ϕc dj, and D = ϕ∗∗ ϕd dj. Using the implicit function theorem to find the comparative statics of the system for a change in G. I find that the following results hold for entry and exit under most parameter values with uniformly distributed productivity:

dϕ∗ dG

< 0,

dϕ∗∗ dG

> 0 and

dϕ∗∗∗ dG

> 0. Please see the appendix for

the derivation of the results.

2.4

Love of Variety: 0 < e < 1

We find Nash Equilibrium by using backward induction, so we begin by choosing quantities q and x to maximize profits. As shown above, q =

d a−beQ−j − ϕ e 2b(1+ 2 (γ−1))

and x =

c a−beX−j − ϕ e 2b(1+ 2 (δ−1)) .

Solving for δ and γ occurs through substitution of the reaction functions for x and q into the model and deriving the first order conditions for δ and γ. Because the reaction functions for x and q have γ and δ in the denominator, the problem quickly becomes complicated. It is not possible to derive closed form solutions here; instead I use the implicit function theorem to derive expected signs for the relevant comparative statics. Please see the appendix for the derivation. The comparative statics results for most parameter values are as follows:

13

dX dG

> 0,

dQ dG

< 0,

dδ dG

3

> 0,

dγ dG

< 0,

dx dG

< 0 and

dq dG

> 0.

Data

The firm and product level data used in this paper was compiled by Capitaline for the period 1990-2005. The product level data is from the CMIE Census of Manufacturers, as reported by Capitaline. Products are each identified by name as well as by eight digit product code. These eight digit product codes approximate the level of aggregation of six digit ISIC codes. There are 2,300 firms over the period 1997-2005 in the data, and 3,500 product codes. The sample is a highly unbalanced panel of annual financial and product level data. Firms which report product level data report data for their products to a rounding error of the total sales figures listed in the financial reports. Highly monitored firms have been identified by matching the sample to the list of monitored firms from the Ministry of Forestry and the Environment. Firms which have had a plant closed because of environmental problems have been omitted from the sample. Table 1: Distribution of Firms in the Sample as of 2004 State Andhra Pradesh Assam Bihar Chattisgarh Gujarat Haryana Jharkhand Karnataka Kerala Madhya Pradesh Maharashtra New Delhi Orissa Pondicherry Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal All India

Produce Both 48 3 3 1 41 11 1 22 6 11 129 21 4 3 13 6 44 22 44 433

Dirty Only 24 0 1 1 23 4 1 8 2 4 71 9 6 0 2 4 12 7 24 203

Clean Only 42 1 0 0 95 35 1 36 15 21 220 83 1 0 25 34 114 31 38 792

All Firms 114 4 1 5 159 50 3 66 23 38 423 116 12 3 40 47 172 61 107 1444

Productivity is calculated in two stages using both ordinary least squares and Levinsohn-Petrin estimation methods. Levinsohn and Petrin (2003) is designed to condition out simultaneity bias

14

from unobserved shocks in the production process which may be correlated with investment. Output is the dependent variable while labor, capital, and materials are the inputs. Nominal output is deflated by a set of wholesale price indices at the 2-digit level from the Reserve Bank of India, while capital inputs are calculated from detailed data on net values of land, buildings, machinery and computers, all deflated by the relevant sector deflators. Labor input is calculated by normalizing the wage bill of each firm by the average wage in a given 2-digit sector in a given year. Materials are deflated by input-output coefficient weighted sector deflators based on the wholesale price index. Estimated coefficients for both the Levisohn-Petrin productivity estimates and the OLS productivity estimates are shown in tables 15 and 16. Each firm is assigned an indicator for whether it belongs to the top decile of the productivity distribution of its industry as of the year it entered the sample, the lowest decile, or the remaining 80% of firms in its 2-digit industry. The productivity level in the first year the firm appears in the data set is then considered the firm’s productivity endowment. These productivity indicators are then interacted with the policy variables, so that we can observe the difference in the effect across the productivity distribution. The product adding variable is assigned as a 1 in the year in which a firm first produces a product, 0 otherwise. The year in which the firm enters the sample is omitted. The product dropping variable is assigned as a 1 if a firm drops a product in a given year and never produces it again, 0 otherwise. The year in which the firm leaves the sample is omitted. Summary statistics are provided in table 3.

3.1

Environmental Policy in India

Following the Bhopal disaster in 1984, there was some progress in terms of legislative reform in environmental policy in India–in 1986 the Environmental Protection Act was passed. However, little enforcement occurred, and the government’s emphasis was primarily on encouraging economic growth across the economy rather than prioritizing the safeguarding of the environment. Therefore, prior to 1997, little progress had been made in India in terms of environmental enforcement. India’s swift and varied rate of increase in enforcement of environmental regulations across its states makes it an ideal environment in which to analyze the effects of changes in environmental enforcement. Following pressure from the World Bank, India underwent intensive changes in the enforcement

15

of their environmental regulations in the late 1990s. Prior to 1997, most of the enforcement of environmental laws occurred through the court system. However, the court system in India is notoriously slow, which makes the threat of judicial action weak, and inspections were ineffective (Pargal, Mani, and Huq (1997)). In 1997 state pollution control boards were given greater regulatory authority and the central pollution control board was given the authority to directly sanction firms where the polluting behavior of the firm was egregious. In addition, new firms were required to be licensed by state pollution control boards if their production was in one of the 17 industries classified as highly polluting by the Ministry of the Environment, and therefore heavily regulated by the pollution control boards. Existing firms are required to apply for licenses in order to expand (The Hindu, 1999). These changes came as a surprise to local firms which were accustomed to low levels of environmental enforcement. Environmental enforcement is most strong against large firms operating in one of the 17 highly polluting industries as listed by the Ministry of the Environment. These industries include aluminum, caustic soda, cement, copper, distillery, dyes and dying industries, fertilizer, iron and steel, leather, pesticide, petrochemicals, pharmaceuticals, pulp and paper, refinery, sugar, triphenylphosphite (TPP), and zinc. Products are classified as either “clean” or “dirty” depending on if they fall into one of the 17 highly polluting industries. The enforcement measures are collected at the state level. They are the percent of heavily monitored firms which have been closed in a given state in a given year by the state pollution control board. They are reported in annual reports by the Indian Ministry of the Environment from 1997 through 2005. The state enforcement measure varies across states and across years. West Bengal has an average of 25% of the heavily monitored plants closed from 1997 to 2005 with a low of 18 percent closed in 2000 and a high of 29 percent closed in 2002, while Gujarat has an average of 3% closed with a low of 1.7 percent closed in 1997 and a high of 4.0 percent closed in 2001. States which increase environmental enforcement do see improvements in environmental quality. Increases in the intensity of local enforcement of environmental regulations are highly correlated with improvements in local water quality. The water quality data is an unbalanced sample of annual station monitoring data from approximately 200 monitoring stations in rivers across India from 1997-2003. Water quality is graded from 1-5 where 5 is excellent quality. Even accounting for

16

Figure 1: State Level Enforcement: 1997 and 2005 fixed effects and time changes across India, we see that the water quality increases more in states where environmental enforcement is higher. Table 2: Environmental Quality Correlation with Enforcement Lagged State Enforcement Station FE? Year dummies? r2 N

Deviation from expected quality -1.433*** -1.342** (0.44) (0.60) Y Y N Y 0.023 0.027 963.000 963.000

Water Quality Change 1.455*** 1.398** (0.50) (0.56) Y Y N Y 0.020 0.036 1210.000 1210.000

*** p<0.01, ** p<0.05, * p<0.1 Dependent variable is the deviation from the expected water quality level at a station. Positive values imply improvements in water quality. Deviation from water quality is expected value minus actual measured value. Fixed effects by station are included. Standard errors clustered by state-year.

4

Estimation Strategy and Results

Environmental enforcement is costly to firms as it changes the cost of production primarily through the fixed cost of investing in abatement technologies. As enforcement intensity increases, firm managers perceive a larger expected cost of non-compliance; the manager’s expectation of having to pay a penalty if he does not invest in the required equipment increases. Firms may have reduced efficiency under increased enforcement if they require increased labor or materials in order to

17

comply, face increased input costs, or change their production function or technologies in order to comply. Multi-product firms often diversify their product mix away from the pollution-intensive product in response to an increase in enforcement as a result. Increased enforcement from the 1997 policy change had a significant impact on the productivity of firms. In order to estimate the impact on firms across manufacturing industries, I interact a dummy for the years affected by the policy change with the input output coefficient summed across the 17 highly polluting industries. Pollution intensity varies from a low of 0.2 for the garment production industry to a high of 0.76 for the coke and petroleum industry. Both national and state level changes in enforcement have a strong negative impact on productivity. I use Olley and Pakes (1996) estimation methods to estimate productivity in order to correct for any selection bias or simultaneity problem in investment with unobserved shocks to the production process. Estimates are presented in table 4. At the mean, I find that a one standard deviation change in the pollution intensity corresponds to a change in productivity from the national policy of 5.7 percent. This corresponds to a decrease in productivity for the least pollution-intensive firms of 0.1 percent, and a decrease in productivity for the most pollution-intensive firms of 23.1 percent. Overall increases in the state enforcement of environmental regulations also has a negative, although somewhat smaller and not significant coefficient. We should expect that state level changes have more of a marginal impact than national changes, however, since firms can compensate for state level changes to some degree by purchasing inputs from other states. As noted in Bernard, Redding, and Schott (2006b), these productivity changes from changes in government policies may not be well estimated at the firm level if multi-product firms systematically respond to the policies by changing their product mix and intrinsic productivity of products is not constant across a firm. Changes in the product composition of overall output has a direct effect on overall firm productivity, since measured productivity is a weighted average over the implicit productivity associated with each product. As shown in section 2, firms respond to changes in environmental policies by optimally adjusting their product portfolio. In addition to affecting the local production of the pollution-intensive product, these portfolio changes in product mix can have an important impact on local industrial composition. In order to find the effect of environmental enforcement, I first estimate the impact of changes in environmental enforcement on the extensive (number of varieties) and intensive (quantity produced per variety)

18

margins of firms producing dirty and clean products. I then differentiate the effect across the productivity distribution of firms, as intrinsic productivity can have an important effect on firm response to changes in enforcement as seen in section 2. I trace the overall impact of these portfolio shifts on the firm by focusing on the effects of environmental enforcement changes on firm revenue and profits. Finally, I examine firm level investment for firms remaining in the dirty industry following an increase in enforcement. The empirical results indicate that firms react to changes in enforcement on the extensive margin (number of varieties) primarily by increasing the rate at which they drop dirty products. High productivity firms are less likely to drop a clean product and more likely to drop dirty products following increases in enforcement. The overall effect on the intensive margin (quantity per variety) at the product level is not statistically significant, however there are important effects for firms which are most impacted by changes industry structure; firms who produce only dirty products increase their production of each variety of pollution-intensive product, while firms that produce only clean goods decrease their production per variety. The percent of production dedicated to clean products at multi-product firms producing both clean and dirty products increases in response to enforcement increases. Increases in environmental enforcement lead to consolidation in the industry. Highly productive firms in the pollution-intensive industry benefit from an increase in environmental regulations. These firms gain market share and increase in profitability as other firms shift their production toward the clean industry. The least productive firms in the clean industry lose market share as the competitive environment in the clean product market becomes more active. There are several factors which may bias the estimates in this paper toward zero. The measured effects at the extensive margin in India may be lower than in other countries, as found by Goldberg, Khandelwal, Pavcnik, and Topalova (2008) in estimating the response at the extensive margin in India for trade liberalization. In addition, firms may expect the changes in regulations and react prior to the actual changes in enforcement or choose a clean product to adopt in a year prior to the change in enforcement. Regulatory authorities may also reduce environmental enforcement when they expect enforcement to have a strong impact on the local economy. Therefore, this paper provides lower bound estimates for the product switching effects. In addition, the measure of environmental enforcement is a relatively blunt proxy for envi-

19

ronmental enforcement. It would be ideal to use a firm manager’s expectation of the costs of non-compliance. This enforcement variable is measured across each state across all industries. It does not capture variation in the costs of compliance across industries, nor does it capture variation in whether compliance requires a fixed cost investment in abatement technology or a marginal cost through additional safety precautions or waste disposal charges. While this paper lays the groundwork for understanding changes in local industrial environment, there is much room for future work to improve estimates through the collection of larger amounts of data on local regulations and differences in regulations across industries. In the conclusion I suggest several avenues for future research which may improve on the estimates of the importance of the product margin of adjustment in the calculation of productivity. In this section I generate empirical tests for the theoretical predictions and examine the impact of these changes in terms of industry composition.

4.1

Extensive Margin Effects

In section 2 we saw that we should expect a large proportion of firm adaptation to occur on the extensive margin. We expect to see firms shifting their portfolios away from the pollution-intensive product and toward the clean product. The empirical estimates support this prediction, as we see firms shift out of pollution-intensive products at higher rates following an increase in enforcement. The estimation model used to estimate the effect on product adding and dropping in both the clean and the dirty market is a linear probability model specification as follows:

ExtensiveM arginChangeif t = αf + β1 StateEnf orcements(t−1) + β2 Zf t + γt

Products are indexed by i, firms are indexed by f, states are indexed by s, and time is indexed by t. Only firms which produced dirty products in the prior year are included in the regression for dropped dirty products, only those which produced clean products are included in the regression for dropped clean products. Zf t contains controls for firm size and age. Firm and year, or firmproduct and year fixed effects are included depending on the specification, and standard errors are clustered by state, year. Results are presented at the firm level in table 5, and in table 6 at the product level. I find that firms drop the dirty product more often in response to a change in enforcement

20

and add the dirty product less often (although the adding effect is not significant at conventional levels). I find that firms are 21 percent more likely to drop a dirty product following a one percent increase in enforcement–this is an increase in probability of dropping a dirty product at the mean from 11 percent probability of a dirty product being dropped to a 32 percent probability of a dirty product being dropped for a ten percent increase in the level of enforcement. Firms are 6 percent more likely to drop a given dirty product line following a one percent increase in enforcement. At the mean of 3.5 percent probability of a given dirty product being dropped by a firm in a given year, this suggests that the probability of a dirty product being dropped increases to 9.5 percent for a 1 percent increase in enforcement. These effects are significant at the 1 percent and 5 percent levels respectively. In order to distinguish the competitive environment effects of the change in policy, I split the sample between firms producing both clean and dirty products and firms specializing in the pollution-intensive product market. Results are presented in table 7. I find that firms specializing in the pollution-intensive product add the dirty product 14 percent more frequently following a 1 percent increase in environmental enforcement; this effect is opposite in sign of the effect for firms competing in both markets. This suggests that firms in the pollution-intensive market are reacting on the extensive margin to changes in the competitive environment. Firms producing in both the pollution-intensive market and the clean market drop pollution-intensive products 14 percent more frequently in response to an increase in environmental enforcement. As pollution-intensive products become more profitable for firms continuing production in that market because of the reduction in competition, those firms increase production of the dirty product. I use the following estimating equation to find the effect across the tiers of the productivity distribution:

ExtM arginChangef t =αf + β1 Enfs(t−1) ∗ I(ϕf ≤ ϕ10pct ) + β2 Enfs(t−1) ∗ I(ϕ10pct < ϕf < ϕ90pct ) + β3 Enfs(t−1) ∗ I(ϕf ≥ ϕ90pct ) + γt

The variable I(ϕf ≤ ϕ10pct ) signifies an indicator for whether the firm belonged to the bottom 10 percent of the productivity distribution of its industry in the first year in which it entered the

21

sample. I(ϕf ≥ ϕ90pct ) signifies an indicator for whether the firm belonged to the top 10 percent of the productivity distribution of its industry in the year it joined the sample, and I(ϕ10pct < ϕf < ϕ90pct ) signifies an indicator for whether the firm was in the remaining portion of the distribution, the middle 80 percent. These indicators are interacted with the lagged state enforcement variable, which enables me to examine the effect of changes in environmental enforcement on the separate sets of firms in each productivity tier. Firm fixed effects are included in order to net out any unobserved firm level time invariant characteristics, and year dummies are included in order to control for any economy wide shocks. Standard errors are clustered by state and year. Results are presented in table 8. High productivity firms react more strongly to the change in enforcement intensity than the rest of the productivity distribution. High productivity firms are 53 percent more likely to drop a dirty product following a one percent increase in enforcement. The average rate of dirty product dropping among high productivity firms is 11.4 percent. This implies that for a 1 percent increase in the enforcement variable, at the mean, a high productivity firm will increase its rate of dropping dirty products to 69 percent. Low productivity firms drop dirty products 95 percent less frequently for a 1 percent increase in the enforcement variable on average. These effects are significant at the 1 percent level. In order to test the significance of the difference in coefficients across the productivity distribution, I generate an F-statistic for the test of the hypothesis that the high productivity and low productivity coefficients equal the mid productivity coefficients. I reject the null that the coefficients equal the mid productivity coefficients at the 1 percent level in both cases. The model predicts that the effect of increased enforcement in the clean market should be the opposite of the effect of enforcement in the dirty market. This is borne out in the data. The magnitude of these effects is smaller, but I find that following a 1 percent increase in enforcement, high productivity firms are 23 percent less likely to drop a clean product (although this effect is not statistically significant) and low productivity firms are 27 percent more likely to drop a clean product (significant at the 10 percent level).

4.2

Intensive Margin Effects

In order to measure the impact of the changes in enforcement on the firm’s product mix, I begin by considering the effect of changes in regulation on the quantity of each variety of the clean and

22

dirty product which are produced. The predictions of the model on quantities produced per variety depended on the competitive environment: in the monopolist model (absolute love of variety) changes in the fixed cost per variety had no effect on the quantity produced per variety. In the Cournot model, we saw that as the market interaction effect became larger, an increase in the fixed cost of the dirty good had a stronger positive effect on the quantity per variety of the dirty good and a negative effect per variety on the clean good. I estimate the effect of state enforcement on the quantity of each variety produced in the equation: logQuantityif t = αif + β1 StateEnf orcements(t−1) + β2 Zf t + γt Products are indexed by i, firms are indexed by f, states are indexed by s, and time is indexed by t. The estimates of the effect of changes in enforcement on the quantity of each product are presented in table 9. Quantity is measured at the level of product within a firm in a given year. Firm-product and year fixed effects are included in the regression as well as controls for firm size and age (Z). The results for the sample as a whole are not significant, as would be expected from the somewhat ambiguous predictions of the model. It could be argued that an estimate of changes in quantity in response to changes in local regulations may be biased as both quantity and local regulations may change through changes in local preferences. In controlling for prices in order to be sure that the measured results are not being unduly affected by demand changes rather than supply changes, I find that the results do not change. I also split the sample to investigate the impact on the firms most directly affected by a change in the competitive environment: the firms producing only clean products and those producing only dirty products. Here, I see that the effects are as predicted by the Cournot model: quantity per variety decreases by 0.9 percent among the clean product producers in response to the increase in competition from firms moving away from the dirty product, and increases by 1.8 percent among the products produced by producers of only pollution-intensive products in response to the decrease in competition in the pollution-intensive market. The model also suggests that the changes in the intensive margin should be different across the productivity distribution: low productivity firms should decrease their production of clean products by the largest margin in response to an increase in competition in the market. In addition, high productivity firms should actually increase their production of dirty products in response to the

23

decrease in competition arising from their competitors shifting toward production in the clean market. In order to test this prediction, I split the sample in the model between the bottom decile productivity firms, the mid 80 percentile of productivity, and the top decile productivity producers.

logQuantityf it =αf + β1 Enfs(t−1) ∗ I(ϕf ≤ ϕ10pct ) + β2 Enfs(t−1) ∗ I(ϕ10pct < ϕf < ϕ90pct ) + β3 Enfs(t−1) ∗ I(ϕf ≥ ϕ90pct ) + γt

The variables I(ϕf ≤ ϕ10pct ), I(ϕf ≥ ϕ90pct ), and I(ϕ10pct < ϕf < ϕ90pct ) are indicators for the bottom decile of the productivity distribution, the top decile of the productivity distribution, and the remaining 80 percent of the productivity distribution as in the estimation of the heterogeneous extensive margin effects. These indicators are interacted with the lagged enforcement variable in order to distinguish the difference in response to changes in enforcement across the productivity distribution. I find an increase in the quantity per variety produced at high productivity firms in the dirty market of 4.2 percent (significant at the 10 percent level) and a strong decrease in the quantity produced per variety in the clean product market of 5.6 percent (significant at the 1 percent level). This is as we expected from the predictions of the Cournot love of variety model. In inspecting the F-statistic for the test of the hypothesis that the high productivity and low productivity coefficients equal the mid productivity coefficients. I can reject the null that the coefficient for the low productivity firms equals the coefficient for the mid productivity firms at the 1 percent level for the firms specialized in production of the clean product.

4.3

Revenue Effects

We are also interested in the effects of enforcement on firm revenue in each product market, as this demonstrates whether a firm is expanding or contracting within a market. This gives us an indication of how the industry is responding as a whole to changes in environmental enforcement. I estimate the equation:

logRevenuef t = αf + β1 StateEnf orcements(t−1) + β2 Zf t + γt

24

Estimates are presented in table 11. Sales revenue in the pollution-intensive product market decreases by 1.1 percent in response to a one percent increase in enforcement. This is a large impact of increased environmental enforcement. As should be expected from the results in section 4.2, the effect of increased enforcement on overall clean sales revenue within a firm is not significant. The effect on revenue in the pollution-intensive market among specialized firms is not significant, as would be expected from some firms increasing their production in response to reduced competition while others drop products in response to increased fixed costs. Testing the effect across the productivity distribution, we find support for the model’s predictions.

logRevenuef t =αf + β1 Enfs(t−1) ∗ I(ϕf ≤ ϕ10pct ) + β2 Enfs(t−1) ∗ I(ϕ10pct < ϕf < ϕ90pct ) + β3 Enfs(t−1) ∗ I(ϕf ≥ ϕ90pct ) + γt

The sample is separated into firm revenue from dirty products and firm revenue from clean products among firms producing in both markets and those specializing in the dirty market. Estimates are presented in table 11. High productivity firms increase market share across the sample. The point estimate for the overall effect of enforcement on revenues in the pollution-intensive market for high productivity firms is 1.2 percent, however, this effect is not significant at conventional levels. Among firms producing only pollution-intensive products, the increase in revenue among high productivity producers is significant. A 1 percent increase in enforcement corresponds to a 0.87 percent increase in revenues for the high productivity firms in the pollution-intensive market. The effects of enforcement on revenues in the clean market across the distribution is even stronger. For a one percent increase in enforcement, we see a 2.8 percent decrease in sales for low productivity firms and a 3.8 percent increase in sales for high productivity firms. These effects are significant at the 5 percent level. In order to estimate the effect of enforcement on the overall product portfolio changes by a firm, I measure the effect of an increase in enforcement on the percent of sales by the firm which are in clean products. I limit the sample to firms which produce both clean and dirty products in the given year in order to examine the marginal impact on product choice, and measure the effects

25

across the productivity distribution.

P ctCleanf t = αj + β1 Enfs(t−1) ∗ I(ϕf ≤ ϕ10pct ) + β2 Enfs(t−1) ∗ I(ϕ10pct < ϕf < ϕ90pct ) + β3 Enfs(t−1) ∗ I(ϕf ≥ ϕ90pct ) + γt

Firm and year fixed effects are included, and errors are clustered by the state and year of observation. Results are presented in table 12. I find a strong positive effect in the high productivity firms. For a 1 percent increase in enforcement, I find a 0.23 percent increase in the percent of revenue coming from sales of clean products in the product portfolios of high productivity firms. The F-statistic shows that the effect among high productivity firms is significantly different from coefficient on mid-range productivity firms at the 1 percent level of significance. In order to test the robustness of the effect on the percent of the product portfolio at the highest productivity firms, I estimate the model with the enforcement variable moved one period forward. In order to find an effect here, the firms would have to be shifting their product portfolios in advance of the changes in enforcement. I find that there is no significant effect in area of the productivity distribution when the enforcement measure is moved forward.

4.4

Process Changes in the Pollution-Intensive Market

The growth in revenue among high productivity firms which remain producers of only pollutionintensive products when enforcement increases is intriguing. As producers of diversified bundles of goods re-weight their product portfolios toward the clean product, high productivity firms specializing in the pollution-intensive product market appear to gain market share. This suggests there is consolidation in the pollution-intensive market toward the high productivity firms. Investigating the investment response of pollution-intensive product producers is revealing. Estimates are presented in table 13. I find that for a 1 percent increase in the enforcement variable, there is a 1.1 percent increase in investment among producers who continue to produce the pollutionintensive product. Separating the effect across the productivity distribution reveals dispersion in the effect, however the coefficients are not statistically significantly different from one another. I do, however, find that high productivity firms have a larger point estimate for capital investment than low productivity firms. Controls for firm size, a foreign dummy, firm fixed effects, and year fixed effects were included in the regressions.

26

From a policy standpoint, it is particularly interesting to consider the effects of enforcement on firm operating profit. These estimates may be biased upwards, particularly for the low productivity portion of the sample, where survival probability is most likely to be at issue. Foster, Haltiwanger, and Syverson (2008) show that selection on demand factors can be a key source of error in productivity estimates. Profit estimates will suffer from similar issues. It is worth investigating the estimates presented in 14. I find that the high productivity producers in the dirty product market actually have increased operating profits following an increase in enforcement. There is an estimated 2.5 percent increase in profitability among high productivity firms for a one percent increase in enforcement. This effect is significant at the 1 percent level. This effect is significantly different from the effect for mid-range productivity firms at the 1 percent level. Although the effect is not significant for high productivity firms producing only dirty products, the point estimate is even larger. This suggests that high productivity firms in the pollution-intensive market may on net benefit from the increase in enforcement. This is due to the reduction in competition in the high pollution industries as firms switch toward cleaner industries.

5

Conclusion

This paper shows that product level adjustment is an important margin of change for firms in response to changes in environmental enforcement.

This has several key policy implications.

Environmental benefits from increases in environmental enforcement may not be limited to the impact of implementation of required abatement technologies. As firms face increased costs of non-compliance, they may shift production away from predominately pollution-intensive products toward cleaner products. As environmental policies generate shifts toward a cleaner local product mix, advances in environmental quality may extend beyond the firms which invest in abatement technologies. In addition, there are important industry structure effects which arise from firms shifting their product portfolios. Increases in environmental enforcement cause industry consolidation toward the most productive firms. High productivity firms in the pollution-intensive market increase market share as other firms substitute out of the industry. Abstracting from firm exit, low productivity clean producers face increased competition from the firms switching into their product market and as a result lose market share.

27

Productivity, measured as the weighted average of revenue minus costs of all products in the firm’s portfolio, may be affected by adjustments made by the firm at the product level. In order to gain a comprehensive understanding of the overall welfare effects arising from changes in environmental regulations, it is important to consider the effects of product portfolio adjustments made by multi-product firms. Firms adapt to changes in regulations by changing their production function, their intensive margin and their extensive margin. The existing environmental economics literature focuses on changes on the intensive margin and changes in the production function in response to changes in environmental regulations and enforcement. This paper adds to the literature by analyzing the product portfolio composition component of firm adaptation to policy changes. I generate a model of product choice by heterogeneous multi-product firms, and show that firms react to regulatory shocks through changes in both their intensive and extensive margin. In addition, the response to changes in regulatory enforcement varies across the productivity distribution. I test the model using firm and product level data from India. India’s rapid increase in environmental enforcement through its state level pollution control boards from 1997-2005 was a strong divergence from the prior policy emphasizing growth across the sectors of the economy over the environment. Lackluster enforcement with authority primarily at the judicial level characterized the environmental policy prior to 1997. However, in 1997 state pollution control boards were given wider authority and the emphasis on pollution reduction was increased. Firms react to changes in environmental enforcement by adapting their product mix. High productivity firms decrease the frequency with which they add dirty products and decrease the frequency with which they drop clean products to a larger degree than other firms in the productivity distribution in response to changes in enforcement. In addition, high productivity firms increase the percentage of their portfolio allocated to clean products, suggesting that they are also reacting on the intensive margin. While I find support for the theoretical model in the data, many of the results are not strongly significant. One key reason for the large standard errors is the level at which the enforcement variable is being measured, and the large grouping of products into pollution-intensive and clean categorizations. The best available measure of enforcement at this time is a state level measure which does not vary by industry. The percentage of companies closed in a state gives no indication of whether the firms perceive regulations as increasing their marginal or fixed costs. Regulations

28

may affect firms differently across different industries; paper industries may perceive increases in enforcement as increased marginal costs as they are forced to improve treatment of the water which they use in production, while steel mills may be forced to make large fixed cost investments. Improved data on regulations and enforcement is necessary in order to obtain good measurements of the effect of enforcement on firm product choice. Cost of non-compliance may also vary across a state, with enforcement being particularly intense in highly populated regions. Measuring the effects at the district, rather than the state level may improve the independent variable in that it would be a better measure of a firm’s expectation of the likelihood of being punished in the event that they are not in compliance. One concern in testing the effects of changes in enforcement is that estimates may be biased if enforcement is weaker in areas where companies are particularly likely to respond to changes in regulation by reducing their product portfolios. This problem would be even larger if the enforcement effects were measured at the district, rather than the state level. Future work on this project will include developing an instrument for local environmental enforcement and testing the robustness of estimates to possible endogeneity arising from political bias away from enforcing regulations against sick plants and industries. Local district water quality is one potential instrument for local environmental enforcement. District water quality is unlikely to affect product choice for firms, but local managers may be more interested in enforcing regulations where public pressure has been instigated by particularly poor water quality in the rivers.

References (2004): World Development Indicators. World Bank. Becker, R., and V. Henderson (2000): “Effects of Air Quality Regulations on Polluting Industries,” Journal of Political Economy. Berman, E., and L. Bui (2001): “Environmental Regulation and Productivity: Evidence from Oil Refineries,” Review of Economics and Statistics, 83. Bernard, A., S. Redding, and P. Schott (2003): “Product Choice and Product Switching,” NBER Working Paper. (2005): “Products and Productivity,” CEPR Discussion Paper.

29

(2006a): “Multi-Product Firms and Product Switching,” NBER Working Paper, (12293). (2006b): “Multi-Product Firms and Trade Liberalization,” CEP Discussion Paper, 769. (2007): “Comparative Advantage and Heterogeneous Firms,” Review of Economics and Studies, 74, 31–66. Christainsen, G., and R. Haveman (1981): “The Contribution of environmental regulations to the slowdown in productivity growth,” Journal of Environmental Economics and Management, 8. Eckel, C., and P. Neary (2006): “Multi-Product Firms and Flexible Manufacturing in the Global Economy,” CEPR Discussion Paper, 5941. Foster, L., J. Haltiwanger, and C. Syverson (2008): “Reallocation, Firm Turnover, and Efficiency: Selection on Productivity or Profitability?,” American Economic Review, 98(1). Goldberg, P., A. Khandelwal, N. Pavcnik, and P. Topalova (2008): “Multi-Product Firms and Product Turnover in the Developing World Evidence from India,” Working Paper. Gray, W. (1987): “The Cost of Regulation: OSHA, EPA, and the Productivity Slowdown,” The American Economic Review, 77. Gray, W., and R. Shadbegian (2002): The Economic Costs and Consequences of Regulationchap. Pollution Abatement Costs, Regulation, and Plant Level Productivity. Ashgate Publishing, Aldershot, UK. (2003): “Plant vintage, technology, and environmental regulation,” Journal of Environmental Economics and Management. 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 Manufacturers,” Journal of Political Economy. Hopenhayn, H. (1992): “Entry, Exit, and Firm Dynamics in Long Run Equilibrium,” Econometrica, 60, 1127–50.

30

Iacovone, L., and B. Javorcik (2008): “Multi-Product Exporters: Diversification and Industry Level Dynamics,” World Bank Working Paper. Levinsohn, J., and A. Petrin (2003): “Estimating Production Functions using Inputs to Control for Unobservables,” Review of Economic Studies, 70(2). List, J. A., D. L. Millimett, P. G. Fredriksson, and W. McHone (2003): “Effects of Environmental Regulations on Manufacturing Plant Births: Evidence from a Propensity Score Matching Estimator,” Review of Economics and Statistics, 85(4). Maleug, D. (1989): “Emission Credit Trading and the Incentive to Adopt New Pollution Abatement Technology,” Journal of Environmental Economics and Management, 16(1). Melitz, M. J. (2003): “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity,” Econometrica, 71(6). Murty, M., and S.Kumar (2002): “Measuring the Cost of Environmentally Sustainable Development in India: a Distance Function Approach,” Environment and Development Economics, 7. (2003): “Win win opportunities and environmental regulation: testing of porter hypothesis for Indian manufacturing industries,” Journal of Environmental Management, 67. Nocke, V., and S. Yeaple (2006): “Globalization and Endogenous Firm Scope,” NBER Working Paper, (12322). OECD (2006): “Environmental Compliance and Enforcement in India: a Rapid Assessment,” OECD Programme of Environmental Cooperation with Asia. Olley, S., and A. Pakes (1996): “The Dynamics of Productivity in the Telecommunications Equipment Industry,” Econometrica, 64(6). Pargal, S., M. Mani, and M. Huq (1997): “Inspections and Emissions in India; Puzzling Survey Evidence about Industrial Pollution,” World Bank Working Paper, 1810. Porter, M., and C. V. der Linde (1995): “Toward a new Conception of the EnvironmentCompetitiveness Relationship,” The Journal of Economic Perspectives, 9.

31

Schoar, A. (2002): “Effects of Corporate Diversification on Productivity,” Journal of Finance. Smith, K., and R. Walsh (2000): “Do Painless Environmental Policies Exist?,” Journal of Risk and Uncertainty, 21(1). WHO (2005): “Air Quality Guidelines: Global Update 2005,” World Health Organization Programme on Health and the Environment.

A

Appendix–Derivation of Comparative Statics Results of the indifference to variety Cournot Model

I first check that the assumptions of the implicit function theorem are satisfied. We have a continuously differentiable function over a non-empty open set, so if the firm produces both products, then the derivative function should be invertible. I take the derivatives of equations 24, 25, and 26 and after some equation manipulation and simplifying, I find: dϕ∗∗ x(ϕ ∗ ∗∗) dϕ∗∗∗ dϕ∗ dG − x(ϕ ∗ ∗) dG = ) dG x(ϕ∗)2 − x(ϕ∗) − C(1+bN bϕ∗2 dϕ∗∗∗

(27)

dϕ∗

1 + x(ϕ ∗ ∗∗) dG + x(ϕ∗) dG dϕ ∗ ∗ = ) d q(ϕ∗∗) dG x(ϕ ∗ ∗) − c(1+bN − + x(ϕ ∗ ∗)2 + 2bπ1+bM bϕ∗∗2  dϕ ∗ ∗ = dG

 x(ϕ ∗ ∗∗) 1 +

x(ϕ∗) C(1+bN ) x(ϕ∗)2 −x(ϕ∗)− bϕ∗

x(ϕ ∗ ∗) +

 − x(ϕ ∗ ∗∗) +

(28)

2d π ϕ∗∗2 d

C(1+bN ) bϕ∗∗∗2

x(ϕ∗)x(ϕ∗∗) C(1+bN ) x(ϕ∗)2 −x(ϕ∗)− bϕ∗



dϕ∗∗∗ dG

−2 (29)

Further manipulation of the equations gives that under the assumption of a uniform distribution, for most parameter values we have: show that

B

dq dG

> 0, but the sign of

dx dG

dϕ∗ dG

< 0,

dϕ∗∗ dG

> 0 and

dϕ∗∗∗ dG

> 0. As a result, we can also

depends on parameter values.

Appendix–Derivation of Comparative Statics Results of the love of variety Cournot Model

Because the denominator of the reaction function for x contains δ and the denominator of the reaction function for q contains γ, the model quickly becomes complicated.

32

The first order conditions for the profit function with the reaction functions (after some simplification) are as follows:  a − beX−j − c    a − beX−j − ϕc dπ c δbe ϕ = a−b(1−e) −beX− −1 1− −Gγ−F (2δ+γ) = 0 (30) dδ 2b(1 + 2e (δ − 1)) 2b(1 + 2e (δ − 1)) ϕ 2b(1 + 2e (δ − 1))

 a − beQ−j − d    a − beQ−j − ϕd dπ d γbe ϕ = a − b(1 − e) − beQ − − 1 1 − − F δ − G(2γ + δ) = 0 e e e dγ 2b(1 + 2 (γ − 1)) 2b(1 + 2 (γ − 1)) ϕ 2b(1 + 2 (γ − 1)) (31)

Because it is not possible to derive closed form solutions for the second stage of the model, I will not be able to fully solve the first stage of the model for the equilibrium values of X and Q. However, we know that the first stage of the model would be solved by setting expected profits equal to 0. On entering, the firm realizes its value of productivity ϕ, and decides to either produce both products, produce only the clean product, produce only the dirty product, or drop out of the market. Which products the firm produces will depend directly on the region of the productivity distribution to which the firm belongs. The entry decision is therefore:

E[π] = E[π|δ > 0, γ > 0] + E[π|δ > 0, γ = 0] + E[π|γ > 0, δ = 0]

(32)

We can see immediately that there will be stratification of the productivity distribution as the lowest productivity firms will only be able to produce the product with the lowest fixed cost. Therefore profits from producing varieties of the low fixed cost product will be bid down, and highly productive firms will not want to produce that product. Firms above the cutoff productivity for producing the high fixed cost product must then decide between producing only the high fixed cost product or both the high fixed cost product and the low fixed cost product. Firms at the top of the distribution will begin by producing only the high fixed cost product, until marginal profits are equal in the production of either product for the marginal firm. The cutoff productivity level for a unit of the pollution-intensive product increases unambiguously when the fixed cost of producing the pollution-intensive product increases. Therefore, there is a decrease in the production of the dirty product and an increase in the production of the clean dQ product when there is an increase in G. That is, dX dG > 0 and dG < 0. I will use the implicit function theorem to generate comparative static results for δ and γ. In taking the derivatives, for simplification purposes I assume that the firm views their reaction dX as a relatively small fraction of the reaction of the group as a whole. That is, dG−j ≈ dX dG and

33

dQ−j dG 

≈

dQ dG .

    dδ 1 + δbe bex c 1 beδ = (a−b(1−e)x−beX− −1) −2 +bex2 ( −1)(1− ) e e e e 2b(1 + 2 (δ − 1)) ϕ 2b(1 + 2 (δ − 1)) 2b(1 + 2 (δ − 1)) 2b(1 + 2 (δ − 1)) dG

 −be 1 −

  b(1 − e) − a − b(1 − e)x − beX − δbe 2b(1 + 2e (δ − 1)) 2b(1 + 2e (δ − 1))

c ϕ

−1

  −x

dγ dδ dX + γ + (G + F ) + 2F dG dG dG

(33)

and



    dγ beq 1 + γbe d 1 beγ (a−b(1−e)q−beQ− −1) −2 +beq 2 ( −1)(1− ) = e e e e 2b(1 + 2 (γ − 1)) ϕ 2b(1 + 2 (γ − 1)) 2b(1 + 2 (γ − 1)) 2b(1 + 2 (γ − 1)) dG

 −be 1 −

  b(1 − e) − a − b(1 − e)q − beQ − γbe 2b(1 + 2e (γ − 1)) 2b(1 + 2e (γ − 1))

d ϕ

 −1

 −q

dQ dδ dγ + δ + (G + F ) + 2G dG dG dG

(34)

I name parts of the equations in order to simplify: ΩC =



    bex c beδ 1 + δbe 1 (a−b(1−e)x−beX− −1) −2 +bex2 ( −1)(1− ) −2F 2b(1 + 2e (δ − 1)) ϕ 2b(1 + 2e (δ − 1)) 2b(1 + 2e (δ − 1)) 2b(1 + 2e (δ − 1))

ψC

ΩD =



  b(1 − e) − a − b(1 − e)x − beX − δbe = −be 1 − 2b(1 + 2e (δ − 1)) 2b(1 + 2e (δ − 1)) 

c ϕ

 −1

 −x

    1 + γbe beq d beγ 1 (a−b(1−e)q−beQ− −1) −2 +beq 2 ( −1)(1− ) e e e e 2b(1 + 2 (γ − 1)) ϕ 2b(1 + 2 (γ − 1)) 2b(1 + 2 (γ − 1)) 2b(1 + 2 (γ − 1))

ψD

 = −be 1 −

  b(1 − e) − a − b(1 − e)q − beQ − γbe 2b(1 + 2e (γ − 1)) 2b(1 + 2e (γ − 1))

d ϕ

 −1

 −q

Then we have ΩC

dX dγ dδ = ψC + γ + (G + F ) dG dG dG

(35)

ΩD

dγ dQ dδ = ψD + δ + (G + F ) dG dG dG

(36)

and

Solving for

dγ dG

we get: dγ ψD dQ δ (G + F )ψD dδ = + + dG ΩD dG ΩD ΩD dG

(37)

  dδ 1 dX dQ = Ω ψ + γΩ + (G + F )ψ + δ D C D D dG ΩC ΩD − (G + F )2 dG dG

(38)

Which yields:

  dγ ψD dQ δ G+F dX dQ = + + Ω ψ + γΩ + (G + F )ψ + δ D C D D dG ΩD dG ΩD ΩC ΩD − (G + F )2 dG dG

34

(39)

It can be shown that for most parameter values ΩC < 0, ΩD < 0, ψC > 0, and ψD > 0. Then we can show that for most parameter values

dδ dG

> 0 and

dγ dG

< 0. We can also use the implicit

function theorem on the reaction functions to see the effect of a change in G on x and q. We get the following results: eb(a − beX−j − ϕc ) dδ dX be dx =− − dG 2b(1 + 2e (δ − 1)) dG (2b(1 + 2e (δ − 1)))2 dG

(40)

eb(a − beQ−j − ϕd ) dγ dq dQ be =− − dG 2b(1 + 2e (γ − 1)) dG (2b(1 + 2e (γ − 1)))2 dG

(41)

Given the comparative static results derived above, this means that

dx dG

< 0 and

dq dG

> 0.

This makes intuitive sense: we would expect the amount produced per variety to change in the opposite direction of the change in the quantity of varieties holding the market effect constant. (As the firm produces more varieties, it sells fewer pieces of each variety and acts more as a monopolist). We would also expect the quantity produced to move opposite of the direction that market quantity moves holding number of varieties constant: as there is more competition in the market, the firm produces a lower quantity per variety.

C

Tables Table 3: Summary Statistics

State enforcement Number dirty products Number clean products Number products Percent clean Added clean Added dirty Dropped clean Dropped dirty

Obs

Mean

13303 21541 21541 21541 21541 21164 21164 16730 8675

.095 .805 2.26 3.06 .696 .137 .041 .172 .110

35

Std. Dev. .077 1.35 2.33 2.80 .421 .343 .199 .377 .313

Min

Max

.008 0 0 1 0 0 0 0 0

.413 17 34 41 1 1 1 1 1

Table 4: Productivity Effect of Changes in Environmental Enforcement OLS Productivity All Companies Heavily Monitored Lagged State Enforcement National Enforcement Change r2 N

-0.031 (0.06)

Olley Pakes Productivity All Companies Heavily Monitored -0.051 (0.07)

0.795 10021.000

-0.375*** (0.05)

-0.494*** (0.11)

0.822 14606.000

0.858 1299.000

0.008 10022.000

-0.376*** (0.05)

-0.664*** (0.11)

0.019 14607.000

0.084 1299.000

*** p<0.01, ** p<0.05, * p<0.1 Dependent variable is the output for OLS productivity with production variables on the RHS. Dependent variable is Olley Pakes Productivity run in a separate stage for second two regressions. The independent variable of interest is the percent of companies closed by regulators in the state in the prior year for state enforcment and for national enforcement an indicator for years in which the increase in national environmental enforcement applied (1997-2005) multiplied by the industry level input output coefficient for the sum of the 17 pollution-intensive industries which are highly regulated in India. Fixed effects by firm and year dummies are included. Standard errors are clustered by state, year.

Table 5: Firm Level Product Adding and Dropping Effects of State Enforcement

Lagged State Enforcement r2 N

Add Dirty -0.007 (0.04) 0.002 13841

Add Clean 0.041 (0.06) 0.004 13841

OLS Drop Dirty 0.208*** (0.07) 0.004 7071

Drop Clean -0.002 (0.10) 0.001 12465

Add Dirty -0.151 (1.95)

Add Clean 0.475 (1.00)

2172

6477

Logit Drop Dirty 3.225* (1.74) 2374

Drop Clean -0.006 (0.91) 6355

*** p<0.01, ** p<0.05, * p<0.1 Dependent variable is an indicator for whether a clean or dirty product was added. Fixed effects by firm and year are included. Standard errors are clustered at the state year level.

Table 6: Product Adding and Dropping, product level regression

Lagged State Enforcement r2 N

Add Dirty -0.015 (0.05) 0.002 21712

Linear Probability Model Add Drop Drop Clean Dirty Clean -0.004 0.060** -0.004 (0.02) (0.03) (0.02) 0.001 0.001 0.000 70439 21712 70439

Add Dirty -0.366 (1.70)

Logit Add Clean -0.239 (0.79)

3319

13469

Model Drop Dirty 2.111 (1.45)

Drop Clean -0.027 (0.69)

4784

15092

*** p<0.01, ** p<0.05, * p<0.1 Standard errors are clustered at the state year level. Regressions include controls for firm size and whether or not the firm is foreign owned. Both the linear probability model and logit model have fixed effects by firm-product and year dummies. Dependent variable is an indicator for the year in which the product was added or dropped.

36

Table 7: Product Adding and Dropping in firms producing both clean and dirty versus only dirty, product level regression

Lagged State Enforcement r2 N

Added Dirty Products General Specialized -0.087 0.140** (0.06) (0.07) 0.002 0.004 15073 4338

Dropped Dirty Products General Specialized 0.140*** -0.056 (0.04) (0.07) 0.002 0.002 15073 4338

*** p<0.01, ** p<0.05, *p<0.1 Standard errors are clustered at the state year level. Regressions include controls for firm size and whether or not the firm is foreign owned. Regression is linear probability model and includes fixed effects by firmproduct and year dummies. Dependent variable is an indicator for taking the value 1 in the year in which the product was added or dropped.

Table 8: Heterogeneous Product Dropping Effects

Lagged Low Productivity Enforcement Effect Lagged Mid Productivity Enforcement Effect Lagged High Productivity Enforcement Effect r2 N F-stat for test that low coef=mid coef p-value F-stat for test that mid coef=high coef p-value

Drop Clean OLS 0.274* (0.15) 0.092 (0.09) -0.234 (0.19) 0.001 12259 0.91 0.342 1.57 0.213

Dropped Drop Clean LevPet 0.439*** (0.16) 0.106 (0.08) -0.472** (0.18) 0.001 12259 2.44 0.120 5.84 0.017

Products Drop Dirty OLS -0.946*** (0.34) 0.041 (0.06) 0.527*** (0.17) 0.005 6652 7.57 0.007 7.37 0.007

Drop Dirty LevPet -1.089** (0.48) 0.052 (0.06) 0.345 (0.25) 0.004 6652 5.99 0.016 1.21 0.273

*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses. Fixed effects by firm and year are included but not reported. Standard errors are clustered by state year. The dependent variable is an indicator that takes the value 1 in the year that a product was dropped, 0 otherwise. The year of firm exit is omitted from the sample. The independent variables are the firm’s productivity grouping as of its first year in the sample interacted with the lagged state enforcement variable.

37

Table 9: Product Quantities, product level regression

Lagged State Enforcement Size Foreign r2 N

Within Variety Product Quantities All Producers Specialized Producers Dirty Prod- Clean Prod- Dirty Only Clean Only ucts ucts -0.093 -0.239 1.761* -0.932** (0.47) (0.24) (1.01) (0.38) 0.418*** 0.335*** 0.442*** 0.387*** (0.06) (0.04) (0.08) (0.05) 0.038 0.091* -0.074 0.049 (0.05) (0.05) (0.15) (0.07) 0.013 0.009 0.023 0.013 9704 25754 2391 14692

*** p<0.01, ** p<0.05, * p<0.1 Dependent variable is the log of quantity produced. Fixed effects by firm-product and year dummies are included. Controls for firm size and foreign ownership included, but not reported. Standard errors are clustered at the state year level.

Table 10: Product Quantities differences across the productivity distribution, product level regression

Lagged Low Productivity Enforcement Effect Lagged Mid Productivity Enforcement Effect Lagged High Productivity Enforcement Effect r2 N F Stat for test that low coef=mid coef p-value F Stat for test that mid coef=high coef p-value

Dirty -0.142 (0.12) -0.081 (0.05) 0.272 (0.21) 0.015 7889 0.34 0.560 2.83 0.095

Within Variety Product Quantities All Producers Specialized Producers Clean Dirty Only Clean Only 0.105 -0.404 -5.638*** (0.21) (2.17) (1.47) -0.017 1.736 -1.295*** (0.04) (1.39) (0.43) 0.037 4.203* -0.621 (0.07) (2.31) (0.56) 0.011 0.023 0.014 20919 2319 14322 0.35 0.71 8.35 0.553 0.400 0.005 0.83 1.23 1.80 0.364 0.269 0.183

*** p<0.01, ** p<0.05, * p<0.1 Dependent variable is the log of quantity produced. Fixed effects by firm-product and year dummies are included. Controls for firm size and foreign ownership included, but not reported. Standard errors are clustered at the state year level. The dependent variable is the log of quantity produced of each product by each firm in the year of observation. The independent variables are the firm’s productivity grouping as of its first year in the sample interacted with the lagged state enforcement variable.

38

Table 11: Revenue from dirty or clean products, firm level regression

Lagged State Enforcement Lagged Low Productivity Enforcement Effect Lagged Mid Productivity Enforcement Effect Lagged High Productivity Enforcement Effect r2 N F-Stat for low coef=mid coef p-value F-stat for mid coef=high coef p-value

Dirty Sales Revenue General Specialized -1.140*** 0.021 (0.43) (0.27) -0.812 -0.835 (1.88) (1.77) -1.367*** -0.120 (0.45) (0.35) 1.214 0.865** (0.93) (0.38) 0.024 0.025 0.118 0.118 3767 3767 1621 1621 0.09 0.15 0.769 0.703 6.60 4.45 0.011 0.037

Clean Sales General -0.233 (0.38)

0.015 4014

-2.754** (1.24) -0.900* (0.46) 3.792** (1.78) 0.018 4014 1.89 0.171 5.74 0.018

*** p<0.01, ** p<0.05, * p<0.1 Dependent variable is the overall revenue from dirty or clean goods sold by the firm in the year of the observation. Fixed effects by firm-product and year dummies are included. The independent variables are the firm’s productivity grouping as of its first year in the sample interacted with the lagged state enforcement variable. Standard errors are clustered at the state year level.

Table 12: Percent Revenue from clean products, firm level regression

Lagged Low Productivity Enforcement Effect Lagged Mid Productivity Enforcement Effect Lagged High Productivity Enforcement Effect r2 N F-Stat low coef=mid coef p-value F-Stat high coef=mid coef p-value

OLS 0.121 (0.19) -0.016 (0.03) 0.231** (0.10) 0.003 4403 0.57 0.451 7.32 0.008

Percent Total Revenue from Clean Sales Robustness: future enforcement LevPet OLS LevPet -0.042 0.263 -0.020 (0.34) (0.21) (0.32) -0.009 -0.022 -0.010 (0.03) (0.03) (0.02) 0.261*** -0.013 -0.018 (0.10) (0.04) (0.04) 0.003 0.001 0.001 4403 5430 5430 0.01 1.78 0.00 0.921 0.184 0.977 8.00 0.04 0.02 0.005 0.851 0.875

*** p<0.01, ** p<0.05, * p<0.1 Dependent variable is the percent of revenue from clean products in the year of the observation. The independent variables are the firm’s productivity grouping as of its first year in the sample interacted with the lagged state enforcement variable. Fixed effects by firm and year dummies are included. Standard errors are clustered at the state year level.

39

Table 13: Investment among Producers of Pollution-Intensive Products, firm level regression

Lagged State Enforcement

Log Investment OLS

LevPet

-0.801 (1.66) 1.081 (0.67) 1.709 (1.06) 0.088 1310 1.00 0.318 0.29 0.589

-4.844 (5.02) 1.103* (0.63) 1.420 (1.00) 0.088 1310 1.39 0.241 0.09 0.771

1.138** (0.49)

Lagged Low Productivity Enforcement Effect Lagged Mid Productivity Enforcement Effect Lagged High Productivity Enforcement Effect r2 N F-stat for test that low coef=mid coef p-value F-stat for test that mid coef=high coef p-value

0.088 1310

*** p<0.01, ** p<0.05, * p<0.1 Dependent variable is the log of investment in the year of the observation. The independent variables are the firm’s productivity grouping as of its first year in the sample interacted with the lagged state enforcement variable. Fixed effects by firm and year dummies are included. Standard errors are clustered at the state year level.

Table 14: Impact of Enforcement on Profitability of Firms Producing the Pollution-Intensive Product

Lagged State Enforcement Lagged Low Productivity Enforcement Effect Lagged Mid Productivity Enforcement Effect Lagged High Productivity Enforcement Effect r2 N F-stat for test low coef=mid coef p-value F-stat for test mid coef=high coef p-value

Firms Producing OLS -0.276 (0.32) -2.143 (1.58) -0.380 (0.36) 2.458*** (0.85) 0.113 0.116 4300 4300 1.60 0.208 7.63 0.007

Dirty LevPet

-3.240 (2.44) -0.243 (0.31) 2.382** (0.99) 0.116 4300 1.71 0.193 5.62 0.019

Firms only Producing Dirty OLS LevPet 0.056 (0.52) -3.149 -6.025 (2.28) (3.92) -0.155 -0.484 (0.56) (0.56) 2.743 4.634 (2.77) (2.80) 0.106 0.110 0.115 1191 1191 1191 1.60 1.98 0.209 0.162 1.00 2.89 0.321 0.092

*** p<0.01, ** p<0.05, * p<0.1 Dependent variable is the log of operating profit in the year of the observation. The independent variables are the firm’s productivity grouping as of its first year in the sample interacted with the lagged state enforcement variable. Fixed effects by firm and year dummies are included. Standard errors are clustered at the state year level.

40

Table 15: OLS Production Coefficients NIC2 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

Industry Food Processing Manufacture of Tobacco Cloth Preparation Garments Manufacture Leather Products Wood Products Paper Products Printing Coke and Fuel Chemicals Plastics and Rubber Concrete and Cement Iron and Steel Metal Products Machinery and Tools Office Machinery Industrial Equipment Electrical Components Medical Equipment Motor Vehicles Major Transport Equipment Furniture

LogLabor 0.240 0.092 0.159 0.174 0.315 0.285 0.282 0.375 0.428 0.279 0.311 0.368 0.272 0.230 0.319 0.546 0.210 0.430 0.543 0.257 0.172 0.103

LogMaterials 0.725 0.839 0.675 0.833 0.677 0.749 0.700 0.611 0.608 0.624 0.644 0.379 0.668 0.784 0.575 0.353 0.708 0.594 0.387 0.627 0.750 0.704

LogCapital 0.029 0.190 0.145 0.033 0.119 0.008 0.111 0.037 0.081 0.136 0.063 0.333 0.068 -0.021 0.116 0.102 0.121 0.024 0.051 0.124 0.149 0.097

Returns to Scale 0.994 1.120 0.979 1.040 1.111 1.042 1.092 1.023 1.117 1.039 1.019 1.080 1.007 0.992 1.010 1.002 1.038 1.048 0.981 1.007 1.072 0.903

The top and bottom 1% observations for capital as a percentage of sales in each industry have been dropped.

41

Table 16: Levinsohn Petrin Production Coefficients NIC2 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

Industry Food Processing Manufacture of Tobacco Cloth Preparation Garments Manufacture Leather Products Wood Products Paper Products Printing Coke and Fuel Chemicals Plastics and Rubber Concrete and Cement Iron and Steel Metal Products Machinery and Tools Office Machinery Industrial Equipment Electrical Components Medical Equipment Motor Vehicles Major Transport Equipment Furniture

LogLabor 0.196 0.235 0.156 0.180 0.121 0.178 0.258 0.389 0.464 0.214 0.217 0.189 0.119 0.157 0.256 0.675 0.177 0.410 0.472 0.232 0.193 0.044

LogMaterials 0.710 0.882 0.741 0.828 0.705 0.711 0.646 0.617 0.520 0.661 0.677 0.427 0.711 0.780 0.582 0.395 0.718 0.646 0.402 0.636 0.704 0.738

LogCapital 0.113 0.253 0.098 0.189 0.865 0.234 0.029 0.393 0.186 0.124 0.173 0.152 0.161 0.014 0.152 0.182 0.126 0.175 0.135 0.154 0.410 0.324

Returns to Scale 1.019 1.370 0.995 1.197 1.692 1.123 0.933 1.398 1.170 0.999 1.067 0.769 0.991 0.951 0.990 1.252 1.021 1.230 1.009 1.022 1.308 1.107

The top and bottom 1% observations for capital as a percentage of sales in each industry have been dropped.

42