NCEE Working Paper
Environmental Regulations and
Technological Change: Pulp and
Paper Mills and EPA's Cluster Rule
Anna Belova, Brendan Casey, Cynthia Morgan,
Carl Pasurka, and Ron Shadbegian
Working Paper 20-03
November, 2020
U.S. Environmental Protection Agency fgf
National Center for Environmental Economics fw
https://www.epa.aov/environmental-economics e n v' r o n m e n ta1Re con o m i cs
-------
Environmental Regulations and Technological Change:
Pulp and Paper Mills and EPA's Cluster Rule
Anna Belova3, Brendan Casey3, Cynthia Morganb, Carl Pasurkab, and Ron Shadbegianc
a ICF, Fairfax, VA 22013
b U.S. Environmental Protection Agency, Washington, DC 20460
c Department of Economics, San Diego State University, San Diego, CA 92182
Abstract: In this paper we propose a new model to calculate changes in pollution abatement
costs (PAC) when data on bad outputs and information on the cost of inputs assigned to
pollution abatement activities are not available. To calculate the PAC of reducing bad output,
we introduce a measure of abatement intensity that captures variation in the technologies used
to reduce water discharges. We then decompose the change in PAC into three components to
identify their relative contribution to changes in PAC: (1) changes in the level of inputs, (2)
technological change, and (3) changes in pollution abatement intensity. These three
components are estimated using data from 1997 to 2007 on a sample of pulp mills required to
comply with effluent limits of the U.S. EPA's Cluster Rule. We find technological change is
consistently associated with declining PAC, while both changes in inputs and abatement
intensity are associated with increasing PAC.
KEYWORDS: technological change, pulp and paper mills, Cluster Rule, pollution abatement costs,
abatement intensity
JEL CODES: 033, Q52, Q53
DISCLAIMER
The views expressed in this paper are those of the author(s) and do not reflect those of the U.S.
Environmental Protection Agency (EPA) or the U.S. Census Bureau. This research was performed at a
Federal Statistical Research Data Center under FSRDC Project Number FSRDC1191. All results using
Census data have been reviewed to ensure that no confidential information is disclosed. Also, the
research described in this paper has not been subjected to the U.S. EPA's required peer and policy
review so no official EPA endorsement should be inferred.
1
-------
I. Introduction1
Most regulations promulgated by the U.S. Environmental Protection Agency (U.S. EPA)
focus on pollution releases into one environmental medium (e.g., land, air or water). However,
often a production process releases pollutants into more than one medium. A good example is
the pulp and paper-making process. Pulping involves separating lignin (glue) from wood pulp
fibers and then mixing these fibers with water to create a slurry. The slurry is washed multiple
times, dried and then whitened with bleach to make paper. Both processes result in
conventional and toxic air and water pollutants.2 Recognizing more cost-effective pollution
reductions could be achieved by jointly setting air and water standards rather than instituting
separate regulations for releases to each medium, EPA issued its first integrated, multi-media
regulation- known as the "Cluster Rule" (CR) - in 1998 (U.S. EPA,1998).
One regulatory objective of the CR was limiting the formation of dioxin and furans from
the pulp and paper-making process. The elemental chlorine used to bleach pulp reacts with
organic compounds in the pulp (lignin) to form dioxins, furans, and chloroform in the
wastewater stream.3 The technology-based approaches considered by the EPA included 1)
reducing the chlorine bleaching required via oxygen delignification (OD) and extended
delignification (ED) and 2) substituting elemental chlorine-free bleaching (ECF) for chlorine
1 We would like to thank Jim Davis at the Boston Research Data Center for his continued help; Wayne Gray at Clark
University for his valuable input and help with the data; Jordan Marvakov, Sergey Kazakov, and Kaushik Ghosh for
excellent research assistance; Lars Vilhuber at Cornell Virtual Research Data Center for help with compilation of R
packages for use at the Research Data Center; and John Haltiwanger for providing us with capital stock data. We
would also like to thank Will Wheeler for his helpful comments and suggestions.
2 Conventional and toxic air and water pollutants include particulate matter, sulfur dioxide, and benzene, while
conventional and toxic water pollutants include biochemical oxygen demand and dioxin.
3 Dioxins and furans, which accumulate in human fatty tissue, are characterized by the EPA as highly toxic and can
cause cancer, reproductive and development problems, damage to the immune system, and can interfere with
hormones (see U.S. EPA 2018a).
2
-------
bleaching. Both OD and ED decrease the amount of lignin in the pulp prior to the bleaching
stage, which reduces the amount of bleaching chemicals needed to brighten the pulp, while ECF
replaces elemental chlorine with chlorine dioxide (CIO2) as the bleaching agent.4 When EPA
published the final CR regulations in 1998, the technology basis for the effluent limits of the CR
required pulp and paper mills to switch from using elemental chlorine as the bleaching agent to
ECF.
We are interested in how this change in bleaching technology affected the cost of
producing pulp and paper. Two approaches have been developed in the extant literature to
calculate the opportunity cost of producers being prohibited from freely disposing of their
undesirable by-products. The first model - the joint production model - specifies the joint
production of good and bad outputs with two technologies - one when the producer can freely
dispose of its bad outputs (i.e., the unregulated technology) and one when the producer may
not freely dispose of its bad outputs (i.e., the regulated technology) (see Fare et al., 2007). One
advantage of this approach is it does not require information on the pollution abatement
technology. Instead, using information on total inputs and the production of the good and bad
output, the opportunity cost of pollution abatement is the foregone good output due to the
bad output not being freely disposable.
In contrast, the second model - the assigned input model - requires information on the
inputs assigned to good output production and pollution abatement, information on good
4 Unlike chlorine, which combines with lignin to create dioxin and furans, chlorine dioxide breaks apart the lignin,
resulting in organic, more water-soluble compounds. While ECF does not completely eliminate dioxin formation, its
use greatly reduces the amount created.
3
-------
output production, but no information about bad output production (see Aiken et al., 2009;
Gray and Shadbegian, 2003). When inputs assigned to good output production are switched to
pollution abatement, the subsequent decrease in good output production reflects the
opportunity cost of pollution abatement.
In our case, we have neither data on bad output production (e.g., dioxin) nor the
quantity of inputs assigned to pollution abatement, so instead of using a joint production or
assigned input model, we use novel measures of pollution abatement intensity to calculate the
opportunity cost of pollution abatement.5 We construct pollution abatement intensity
measures based on the regulated technology, which incorporates the non-chlorine bleaching
technologies used by mills to reduce dioxin. We have a unique dataset that allows us to
identify when OD and ED systems were installed and the extent of use of ECF or CIO2
substitution at mills subject to the effluent limits of the Cluster Rule starting in 1997. Using
these data, we introduce three measures of pollution abatement intensity that capture
variation in the levels of ED, OD and ECF employed by pulp mills to comply with CR limits.
These measures of abatement intensity serve as proxies for the reduction in dioxin releases by
mills. That is, we assume that as more pulp production is subject to these technologies,
abatement intensity increases which results in less dioxin being generated and released.
Gollop and Roberts (1983) developed a measure of abatement intensity to estimate the
effect of restrictions on sulfur dioxide emissions on plant-level productivity of U.S. electric
5 The EPA's Toxic Release Inventory (TRI) database contains facility-level annual data on the quantity and type of
toxic releases for an extensive range of hazardous substances. EPA did not require monitoring of dioxins before
the Cluster Rule and releases of dioxin and dioxin-related compounds were not collected by the TRI until 2000, by
which time many facilities had already achieved the reductions required by the CR. For information on EPA's TRI
database, please see https://www.epa.gov/toxics-release-inventorv-tri-program.
4
-------
utilities. They found that an increase in regulatory intensity increases abatement costs. We are
also interested in how an increase in regulatory intensity affects pollution abatement costs.
The increase in the level of abatement needed to comply with the CR limits increases
abatement costs as more inputs are switched from producing the good output to reducing the
bad output. However, technological change may moderate the increases in pollution
abatement costs (PAC) associated with the regulation.6
To identify the relative importance of the factors that affect changes in PAC, we
decompose PAC into the following components (1) changes in inputs, (2) changes in technology,
and (3) changes in pollution abatement intensity.7 The hypothesis of whether technological
change can moderate pollution abatement costs can be evaluated by examining the magnitude
of the technological change component of the decomposition. To our knowledge, because of
our distinct dataset, we are the first to develop a measure of regulatory stringency to examine
how changes in abatement technology affected the costs of complying with the effluent limits
of the CR.
Several studies have explored how the CR affected technological change in the pulp and
paper industry. Norberg-Bohm and Rossi (1999) argue the use of ECF by pulp and paper mills
represents an incremental change in technology and the EPA, in designing the CR, did not push
for more sweeping innovations such as totally-chorine free bleaching (TCF). 8 However, the
6 Some have referred to this as the 'weak' version of the Porter Hypothesis - see Jaffe and Palmer (1997).
7 Fare et al. (2016) decompose changes in PAC according to (1) changes in inputs, (2) technical change, (3) and
changes in bad output production. Instead of changes in bad output production, changes in PAC are affected by
changes in pollution abatement intensity in our decomposition.
8 Mills could have used TCF, which uses hydrogen peroxide and ozone as bleaching agents to eliminate the
formation of dioxin, to meet the CR water standards. However, EPA demonstrated that TCF was not technically
feasible for all mill categories, i.e., ammonia-based sulfite mills, whereas ECF was both technically and
economically feasible. Because TCF was feasible for some segments, EPA included voluntary alternative standards
5
-------
incremental technological change to ECF had an upstream effect, namely changing the
technology used to produce chlorine. Snyder et al. (2003) find the CR decreased the demand
for chlorine, resulting in a significant increase in the number of closures among plants that did
not adopt the cleaner, membrane technology to produce chlorine. The CR also had a
downstream effect, changing the mix of bleached and unbleached products. Elrod and Malik
(2017) find evidence that some mills, which faced the both air and water regulations of the CR,
substituted away from bleached products instead of changing their production processes
compared to mills that faced only the air regulations.
Using patent data on technologies designed to reduce dioxin from the pulp bleaching
process in Canada, Finland, Japan, Sweden and the United States, Popp et al. (2011) find public
pressure played a larger role than environmental regulation in the development and diffusion
of ECF and TCF technologies. Because of growing health concerns about dioxin, public pressure
led many mills to voluntarily adopt technologies that used less elemental chlorine (Maynard
and Shortle, 2001). For example, some mills installed bleach-reducing technologies such as OD
or ED systems in the early 1990's.
The remainder of our paper is organized as follows. Section II provides an overview of
the pulp and paper industry and the Cluster Rule, while Section III specifies the regulated and
unregulated production functions used to calculate PAC. Section IV presents the data, Section V
discusses our estimation approach and the results, and Section VI summarizes our findings.
using TCF in order to encourage mills to use the bleaching technology whenever it is consistent with the type of
pulping process they used. In the end, only the Samoa Mill in California used TCF.
6
-------
II. Pulp and Paper Industry and EPA's "Cluster Rule"
While the pulp and paper industry confronts substantial levels of environmental
regulation, the regulatory burden varies among mills depending on their age, location,
production process used, whether or not the mill includes a pulping facility, and the regulatory
intensity directed at a mill.9 Prior studies, including Gray and Shadbegian (2003), found the
main determinant of regulatory impact on a mill is whether or not it contains a pulping facility,
since the pulping process is significantly more pollution intensive than the paper-making
process.10
There are two methods employed to produce pulp from wood chips - mechanical and
chemical wood pulping - and each process produces a different quality of paper and generates
different types of pollutants. Mechanical pulping uses mechanical energy to separate the lignin
from the wood fibers resulting in more air pollution from a power-generating boiler used to
generate energy. But mechanical pulping generally produces lower quality pulp and
consequently, is used to make lower quality printing paper such as newsprint. Kraft chemical
pulping involves using "cooking" chemicals to separate the lignin from wood fibers. Bleaching
the kraft pulp with elemental chlorine creates dioxins, furans and chloroform. Bleached kraft
pulp is used to produce higher quality printing paper such as copy paper and magazine paper.
On December 5th, 1982 the Meramec River flooded Times Beach, Missouri
contaminating virtually the entire town with dioxin that was in the oil the town contracted to
9 Integrated mills make their own pulp, while non-integrated mills buy pulp or use recycled wastepaper.
10The most important environmental concerns during the paper-making phase are air pollution, if the mill
produces electricity via a cogeneration system, and with the residual water pollution produced throughout the
drying process.
7
-------
have sprayed on its unpaved roads to alleviate dust in the early 1970's. As a result of this
incident, two influential environmental groups, the Environmental Defense Fund (EDF) and the
National Wildlife Federation (NWF), sued the EPA for not adequately protecting the U.S. public
from the risks caused by dioxin. EPA, as part of a 1988 settlement with the EDF and NWF,
agreed to examine the health risks of dioxin and to promulgate regulations to decrease dioxin
emissions (Environmental Defense Fund and National Wildlife Federation vs. Thomas, D.D.C.
No. 85-0973). In 1997, as part of the CR, the EPA implemented regulations that included dioxin
reductions.
EPA promulgated the CR on April 15,1998, and while most air provisions became
effective on April 16, 2001, best available technology (BAT) standards for water effluents
became effective the next time an existing mill's National Pollutant Discharge Elimination
System (NPDES) permit was renewed.11 Because NPDES permits are generally renewed every
five years, the effective BAT compliance dates for the water provisions were spread over the
1998 to 2003 period. The CR rule affected pulp and paper mill in three subcategories: (1)
bleached papergrade kraft and soda (BPK) mills; (2) papergrade sulfite (PS) mills, and (3) semi-
chemical mills. The rule set limits on releases of toxic, conventional, and nonconventional
pollutants (e.g., chlorine, nitrogen, and phosphorus) to both air and water from the pulp and
paper mills in these subcategories.
11 The rule and implementation information for the air portion of the Cluster Rule can be found at U.S. EPA
(2018b), while information about the Effluent Guidelines for the Cluster Rule can be found at U.S. EPA (2018c).
Rule information can also be found in the Federal Register: https://www.govinfo.gov/content/pkg/FR-1998-04-
15/pdf/98-9613.pdf.
8
-------
For air pollutants, EPA established maximum achievable control technology (MACT)
standards requiring existing mills to capture and treat toxic air pollutant emissions produced
during the cooking, washing, and bleaching stages of the pulp manufacturing process.
Specifically, the CR called for hazardous air pollutants to be reduced by almost 60%. Additional
provisions of the CR included reducing releases of sulfur, volatile organic compounds, and
particulate matter by 47%, 49%, and 37%, respectively. For water pollutants, EPA set effluent
limits - Best Available Treatment (BAT) and Pretreatment Standards for Existing Sources (PSES)
requirements- based on mills substituting ECF for elemental chlorine in the bleaching process.
The CR's technology-based water regulations required a 96% reduction in dioxin and furan, and
a 99% reduction in chloroform (see Morgan et al., 2014, for a detailed discussion of the Cluster
Rule).
Initially, the EPA expected approximately 490 pulp and paper mills would be subject to
the CR air regulations. EPA expected 155 of the 490 pulp and paper mills would need to comply
with the more stringent MACT emission standards, and of those 155 mills, the 96 using
chemical pulping techniques would also be required to comply with a new set of BAT effluent
standards. In the economic analysis conducted in support of the rule, the EPA estimated the
cost for these 96 mills to comply with the BAT standard, expressed in 1995 dollars, was $1,039
billion in capital and $0,158 billion in annual operation and maintenance (O&M) expenses (U.S.
EPA, 1997).
In anticipation of the CR, as well as some state requirements and public pressure to
reduce dioxin releases, some mills installed OD and ED systems in the early 1990's. Of the 37
mills that installed OD systems from the early 1990's through 2008, 24 mills installed this
9
-------
system by 1995. Similarly, 23 of the 33 mills that installed ED systems between 1987 and 2008
did so by 1995 (Beca AMEC, 2013b). Although some mills installed ED/OD systems, other mills
delayed adopting cleaner pulping technologies during the prolonged development of the CR
regulations (the first CR regulations were proposed in 1993). Maynard and Shortle (2001)
found mills delayed investing due to the uncertainty associated with making large irreversible
investments prior to the EPA issuing its final CR regulations.
Converting pulp lines from elemental chlorine bleaching to ECF takes time. Because
chlorine dioxide gas is flammable and may not be transported within the US, chlorine dioxide is
made onsite at the mill. Because it takes from 12 to 24 months to install a new chlorine
dioxide generator and adsorption tower (which optimizes the generation process) or expand
the capacity of the existing generator, full implementation of 100% chlorine dioxide
substitution (i.e., switching to ECF) depends on a mill's initial on-site chlorine dioxide
generation (U.S Congress, 1989, Chapter 4). In 1998, the year the CR was promulgated, only
28% of active mills had completely switched to ECF (50% of active mills had switched at least
50% of their total pulp production to ECF). In 2001, 71% of mills had adopted 100% ECF (85% of
active mills had switched at least 50% of their total pulp production to ECF), and in 2002, 90% of
active mills had completely switched (Beca AMEC, 2013a). Thus, the level of dioxin discharged
into surface waters was below the effluent limitations and standards required by the CR for
most mills by 2001.12
12 The CR limits for NPDES dischargers would have been written into permits upon next issuance, which is generally
every five years. Using the renewal dates from EPA's Permit Compliance System (PCS) to determine when the CR
limits were written into the permits, 60% of the 96 mills permits were renewed between April 15,1998 and 2001.
By 2003, 80% of the permits had been renewed and by 2007,100% of the permits had been renewed.
10
-------
III. Model
In this paper we use a data envelopment analysis (DEA) framework, combined with our
measures of pollution abatement intensity, to calculate the opportunity cost of changing the
bleaching process to comply with the CR. Specifically, we model two production technologies -
one when the bad output is unregulated (or least regulated) and another when the bad output
is regulated - where the regulated technology captures the effect of variation in use of chlorine
dioxide substitution, as well as use of extended and oxygen delignification, among mills. In
other words, the bad output is freely disposable with the unregulated technology, but not
freely disposable with the regulated technology. In this model, the difference in maximum
good output production between the unregulated and regulated technologies constitute the
PAC of reducing dioxin, furans and chloroform loadings to water.
We will now formally specify the unregulated and regulated technologies. As an
example, we present the unregulated and regulated production functions for period t. If we
have k=l,...,K mills with information on xn inputs (n=l,..., N), and one good output yk for t=l,...,T
periods, the unregulated production function for mill k' (FUR) in period t is (see Aiken et al.,
2009):
K
FURfV, t) = max j yŁ
(1)
k=l
K
n = 1 N
k=l
K
k=l
11
-------
> 0 k = 1, , K
The unregulated production technology for mill k' is constructed from all observations (K) that
are available in period t. The linear programming (LP) problem calculates the maximum good
output of mill k' subject to its input constraints and a constraint that imposes non-increasing
returns to scale (i.e., the summation of the weights assigned to each mill when constructing the
production frontier, z, is less than or equal to unity).13 Hence, linear combinations of all
observations subject to the LP constraints are used to calculate the maximum good output of
mill k' in period t. Guaranteeing the maximum good output of the unregulated technology
equals or exceeds the maximum good output of the regulated technology requires employing
sequential frontiers, which construct the frontier for period t with observations from period 1
up to and including period t. In other words, using all observations to construct the
unregulated frontiers eliminates the possibility of the model identifying negative pollution
abatement costs.
To capture variation in chlorine dioxide substitution and the use of OD and/or ED among
pulp lines at a mill, we define the regulated technology by the share of total pulp production in
period t subject to ECF (m=l), OD (m=2), and ED (m=3).14 Specifically, for each line i=l,...,l,
within a mill, we multiply the percent of abatement technology used on line i, aim, by the pulp
production on line i, pi, which yields the amount of pulp production on line i subject to
abatement technology m. For each mill, we sum this value over all its lines of production and
13The occasional occurrence of infeasible LP problems in the decomposition with variable returns to scale led us to
specify non-increasing returns to scale.
14 The pollution abatement intensity measures are patterned after the measure of regulatory intensity developed
by Gollop and Roberts (1983).
12
-------
divide by its total pulp production, which yields the weighted average of total mill production
subject to abatement technology m. Hence, for mill k', the abatement intensity (i.e., the share
of pulp production subject to ECF, OD, or ED) in period t is written as:
Zi=l (aim * Pi) t n n
~t m=l,2,3 (2)
A i=i Pi
Using equation (2) to define the share of pulp production subject to each abatement
technology, the regulated production function (FR) for mill k' in period t is specified as:
K
FR(xt,at,t) = max ^ zŁ yŁ (3)
k=l
S-t Ek=lZk4n n = l, ,N
Zk=lZi=l zk(akim X Pki) ^ t , no
yK yl ytnt ^ ak'm m = 1,2,3
Z,k=l^i=izkPki
k=l
> 0 k = 1, , K
or rewriting
K
FR(xt, at, t) = max^ zŁ (3')
k=l
K
S-t- ^zk4n ^ 4'n< n = 1'
-------
K
-1
k=l
> 0 k = 1, , K
Like the unregulated technology, the regulated production technology for mill k' is also
constructed from all observations (K) that are available in period t via sequential frontiers.
However, in addition to the input constraint(s) and non-increasing returns to scale constraint,
the regulated production function is also subject to constraint(s) on abatement intensity where
the abatement intensity for mill k' is equal or less than that of the best-practice plant (i.e., the
left-hand side of the second constraint).15 These abatement intensity constraints reflect
degrees of abatement intensity for the three abatement technologies used by the regulated
mills.
Having defined the unregulated and regulated production functions, PAC is the ratio of
the maximum good output production of the unregulated (FUR) and regulated (FR)
technologies for plant k' in periods t and t+1.16 That is, in period t pollution abatement costs
(PAC*) is defined as follows:
PAC1 = FU R(x\ t)/ FR(x\ a\ t) (4)
and
m SLiSU zk(akim x Pki)
ib In equation (3), the numerator of ^^ — calculates the amount of pulp production of the best-
ial ^i=i zkPki
practice frontier for mill k' that is subject to abatement process m, while the denominator calculates the total pulp
produced by the best-practice frontier for mill k'.
16 Because we define PAC as the ratio of a mill's good output production when the bad output is unregulated (i.e.,
the mill is producing on its unregulated frontier) to its good output production when the bad output is regulated
(i.e., the mill is producing on its regulated frontier), the foregone good output production associated with technical
inefficiency is excluded.
14
-------
PACt+1 = FUR(xt+1, t+l)/FR(xt+1, at+1, t+1) (4')
is PACt+1 in period t+1.
In Figure 1, FR(x\ a*, t), the regulated frontier, represents the maximum good output
produced using input vector x*, abatement intensity vector a*, and technology in period t, while
FR(xt+1, at+1, t+1) represents the maximum good output produced using input vector xt+1,
abatement intensity vector at+1, and regulated technology in period t+1. The unregulated
FUR(xt, t) and FUR(xt+1, t+1) frontiers represent the maximum good output produced by the
vector of inputs and unregulated technology in periods t and t+1, respectively. Using Figure 1,
the PAC for period t (equation 4) is (aB/aA) and (bD/bC) for period t +1 (equation 4').
To examine how pollution abatement cost changes overtime, we calculate the ratio of
good output production between the regulated and unregulated frontiers in period t and t+1.
We define the change in pollution abatement costs between period t and t+1 as:
rFUR(xt+1, t+l)/FR(xt+1, at+1, t+1)
APACtt+1 =
FUR(xt, t)/FR(xt a1, t)
(5)
.t+i (bD/bC) AnA„f+i
(aB/aA)
Using Figure 1, APACf = - TheAPACf is unity if there is no change in PAC, greater
than unity when there is an increase in PAC, and less than unity when there is a decrease in
PAC.
Using shifts in both the regulated and unregulated frontiers, we decompose changes in
PAC into three components - technological change (TC), changes in inputs (IC), and changes in
pollution abatement intensity (Al). Since using period t or period t+1 as the reference
technologies typically yield different values for the relative importance of the factors associated
15
-------
with changes in PAC, we use the arithmetic means of period t and t+1 as reference technologies
when specifying the mixed-period LP problems:17
APACf+1 = flY [FUR(xt+1,t+l)/FUR(xt+1,t)] \ / [FURCxt.t+lj/FURCxt.t] }
1 UV[FR(xt+1,at+1,t+l)/FR(xt+1, at+1,t)]/ V[FR(xt, a^t+iyFRfx1, at,t)]/J ' J
f|Y [FUR(xt+1,t)/FUR(xt,t)] \ / [FUR(xt+ 1,t+l)/FUR(xt,t+1] \1 }
UV[FR(xt+1, at+i.Q/FRfx1, at+1,t)]/ V[FR(xt+1, a'.t+iyFRCx1, a'.t+l)]/] ' J
(•[¦/ [FURCxt.Q/FURCxt.t)] \ / [FUR(xt+1, t+l)/FUR(xt+1, t+1] \1 }
ILvCFRCx1, at+1, t)/FR(xl, a1, t)]/ V[FR(xt+1, at+1, t+l)/FR(xt+1, a1, t+l)]/J ' J
= (TCur/TCr) x (ICur/ICr) x (AIur/AIr) (6)
where (TCur/TCr) represents the change in PAC associated with TC, (ICur/ICr) is the change in
PAC associated with IC and (AIur/AIr) is the change in PAC associated with Al.
More precisely, (TCur/TCr), measures the change in good output production associated
with technological change of the unregulated technology relative to the change in good output
production associated with technological change of the regulated technology. (ICur/ ICr),
measures the change in good output production of the unregulated frontier relative to the
regulated frontiers associated with changes in the level of inputs. Finally, because Al does not
affect the unregulated frontier, FUR(#) cancel out in the numerator. The Al component then,
reflects movements along the regulated frontier associated with changes in the share of inputs
assigned to good output production and pollution abatement. To illustrate, as shown in Figure
1, equation (6) can be written as
APAC;+1=f[c^)+ ^^1x0.5]
f IL (bC/bG) (aH/aA) J J
x {[M/!B) + « x o.5]
IL (bG/aF) (bl/aH) J J
17 The expressions for the mean of the ratios for TC, IC, and Al when using period t and period t+1 as reference
technologies are not calculated using FR(-) and FUR(*) values for individual mills (i.e., observations) and then
aggregating those ratios. Instead, the FR(-) and FUR(*) values used in our calculations are summations of values
for all observations from periods t and t+1 in the dataset.
16
-------
X
(T(aB/aB)
U(aF/aA)
(bD/bD)
(bC/bl)
(6')
For TC, IC, and Al, a value exceeding unity indicates the component is associated with increasing
PAC between period t and period t+1. A value less than unity signifies the component is
associated with declining PAC. Finally, a value of unity indicates the component is associated
with no change in PAC.
The EPA released a list identifying the 96 pulp and paper mills that were expected to
meet the MACT air and BAT water standards of the CR rule.18 In order to model a
homogeneous production technology, we focus on the kraft (i.e., sulfate) and sulfite pulp and
paper mills on that list. In this section, we describe the production and abatement intensity
data for these mills.
IV. 1 Production Data
Information on mill-level inputs and outputs comes from restricted-use establishment-
level data accessible at Federal Statistical Research Data Centers (RDCs).19 Specifically, we use
production information collected by the quinquennial Census of Manufactures (CM) for 1967-
2007 (U.S. Census Bureau, various years), the Annual Survey of Manufactures (ASM) in non-
census years for 1973-2014 (U.S. Census Bureau, various years), which are linked together
18 The EPA also released a list identifying the 155 pulp and paper mills with sufficiently large air toxic releases to
qualify for the MACT standards of the CR rule. The 96 mills subject to the BAT requirements were a subset of the
155 mills subject to the MACT requirements. Because the process changes required by the BAT requirements were
substantial, we do not include the 59 mills subject only to the MACT standards in our core analysis of the
regulatory cost. See Appendix A, U.S. EPA (1998b) for list.
19 See U.S. Census Bureau (2018).
IV. Data
17
-------
using the Longitudinal Business Database (LBD), as described in Jarmin and Miranda (2002). We
also use data from the Manufacturing Energy Consumption Survey (MECS) that is available
triennially for 1985-1994 and quadrennially for 1998-2010 (U.S. Department of Energy, various
years).20
We measure the good output (y) using the total value of shipments (TVS) in dollars from
CM/ASM, adjusted for inventories and work in progress. This output measure excludes the
value of on-site co-generated electricity sold, which is computed as the product of the quantity
of electricity sold and the implied electricity price for each mill/year observation (i.e.,
expenditures on purchased electricity divided by the quantity of purchased electricity). By
excluding revenue generated by the sale of electricity from the revenue measure of output, the
good output of the mill consists solely of the revenue generated from the sale of pulp and
paper.
We use the following five inputs: labor, materials, capital, electricity and fuels. Labor (L)
is measured by the total production worker hours from the CM/ASM. Materials (M) is
represented by the dollar expenditures on materials, resale, and contract work from CM/ASM.
Capital (K) is represented by mill-specific estimates of real capital stock (equipment and
structures, expressed in dollar terms). We rely on an LBD-linked database that uses the
perpetual inventory method to calculate establishment-specific real capital stock from annual
20 In addition, the 59 MACT mills may not be analyzed separately because the number of mills with complete LBD
data does not meet Census disclosure requirements.
18
-------
data on new capital expenditures from the CM/ASM, the nominal capital gross book value from
CM, and other data (Foster et al., 2014).21,22
Electricity (E) is measured as the quantity of purchased electricity in British thermal units
(BTU) from the CM/ASM. Fuel (F) is measured as consumption of fuels produced off-site in BTU,
which is estimated using data from MECS. While fuel transfers are not a significant share of fuel
for our plants, our analysis includes measures of fuel with and without transfers.23 Because
MECS data are not available annually, we interpolate fuel consumption for the non-MECS years
using annual fuel expenditures from CM/ASM.24 This approach closely follows the value-based
interpolation used by the U.S. Department of Energy (1992). Finally, our measures of labor,
materials, capital, and energy consumption may include resources used for electric power co-
generated by the mill, which may introduce measurement errors into the input data.
IV.2 Pollution Abatement Intensity Data
Information on mill name, location, annual bleaching sequence, total annual pulp
production, and annual data on the percent of chlorine dioxide substitution by pulp line for
every pulp mill subject to the ELGs of the CR from 1997 to 2007 comes from Beca AMEC
(2013a). Because most mills phased in chlorine dioxide (CIO2) over a period of months (see U.S.
EPA, 2000), chlorine dioxide substitution for a pulp line may range from zero to 100% in any
21 We thank John Haltiwanger for providing access to this database.
22 All variables expressed in dollar terms (i.e., output, materials, and capital stock) are converted into real terms—
2009 dollars—by dividing those variables with an appropriate industry-specific price index (Bartelsman and Gray,
1996).
23 MECS defines transfers as quantities 1) delivered from any other establishment in your company, 2) transferred
from other establishments of your company for which payment was not made, 3) purchased centrally within your
company, separate from this establishment, or 4) for which payment was made in-kind.
24 Because some of our mills were not present in the 2010 MECS, we estimated these values by extrapolating their
2007 fuel consumption (in BTUs).
19
-------
given year. The Beca AMEC data (2013b) is also used to identify the year a mill started using OD
and/or ED technologies on its pulp lines.25 However, because the standards for the CR are
based on ECF, many lines at the 96 mills subject to the BAT provisions do not use OD or ED.26
Unlike the measure of CIO2 abatement intensity for a pulp line, aim(m=l), that can
assume a value between zero and one, abatement intensity for OD and/or ED assumes a value
of either zero or one. That is, the abatement intensity of a pulp line, aim (m=2,3), is equal to
one when a pulp line employs OD and/or ED, respectively, on the line, otherwise its intensity is
set to zero. As shown in equation 2, the measures of abatement intensity for ECF, OD and ED at
the mill level are calculated by summing pulp production across lines subject to each
abatement technology, respectively, and dividing that value by total pulp production at each
mill.27
IV.3 Linking Data Sources
Using mill name and location, we match the regulatory intensity measures from the
Beca AMEC data to the list of 96 mills EPA expected to comply with the water technology
standards of the CR. We then use a probabilistic matching method based on Felligi and Sunter
(1969) to link this data to the LBD, which allows us to link all relevant production data from the
CM, the ASM, the MECS, and Haltiwanger capital stock estimates (Foster et al., 2014).28 While a
link to the LBD was established for approximately 70 CR pulp and paper mills, only 50 mills had
25 Pulp lines may use only ECF or a combination of ECF and/or ED/OD.
26 For a more detailed description of the Beca AMEC data, see Morgan et al. (2014).
27 Since abatement intensity is based on pulping operations, our results may be biased because of heterogeneity
among plants in terms of the share of production associated with pulping operations.
28 We use a probabilistic matching method to account for differences in mill names and addresses between our
different data sources.
20
-------
complete CM/ASM production data that were consistently available between 1997 and 2007.29
These 50 mills constitute our balanced panel dataset.
IV.4 Treatment of Non-Pulping Mills
Among our 50 CR mills, several switched from pulping to non-pulping operations during
1997-2007. Because disclosure rules established by the U.S. Census Bureau prohibit removing
the non-pulping mills from the sample, we treat them as unregulated mills (i.e., we set akim=0)
for year(s) that mill k has no pulping operations. Even though these mills are regulated by the
CR, since they are no longer pulping, they are in essence, unregulated. By setting akim=0, we
maintain a balanced panel for both the regulated and unregulated technology.30
V. Linear Programming Problems and Results
We construct four models using different combinations of input and abatement
intensity vectors.31 The first model (Ml) is defined using an input vector x*, in which energy use
in BTU is represented by the sum of purchased electricity and fuels produced off-site without
fuel transfers, and an abatement intensity vector a*, where all three abatement intensity
measures are represented. Model 2 (M2) differs from Ml in that we use the sum of purchased
electricity and fuels produced off-site which includes fuel transfers. Since CR limits are based on
29 We focused on this period for two reasons. First, compliance intensity data for the regulated mills are available
starting from 1997. Second, there were considerably fewer mills with production data continuously available for
1997-2008 and 1997-2009, compared to 1997-2007.
30 Because we are not allowed to drop the mills that stopped pulping from our sample, including them may affect
the results for the pulping mills. The unregulated production frontier, which is constructed from all available
observations, may include a non-pulping mill. In addition, if all akim for pulping mill i are less than unity, then its
regulated production frontier might include an observation with akim = 0, which can either be an unregulated
pulping mill or a non-pulping mill. For more details on this issue, please contact the corresponding author for
Appendix.
31 The estimation process was implemented within the R language and environment (R Development Core Team,
2016) using the open-source linear programming package Rsymphony (Harter et al., 2016).
21
-------
chlorine dioxide substitution, we examine two additional models (M3 and M4) where the
abatement intensity vector, a*, contains only the share of pulp production subject to chlorine
dioxide bleaching. The model M3 uses the same input vector as Ml, while M4 uses the input
vector of M2.
For each model, we solve for the maximum good output for the unregulated frontier in
year t and year t+1, i.e., FUR(xt, t) and FUR(xt+1, t+1), as specified by LP problem (1). We also
solve for the maximum good output for the regulated frontier in year t and year t+1, i.e., FR(xt,
a*, t) and FR(xt+1, at+1, t+1), as specified by LP problem (3) for each model. The decomposition
strategy (6) requires two additional unregulated frontiers and four additional regulated
frontiers. Both the unregulated and regulated frontiers for each period are constructed using
sequential frontiers, which are comprised of observations from 1997 up to year t or year t+1.32
As a result, there are ten two-year pairs from 1997-1998 to 2006-2007 associated with each mill
based on equations (4) and (5).
Due to data confidentially concerns, mill-level results are not released by the Census
Bureau; however, summary statistics of the data, and aggregate results allow us to calculate
indexes of change for pollution abatement costs and the factors associated with changes in PAC
- technological change, input change, and abatement intensity - for each two-year pair from
1997-1998 to 2006-2007 for the fifty mills can be released. While we are unable to present the
results for APAC and its components for individual mills, the decomposition procedure allows to
explore the relative importance of factors associated with changes in pollution abatement costs
for the 50 mills.
32 Sequential frontiers do not allow the possibility of technical regress (i.e., inward shifting production frontiers).
22
-------
Table 1 presents summary statistics of the data for 1997, 2002, and 2007. For the 50
mills in our sample, output increased. This increase in output is associated with fewer workers
and reduced capital stock. These decreases reflect trends observed in the pulp and paper
industry over the same period. Using annual industry-level data from the NBER-CES
Manufacturing Industry database (see http://data.nber.org/nberces/). the number of workers
decreased by 27% while capital stock decreased by 33% from 1997 to 2007. However,
substantial differences are observed between pulp mills (SIC 2611) and paper mills (SIC 2621).
For pulp mills, employment fell by less than 10%, but capital stock declined by 32% between
1997 and 2007, whereas paper mills experienced a decline in employment of 42% while capital
stock decreased by 12% over the same period.
Table 2 presents summary statistics of each measure of regulatory intensity. Regulatory
intensity for chlorine dioxide increases each year until it reaches a peak in 2004, an increase of
44% from 1997, before slightly decreasing, and then remaining constant through 2007. This
increase in intensity reflects the trend in CIO2 substitution adopted by mills over the period. As
Morgan et al (2014) point out, only 28% of mills were using 100% CIO2 substitution in 1998,
67% switched by 2000, and at least 95% switched to 100% CIO2 substitution by 2005. While
there are some year-to-year fluctuations, regulatory intensity for both ED and OD slightly
increased over the sample period. The small increase in pulp production subject to ED and OD,
roughly 5% and 16%, respectively, over the 1997-2007 is not surprising given only nine mills
installed OD during 1998-2008 while four mills installed ED over the same period (Morgan et al,
2014).
23
-------
The APAC and its components for each model are presented in Tables 3, 4, and 5.33
Table 3 presents the results for model Ml. The APAC ranges from a minimum of 0.9725 for
2000-2001 to a maximum of 1.0527 for 2002-2003, with an average increase of 0.26% over the
1997-2007 period. Across all two-year periods in our sample, the increase in APAC in 2002-
2003 is associated with the largest increases in PAC due to TC and IC, 3.55% and 1.96%,
respectively, and the largest decrease in PAC is associated with Al of 0.28%.
For models M3 and M4, whose results are presented in Tables 4 and 5, overall changes
in PAC, TC, IC, and Al generally follow similar patterns and magnitudes over the entire period.
The maximum APAC for both models occur during the 1999-2000 period, while the minimum
APAC for both models occur in the next period, 2000-2001. Unlike model Ml, where average
APAC increased over the 1997-2007 period, average APAC decreased by 0.33% and 0.47% over
the entire period for models M3 and M4, respectively.
For all models, TC is the only component that is associated with reducing average PAC
over the entire 1997-2007 period. More specifically, for model Ml, where the abatement
intensity vector includes all three abatement measures, TC is associated with a 1.59% decrease
in PAC. For models M3 and M4, where the abatement intensity vector only includes ECF, TC is
associated with a 0.93% and 1.06% decrease in PAC for models M3 and M4, respectively. In
general, for models M3 and M4, the fluctuations over time with PAC are associated with similar
fluctuations in TC. That is, in years when PAC increased, TC is associated with increased PAC.
33 Unfortunately, we did not catch a mistake in our code and disclosed negative outputs for model M2, preventing
us from evaluating PAC and its component. Census Bureau policy prevents us from accessing our data at this time
to correct the code and re-run the model.
24
-------
The relationship between PAC and TC reflects relative shifts in the unregulated and
regulated frontiers. The decline in PAC associated with TC may reflect the reallocation of R&D
resources from the unregulated technology to regulated technology (e.g., ECF bleaching
process to reduce production of dioxin). Because the CR required mills to substitute ECF for
chlorine bleaching, mills were able to adopt the new technology while simultaneously
expanding production of pulp and paper. The reallocation of R&D resources yields slower TC
for the unregulated technology and faster TC for the regulated technology. This results in the
regulated frontier shifting outward more rapidly than the unregulated frontier which reduces
the foregone output associated with reducing dioxin in period t+1 relative to period t, leading
to values of APAC less than one.
Unlike TC, both IC and Al are associated with increases in PAC for all models. The
increase in PAC is the highest for Ml where IC and Al are associated with an average annual
increase in PAC of 1.01% and 0.90%, respectively. Except for 2004-2005, when Al is less than
unity for M3 and M4, and thus associated with decreases in PAC, Al is associated with increases
in PAC. While the year-to-year fluctuations in IC are similar between M3 and M4, the
relationship between PAC and IC is not as clear. In both models, IC and APAC move in opposite
directions between 2000-2001 and 2003-2004. That is, an increase in IC is associated with a
decrease in PAC and vice versa. Overall, for models M3 and M4, IC is associated with an
average annual increase in PAC of 0.14% and 0.10%, respectively, while Al is associated with an
average annual increase in PAC of 0.47% and 0.50%, respectively.
Given that employment and capital stock were decreasing in the pulp and paper
industry over our sample period, we were surprised to find inputs are associated with increases
25
-------
in PAC. This result made us examine more closely, industry trends in our other two inputs -
material and energy. For the industry, the cost of materials remained stable for paper mills
from 1997 to 2002, only increasing roughly 2% over the period. However, material costs
increased nearly 40% for pulp mills. On the other hand, both pulp and paper mills experienced
large increases in energy costs over the time period, roughly 50% and 29%, respectively.34 The
large increase in energy costs for pulp mills may reflect increases in CIO2 substitution. Because
the production of CIO2 is unstable and explosive, it is generally made on site and the electricity
requirements to make CIO2 are sizeable.35
We are interested in the behavior of PAC over time. Recall, PAC is the ratio of the
maximum good output production of the unregulated and regulated technologies in period t as
given by equation 4.36 Table 6 presents PAC in period t for each model. For model Ml, PAC
starts at 9.68% in 1997, decreases until 1999, increases in 2000 before decreasing to 5.44% in
2002. After 2002, PAC increases again. On average, PAC for model Ml was roughly 9.4%,
which seems high relative to historical abatement costs in the pulp and paper industry. The
results are strikingly different for models M3 and M4. PAC starts much lower for M3 and M4,
at 3.49% and 4.97%, respectively. And, except for 2004, when PAC increases slightly, pollution
abatement costs decrease over the time period in both models. On average from 1997 to 2006,
PAC is 1.69% and 1.96% for models M3 and M4, respectively which is more in line with
historical abatement costs in the pulp and paper industry.
34 See http://data.nber.org/nberces/
35 See http://www.paperenvironment.org/PDF/chcompounds/energy/CC E Electricitv.pdf.
36 Except for 1997, the values for PAC(t+l) given by equation 4' are the same values as PAC(t), just shifted one
year.
26
-------
Intuitively, smaller estimates of PAC for M3 and M4 are expected. Recall, unlike model
Ml which has three abatement intensity constraints for each abatement technology, models
M3 and M4 only have one abatement intensity constraint for CIO2 substitution. Additional
constraints limit the good output a mill produces using the regulated technology, and because
the output produced by the unregulated technology remains the same, lower output for the
regulated technology when subject to additional constraints means overall, PAC will be higher,
as exhibited by PAC for Ml.
To put our pollution abatement costs in perspective with other estimates of the cost of
implementing the Cluster Rule, we examined some other estimates of pollution abatement
costs. Table 7 shows industry estimates of pollution abatement capital expenditures from
1997-2002 from the National Council for Air and Stream Improvement's (NCASI) survey of pulp
and paper firms.37,38 According to results published by NCASI, the percentage of total capital
expenditures for environmental protection spent on water pollution abatement ranges from
32% to 56% over the 1998 -2002 period, which amounts to approximately $1.5 billion dollars.
NCASI also reports the percentage of total capital expenditures spent on environmental
protection each year (see Table 7). Multiplying that percentage by the percent spent on water
pollution yields an average of roughly 6% of total capital expenditures are spent on water
pollution abatement.
37 NCASI is the main trade association of the forest products industry and has been conducting a pollution
abatement capital cost survey since 1970. NCASI receives survey responses from a subset of companies and
extrapolates to the entire U.S. pulp and paper industry assuming the companies that did not report expenditure
data, spend at a similar rate. NCASI stopped conducting this survey in 2002.
38 See Morgan, Pasurka, and Shadbegian (2014) for more information on the CR and NCASI Survey.
27
-------
Looking at historical pollution abatement operating costs (PAOC) and capital
expenditures (PACI) for water from the Pollution Abatement Costs and Expenditures Survey
(PACE) conducted by the U.S. Census Bureau, the percent of PAOC for water compared to total
value of shipments and PACI for water compared to all new capital expenditures is small.39 As
shown in Table 8, for various years from 1992 to 2005, the percent of water PAOC to value of
shipments ranges from 0.19% to 0.62% while the percent of water PACI to new capital
expenditures ranges from 2.7% to 4.7%. While water pollution abatement costs are a small
percentage of other costs for pulp and paper mills, those costs are a larger percent of total
pollution abatement costs. The percent of water PACI compared to total PACI ranges from 21%
to almost 40%.
The fluctuations over time and the magnitude of pollution abatement costs from other
sources are not unlike what we find, especially for models M3 and M4. The PAC captured by
our models are the opportunity costs of pollution abatement activities, which is a more
inclusive measure of the cost of pollution abatement than the accounting costs captured by the
NCASI and PACE surveys. More specifically, opportunity costs measure the forgone
opportunity of using inputs for pollution abatement as opposed to producing the good output,
thus we may also capture some additional costs of pollution abatement activities (e.g. changes
in production processes) that are not captured by these surveys. Moreover, as is often the case
with DEA models, the cost estimates from our models may exceed those found from industry
and government surveys because of outliers in our data.
39 The PACE survey was conducted annually from 1973-1994 (except 1987), and then in 1999 and 2005.
28
-------
VI. Conclusions
Two approaches have been developed in the literature to calculate pollution abatement
costs (PAC): the joint production model, which requires data on bad outputs and the assigned
input model, which requires information on the inputs assigned to pollution abatement. The
main contribution of this paper is we are the first to demonstrate how to use information on
pollution abatement intensity to measure PAC when there are no data on bad output
production, or quantities of inputs assigned to pollution abatement. Using the U.S. EPA's
Cluster Rule as a case study, we employ a unique dataset that identifies the year and extent of
adoption of different pollution abatement technologies by mills to calculate the opportunity
cost of pollution abatement. This data allows us to include the degrees of pollution abatement
intensity, captured by variation in the levels of adoption of extended delignification (ED),
oxygen delignification (OD) and elemental chlorine free (ECF) or chlorine dioxide substitution,
into our specification of the regulated technology. Specifically, we include three intensity
constraints when modeling the regulated mills: 1) the share of pulp production subject to ECF
or chlorine dioxide substitution, 2) the share of pulp production subject to OD, and 3) the share
of pulp production subject to ED. These shares range from zero to one and change over our
time period as mills increase the share of pulp production subject to these abatement
technologies.
Using these constraints, we calculate the maximum good output production when the
bad output is regulated and when it is unregulated, where the unregulated technology excludes
the constraints on abatement intensity. The difference in good output production between the
regulated and unregulated technologies is the PAC of reducing dioxins, furans, and chloroform
29
-------
loadings to water. We then investigate the relative importance of changes in the quantities of
inputs used, technological change, and changes in pollution abatement intensity in explaining
changes in PAC.
While there were fluctuations in pollution abatement cost over the 1997-2007 period
for our 50 pulp and paper mills, on average pollution abatement costs appear to decrease,
especially for our two models that focus on the abatement intensity changes in pulp production
subject to ECF. Most importantly, our findings suggest that technological change moderated
the increase in pollution abatement costs associated with reducing releases of effluents to
comply with the CR. On the other hand, input change and abatement intensity are associated
with increases in PAC. These results are similar across all models.
30
-------
References
Aiken, Deborah Vaughn, Rolf Fare, Shawna Grosskopf, and Carl Pasurka (2009), "Pollution
Abatement and Productivity Growth: Evidence from Germany, Japan, the Netherlands, and the
United States," Environmental and Resource Economics, 44, No. 1 (September), 11-28.
https://doi.org/10.1016/S0928-7655(03)00042-3
Bartlesman, Eric J. and Wayne Gray (1996), "The NBER Manufacturing Productivity Database,"
National Bureau of Economic Research, Technical Working Paper 205, October.
(http://www.vrdc.cornell.edu/info747/2007/Readings/bartelsman-gray-1996.pdf)
Beca AMEC. (2013a), "Annual US pulp line MTPY" spreadsheet, Tony Johnson, ed.
Beca AMEC. (2013b), "Pulp mill data - delig study" spreadsheet, Tony Johnson, ed.
Elrod, Aaron and Arun Malik (2017). "The effect of environmental regulation on plant-level
product mix: A study of EPA's Cluster Rule," Journal of Environmental Economics and
Management, 83, 164-184. https://doi.Org/10.1016/i.ieem.2017.03.002
Fare, Rolf, Shawna Grosskopf, and Carl Pasurka (2007), "Pollution Abatement Activities and
Traditional Productivity," Ecological Economics, 62, No. 3-4 (15 May), 673-682.
https://doi.Org/10.1016/i.ecolecon.2006.08.014
Fare, Rolf and Shawna Grosskopf, and Carl Pasurka (2016), "Technical Change and Pollution
Abatement Costs," European Journal of Operational Research, 248, No. 2 (16 January), 715-
724. https://doi.Org/10.1016/i.eior.2015.07.040
Felligi, Ivan P. and Alan B. Sunter (1969), "A Theory for Record Linkage," Journal of the American
Statistical Association, 64. No. 328 (December), 1183-1210.
https://doi.org/10.1080/01621459.1969.10501Q49
Foster, L., Grim, C., & Haltiwanger, J. (2014), "Reallocation in the Great Recession: Cleansing or
Not?" (No. w20427). National Bureau of Economic Research.
http://www.nber.org/papers/w20427
Gollop, F.M. and M.J. Roberts (1983), "Environmental Regulations and Productivity Growth: The
Case of Fossil-fueled Electric Power Generation," Journal of Political Economy, 91, No. 4
(August), 654-74. https://doi.org/10.1086/261170
Gray, Wayne B, and Ronald J. Shadbegian (2003), "Plant Vintage, Technology, and
Environmental Regulation," Journal of Environmental Economics and Management, 46, No. 3
(November), 384-402. https://doi.org/10.1016/S0095-0696(03)00031-7
31
-------
Harter, R., Hornik, K., Theussl, S., Szymanski, C. (2016), An R interface to the SYMPHONY solver
for mixed-integer linear programs, https://cran.r-
proiect.org/web/packages/Rsymphonv/index.html.
Jarmin, Ron S., and Javier Miranda (2002), "The Longitudinal Business Database," Center for
Economic Studies Discussion Paper CES-WP-02-17.
https://www.census.gov/ces/dataproducts/datasets/lbd.html
Maynard, Leigh J. and James S. Shortle (2001), "Determinants of Cleaner Technology
Investments in the U.S. Bleached Kraft Pulp Industry," Land Economics, 71, No. 4 (November),
561-576. https://doi.org/10.2307/3146941
Morgan, Cynthia, Carl Pasurka, and Ron Shadbegian (2014), "Ex Ante and Ex Post Estimates of
the Cluster Rule and MACT II Rule," Journal of Benefit-Cost Analysis, 5, No. 2 (June), 195-214.
https://doi.org/10.1515/ibca-2014-0026
National Council for Air and Stream Improvement, Inc. (various years), "A survey of pulp and
paper industry environmental protection expenditures," Research Triangle Park: National
Council for Air and Stream Improvement, Inc.
Norberg-Bohm, Vicki and M. Rossi (1998), "The Power of Incrementalism: Environmental
Regulation and Technological Change in Pulp and Paper Bleaching in the US," Technology
Analysis & Strategic Management, 10, No. 2 (June), 225-245.
https://doi.org/10.1080/095373298Q8524313
Popp, David, T. Hafner, and N. Johnstone (2011), "Environmental Policy vs. Public Pressure:
Innovation and Diffusion of Alternative Bleaching Technologies in the Pulp Industry," Research
Policy, 40, No. 9 (November), 1253-1268. https://doi.Org/10.1016/i.respol.2011.05.018
R Development Core Team (2016), R: A language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/.
Snyder, Lori D., Nolan H. Miller, and Robert N. Stavins (2003), "The effects of environmental
regulation on technology diffusion: the case of chlorine manufacturing," American Economic
Review, 93, No. 2 (May), 431-435. https://doi.org/10.1257/00028280332194747Q
U.S. Census Bureau (various years), Annual Survey of Manufactures.
U.S. Census Bureau (various years), Census of Manufacturers.
U.S. Census Bureau (2018), "Federal Statistical Research Data Centers,"
https://www.census.gov/fsrdc
32
-------
U.S. Bureau of the Census, (various issues), Pollution abatement costs and expenditures.
Current Industrial Reports (MA200). Washington, DC: U.S. Government Printing Office.
U.S. Congress, Office of Technology Assessment (1989), "Technologies for Reducing Dioxin in
the Manufacture of Bleached Wood Pulp", OTA-BP-O-54, Chapter 4, May.
https://www.princeton.edu/~ota/diskl/1989/8931/8931.PDF
U.S. Department of Energy, Energy Information Administration (1992), "Derived Annual
Estimates of Manufacturing Energy Consumption 1974-1988," Energy Consumption Series.
August 1992. DOE/EIA-0555(92)/3
U.S. Department of Energy, Energy Information Administration (various years), Manufacturing
Energy Consumption Survey.
U.S. Environmental Protection Agency (1997), "Economic analysis for the national emission
standards for hazardous air pollutants for source category: pulp and paper production; effluent
limitations guidelines, pretreatment standards, and new source performance standards: pulp,
paper, and paperboard category-phase 1." Research Triangle Park, NC: Office of Air and
Radiation, Office of Air Quality Planning and Standards, Innovative Strategies and Economics
Group and Washington, DC: Office of Water, Office of Science and Technology, Engineering and
Analysis Division, October.
https://www.epa.gov/sites/production/files/2015-10/documents/pulp-paper economic-
analysis-cluster-rule 1997.pdf
U.S. Environmental Protection Agency (1998a), National emission standards for hazardous air
pollutants for source category: pulp and paper production; effluent limitations guidelines,
pretreatment standards, and new source performance standards: pulp, paper, and paperboard
category. Federal Register, 63[72), 18504-18751.
U.S. Environmental Protection Agency (1998b), Pulp and Paper NESHAP: A Plain English
Description, EPA-456/R-98-008, November.
U.S. EPA (2000), "Permit Guidance Document: Pulp, Paper and Paperboard Manufacturing
Point Source Category", EPA-821-B-00-0003, May.
https://www3.epa.gov/npdes/pubs/pretreatment pulp.pdf
U.S. Environmental Protection Agency (2018a), "Learn about Dioxin,"
https://www.epa.gov/dioxin/learn-about-dioxin
U.S. Environmental Protection Agency (2018b), "Pulp and Paper Production (MACT I & III):
National Emissions Standards for Hazardous Air Pollutants (NESHAP) for Source Categories"
https://www.epa.gov/stationarv-sources-air-pollution/pulp-and-paper-production-mact-i-iii-
national-emissions-standards
33
-------
U.S. Environmental Protection Agency (2018c), "Pulp, Paper and Paperboard Effluent
Guidelines" http://water.epa.gov/scitech/wastetech/guide/pulppaper/index.cfm
34
-------
X
Figure 1. Unregulated and Regulated Frontiers in Periods t and t+1
35
-------
Table 1. Summary Statistics for Output and Inputs
Units Mean Std. Deviation
50 Pulp and Paper Mills, 1997
Output
in millions, 2009$
435,100
196,700
Labor
Workers
1553
750
Capital
in millions, 2009$
517,300
283,900
Materials
in millions, 2009$
185,600
79,180
Qty of Energy
Btu (in billions)
6,778,000
4,070,000
Qty of Energy plus Fuel Transfers
Btu (in billions)
7,704,000
4,192,000
50 Pulp and Paper Mills, 2002
Output
in millions, 2009$
450,400
214,500
Labor
Workers
1287
558
Capital
in millions, 2009$
421,500
219,500
Materials
in millions, 2009$
171,400
77,350
Qty of Energy
Btu (in billions)
7,547,000
5,351,000
Qty of Energy plus Fuel Transfers
Btu (in billions)
7,713,000
5,526,000
50 Pulp and Paper Mills, 2007
Output
in millions, 2009$
426,500
231,900
Labor
Workers
1160
544
Capital
in millions, 2009$
344,700
189,200
Materials
in millions, 2009$
167,400
78,020
Qty of Energy
Btu (in billions)
6,978,000
4,557,000
Qty of Energy plus Fuel Transfers
Btu (in billions)
7,050,000
4,530,000
36
-------
Table 2. Annual Means for Regulatory Intensity (standard deviation)
Year
CI02
ED
OD
1997
0.5161 (0.378)
0.2600 (0.443)
0.3139 (0.443)
1998
0.5835 (0.390)
0.2600 (0.443)
0.3179 (0.445)
1999
0.7005 (0.377)
0.2600 (0.443)
0.3301 (0.444)
2000
0.7631 (0.367)
0.2593 (0.442)
0.3380 (0.452)
2001
0.8081 (0.342)
0.2593 (0.442)
0.3380 (0.452)
2002
0.9050 (0.280)
0.2593 (0.442)
0.3398 (0.453)
2003
0.9050 (0.280)
0.2737 (0.445)
0.3377 (0.452)
2004
0.9176 (0.274)
0.2738 (0.445)
0.3683 (0.467)
2005
0.8800 (0.328)
0.2741 (0.446)
0.3665 (0.464)
2006
0.8800 (0.328)
0.2745 (0.446)
0.3658 (0.464)
2007
0.8800 (0.328)
0.2745 (0.446)
0.3754 (0.463)
Table 3. Decomposition of APAC for two-year pairs for Ml
(BOLD = maximum value and ITALICS = minimum value)
Two-year pairs
APAC
TC
IC
Al
1997-1998
1.0000
0.9762
1.0116
1.0127
1998-1999
0.9895
0.9492
1.0009
1.0428
1999-2000
1.0200
0.9848
1.0153
1.0204
2000-2001
0.9725
0.9525
1.0176
1.0036
2001-2002
0.9794
0.9654
1.0075
1.0070
2002-2003
1.0527
1.0355
1.0196
0.9972
2003-2004
1.0091
0.9876
1.0163
1.0054
2004-2005
0.9851
0.9835
1.0023
0.9994
2005-2006
0.9867
0.9842
1.0026
0.9999
2006-2007
1.0310
1.0219
1.0075
1.0014
Arithmetic Mean
1.0026
0.9841
1.0101
1.0090
Note: subtracting unity from values in this table and multiplying by 100 yield percentage changes.
Note: For Tables 3, 4, and 5, the arithmetic means are calculated by taking unweighted means of the 10
2-year pairs for APAC and its three components. However, using unweighted means to calculate APAC,
TC, IC, and Al for each 2-year pair results in a slight discrepancy when compared to calculating APAC and
its three components using the sum of maximum good output values for all mills for the 10 2-year pairs.
For model Ml, the largest discrepancy is for the Al component, where the mean value of Al found using
values for all mills across the 10 2-year pairs is 1.0084.
37
-------
Table 4. Decomposition of APAC for two-year pairs for M3
(BOLD = maximum value and ITALICS = minimum value)
Two-year pairs
APAC
TC
IC
Al
1997-1998
0.9966
0.9803
1.0059
1.0107
1998-1999
0.9891
0.9694
0.9980
1.0226
1999-2000
1.0081
0.9961
1.0034
1.0085
2000-2001
0.9855
0.9792
1.0038
1.0026
2001-2002
0.9991
0.9979
0.9986
1.0026
2002-2003
0.9970
0.9922
1.0048
1.0000
2003-2004
1.0008
1.0000
1.0005
1.0003
2004-2005
0.9960
0.9979
0.9984
0.9997
2005-2006
0.9960
0.9963
0.9996
1.0000
2006-2007
0.9985
0.9981
1.0004
1.0000
Arithmetic Mean
0.9967
0.9907
1.0014
1.0047
Note: subtracting unity from values in this table and multiplying by 100 yield percentage changes.
Table 5. Decomposition of APAC for two-year pairs for M4
(BOLD = maximum value and ITALICS =
minimum value)
Two-year pairs
APAC
TC
IC
Al
1997-1998
0.9835
0.9705
1.0006
1.0129
1998-1999
0.9936
0.9741
0.9979
1.0223
1999-2000
1.0068
0.9911
1.0052
1.0105
2000-2001
0.9825
0.9763
1.0035
1.0029
2001-2002
0.9987
0.9970
0.9998
1.0020
2002-2003
0.9963
0.9919
1.0045
1.0000
2003-2004
1.0005
0.9999
1.0005
1.0001
2004-2005
0.9967
0.9988
0.9981
0.9998
2005-2006
0.9954
0.9961
0.9993
1.0000
2006-2007
0.9987
0.9981
1.0006
1.0000
Arithmetic Mean
0.9953
0.9894
1.0010
1.0050
Note: subtracting unity from values in this table and multiplying by 100 yield percentage changes.
38
-------
Table 6. Pollution Abatement Costs in period t for each model
(BOLD = maximum value and ITALICS = minimum value)
Model Ml
Model M3
Model M4
Year
PAC'
PAC'
PAC
1997
1.0968
1.0349
1.0497
1998
1.0969
1.0314
1.0324
1999
1.0853
1.0202
1.0257
2000
1.1070
1.0284
1.0326
2001
1.0766
1.0135
1.0146
2002
1.0544
1.0125
1.0133
2003
1.1100
1.0095
1.0095
2004
1.1201
1.0103
1.0100
2005
1.1034
1.0063
1.0067
2006
1.0888
1.0023
1.0021
Arithmetic Mean
1.0939
1.0170
1.0197
Note: subtracting unity from values in this table and multiplying by 100 yield percentage changes.
Table 7. Environmental Protection Expenditures for Pulp and Paper Industry (NCASI)
Year
Total Capital Expenditures
for Environmental
Protection (EP) (millions of
dollars)
% of Capital
Expenditures
Assigned to Water
Protection
Total Capital
Expenditures for
Water EP (millions
of dollars)
% of Total Capital
Expenditures
Assigned to EP (%
water protection)
1997
560
56
314
1998
630
50
315
13 (6.5)
1999
790
42
332
17 (7.1)
2000
1220
34
415
23 (7.8)
2001
612
32
196
12 (3.8)
2002
363
56
203
9(5.0)
Source: National Council for Air and Stream Improvement, Inc. (various years)
39
-------
Table 8. Pollution Abatement Costs and Expenditure Survey - Water
Year
Water PAOC/Value of Shipments
(in percent)
Water PACI/New Capital
Expenditures (in percent)
Water PACI/
Total PACI (in
percent)
1992
0.62
4.7
37.2
1994
0.58
2.7
30.8
1999
0.19
3.9
39.4
2005
0.35
2.7
21.1
PAOC: Pollution Abatement Operating Costs
PACI: Pollution Abatement Capital Expenditures
Sources: U.S. Bureau of the Census, Pollution Abatement Costs and Expenditures Survey (various years),
Annual Survey of Manufactures (various years)
40
------- |