environmental regulations and technological change in the
TRANSCRIPT
DRAFT
Environmental Regulations and Technological Change
in the Offshore Oil and Gas Industry:
Rethinking the Porter Hypothesis
Shunsuke Managi1, James J. Opaluch1, Di Jin2 and Thomas A. Grigalunas1
1Department of Environmental and Natural Resource EconomicsUniversity of Rhode Island
Kingston, Rhode Island 02881
2Marine Policy CenterWoods Hole Oceanographic Institution
Woods Hole, Massachusetts 02543
This research was funded by the United States Environmental Protection Agency STAR grantprogram (Grant Number Grant Number R826610-01) and the Rhode Island AgriculturalExperiment Station (AES # XXXX), and is Woods Hole Contribution Number 10704. Theresults and conclusions of this paper do not necessary represent the views of the fundingagencies.
April 2002
DRAFT
1
Environmental Regulations and Technological Change in the Offshore Oil and Gas Industry: Rethinking the Porter Hypothesis
Shunsuke Managi, James J. Opaluch, Di Jin and Thomas A. Grigalunas
ABSTRACT: Technological progress can play a key role in maintaining a high standard of living
while addressing environmental problems. Well-designed environmental regulations encourage
technological innovation, while poorly designed regulations can inhibit technological progress.
Indeed, the well-known Porter hypothesis suggests that environmental regulations could
conceivably spur technological growth, leading to a win-win solution with a long-term increase
in productivity of market outputs, while also providing greater environmental protection. We
recast the Porter hypothesis to consider productive efficiency of multi-product production
function, which includes both market and non-market outputs. We apply Data Envelopment
Analysis (DEA) techniques to a unique field-level data set to measure changes in total factor
productivity (TFP) in the oil and gas production in the Gulf of Mexico outer continental shelf.
We test the direction of causality between technological innovation and environmental
regulation, and find support for the recast version of the Porter hypothesis.
JEL codes: O32, O38, L71
Keywords: Total Factor Productivity; Technological change; Porter hypothesis; Offshore oil
and gas industry.
DRAFT
2
I. Introduction
Technological progress can play a key role in the resolving environmental problems
while maintaining productivity. However the extent of its contribution depends on how well
environmental policies are designed and implemented. Successful environmental policies can
contribute to technological innovation and diffusion (Kneese and Schultze, 1978; Jaffe, Newell
and Stavins, 2002), while poor policy designs can inhibit innovation.
Conventional wisdom suggests that environmental regulations impose significant costs
on industry, adversely impacting productivity (e.g., Palmer, Oates and Portney 1995). Recently,
however, researchers have challenged this conventional view with an alternative hypothesis that
environmental regulations can encourage innovation, potentially resulting in increased
productivity and ultimately higher profits. This is the well-known Porter hypothesis (Porter,
1991; Porter and van der Linde, 1995)1. Some studies have confirmed that, in principle, market
failure associated with technological innovation (e.g., Romer, 1990) can imply circumstances
under which environmental regulations can benefit industry (Ulph 1996; Simpson and Bradford
1996; Bovenberg and Smulders 1996; Xepapadeas and De Zeeuw 1999; Mohr, 2002).2 Where
these circumstances hold, well designed environmental regulations can lead to solutions whereby
efficiency is increased in both market and nonmarket outputs. However, there is a broader class
of problems where environmental regulations can potentially result in increased social efficiency
1 Jaffe, Newell and Stavins (2000) and Kemp (1997) provide thorough surveys of the literature relating policy,
technological change, and the environment.2 Simpson and Bradford (1996) and Ulph (1996) rely on a two-country, game-theoretic framework where each
country hosts one producer. Gains come from exploiting market power, so the models apply only to a subset ofenvironmental industries. Bovenberg and Smulders (1996) explore the role of environmental policy in anendogenous growth model and characterize conditions under which more stringent policies are likely to leadhigher growth. Mohr (2002) derives the results using a less restrictive model starting with the observation thatnew productive capital is often less polluting than prior generations of capital as in Xepapadeas and De Zeeuw(1999). Using a general equilibrium model with a large number of agents, the model shows that environmentalpolicy can simultaneously increase productivity and welfare. These gains come even without accounting for thevalue of a cleaner environment.
DRAFT
3
in the joint production of market and non-market products, so that if there is a decrease in in
productive efficiency of market outputs, it is made up for by increased productivity in non-
market, environmental outputs.
This paper explores these issues using a unique micro-level data set for offshore oil and
gas production in the Gulf of Mexico. We recast the Porter hypothesis to calculate productivity
change in a joint production model, with a vector of market and non-market outputs. We also
test for the causal directions between technological change and environmental regulations. We
find no support for the increased productivity of market outputs as stated by the standard Porter
hypothesis, but we find support for the Porter hypothesis recast in terms of joint production of
/market goods and environmental commodities.
An important challenge faced in empirical tests of the Porter hypothesis is identifying the
direction of causality between technological innovation and environmental regulations. New,
tougher environmental regulations might spur research and development efforts leading to
innovation. But at the same time, technical innovations, especially those in pollution control
technologies, may lead federal agencies to develop tougher environmental regulations that
capitalize on these new technologies (e.g., Meyer, 1993). For example, U.S. EPA’s technology-
based standards are based on concepts like Best Conventional Technology (BCT) or Best
Available Technology (BAT). The stringency of new regulations will reflect the technologies
known at the time of implementation. Because causality between regulations and innovation
may go in either (or both) directions, it is important to identify the direction of causality between
regulations and advances in environmental technologies.
Literature Review
Historically, technological change was modeled by simply observing that productivity
changed over time (e.g., Solow, 1957). The use of time as the sole “explanatory” variable for
DRAFT
4
technological change has been recognized as merely a “confession of ignorance” (e.g., Arrow,
1962), in that it doesn’t contribute to an understanding the innovation process. Clearly, new
technologies do not occur exogenously over time, but are, in part, a result of conscious and
costly efforts in research and development (R&D).
Many empirical studies have used patents and R&D expenditures as proxies of
technological innovation. While these proxies provide some insights into technological change,
they also have important limitations. R&D expenditures indicate the effort expended in the
search for new technology, and so provides a measure of inputs to innovation, but R&D
expenditures are not necessarily a good proxy for innovation. Many firms conduct R&D
fruitlessly for years, and some innovative firms create major breakthroughs with little officially
recorded R&D. The measurement issues are especially troublesome for analyses that capitalize
on long time series of data. New discoveries based on past knowledge may lead to positive
feedbacks that imply an accelerating productivity of R&D (e.g., Romer, 1990). Or there could
be eventual limits to productivity, which imply a diminishing rate of productivity (e.g., Griliches,
1994). For these or other reasons, the relationship between R&D expenditures and innovation
may vary systematically over time so that R&D expenditures could be a misleading measure of
innovation rates.
Similarly, use of patents as a proxy for technological change also has important
drawbacks. Many patents never see commercial application, many innovations are not patented,
and some are subdivided into multiple patents, each covering one or more aspects of the
innovation. Changes in patent policies over time may again make patent counts a misleading
measure of innovation, particularly over long time periods. In response to these issues,
DRAFT
5
refinements of patent counts use citations as a weight to the patent (see Hall, Jaffe and
Trajtenberg, 2001).
Moss (1993) and Cuddington and Moss (2001) provide further refinement of measures
of technological change by counting the number of innovations actually implemented each year
by companies in an industry, as reported in trade journals. This represents a significant advance,
but a simple innovation count treats all innovations as having an equivalent importance. In fact,
various new technologies have vastly different levels of significance. A small number of major
breakthroughs may have larger productivity effects than a larger number of incremental
innovations.
We extend innovation counts as a measure of technological change into an importance-
weighted index using the results of an industry survey regarding the importance of specific
technological innovations, both in the short term and the long term. Our technological index is
described in more detail further below.
Empirical investigations of the relationship between the stringency of environmental
regulation and development of new technologies has increased over the last few decades.
Limited evidence that suggests that patent counts and research & development (R&D)
expenditures increase with stringency of environmental regulation. For instance, Lanjouw and
Mody (1996) conducted a study of Germany, Japan, and the U.S., which finds a positive
relationship between environmental compliance cost (a proxy for environmental regulation
stringency) and patenting of new environmental technologies3. In addition, Jaffe and Palmer
(1997) used US data to investigate the relationship between environmental compliance
3 Lanjouw and Mody (1996) analyze the impact of increase in environmental compliance cost of the patenting of
environmental technologies using international data. Lanjouw and Mody provide support the “weak” version ofthe hypothesis which Jaffe and Palmer (1997) defined the weak version of the Porter hypothesis is thatenvironmental regulation stimulates only certain kind of innovations, which are environmental innovations.
DRAFT
6
expenditures and R&D expenditures. Here results show no significant relationship between
environmental compliance cost and patents. However, they found a significant relationship
between compliance costs and R&D expenditure (Jaffe and Palmer, 1997)4.
Jaffe et al. (1995) review empirical studies on this subject, and show that environmental
regulations reduce productivity in the industries. For example, Jorgenson and Wilcoxen (1990)
develop a model to analyze the impact of environmental regulations in the U.S. and conclude
that a long-run cost of environmental regulation is a reduction of 1.91% in the level of the U.S.
gross national product. The recent study by Berman and Bui (2001) of the US oil refiners
indicates suggests that environmental regulation is productivity enhancing. These studies,
however, have been criticized for not considering the full range of impacts of environmental
regulations, including possible positive external impacts on other producers (e.g., Barbera and
McConnell, 1990; Repetto, 1996).
Substantial efforts have been made to regulate pollution in most industrialized countries,
and the stringency of pollution regulations have continued to increase worldwide. And many
economists subscribe to the idea that painful consequences of environmental regulations cannot
easily be avoided, since environmental regulation involves additional cost to industry (Jorgenson
and Wilcoxen, 1990; Portney 1994; Palmer, Oates and Portney 1995; and Jaffe et. al. 1995).
What appears to be the important issue is how to foster compatibility between environmental
regulation and technological innovation (& productivity) in order to control the adverse impact
on industry to the extent feasible. Thus, environmental regulations that encourage technological
change can contribute to a least-cost resolution of environmental problems, while regulations
4 Jaffe and Palmer (1997) consider two measures of innovative activity: total private expenditures on R&D and
the number of successful patent applications by domestic firms in an industry. Both of them includesenvironmental and non-environmental value.
DRAFT
7
that inhibit technological change may lead to substantially higher costs. Success in this regard
will clearly depend on how environmental regulations are designed and implemented.
The controversy regarding the Porter Hypothesis is generated, in part, by a lack of
convincing empirical evidence. It is therefore necessary to examine the economic and social
impacts of specific regulations. In this paper we use a unique micro-level data on offshore oil
and gas production in the Gulf of Mexico to measure technological change to test the Porter
hypothesis. We estimate the relationship between environmental regulations and productivity in
the offshore oil and gas industry in the Gulf of Mexico, considering the impacts on market
outputs and on joint production of market and environmental outputs.
We apply a mathematical programming technique called Data Envelopment Analysis
(DEA) (see, for example, Charnes et al, 1978, Färe et al, 1985) to compute the total change in
productivity over time. We decompose these components to provide a better understanding of
the relative importance of various productivity effects over the study period, so as to identify the
relative impact of innovation, or the creation of new technologies, diffusion of existing
technologies and non-structural productivity effects, such as so-called learning-by-doing.
We then apply these indices to analyze the link between environmental regulation and
industrial performance. In the short term, more stringent environmental regulations clearly
reduce total factor productivity (TFP) in production of market outputs. With technological
change, however, the short-run cost of regulation could conceivably be offset if they stimulate
innovation and increase productivity in long term. In this context, environmental regulation may
provide a win-win solution, as suggested by the Porter hypothesis. We allowed for the dynamic
aspect of the Porter hypothesis to be tested by examining the impact of environmental regulation
adopted in a given year on technological change and productivity through the period when the
DRAFT
8
impact dissipates. This allows us to identify both the immediate impact of regulations on
productivity, as well as the longer-term impacts.
Some scholars challenge Porter’s hypothesis arguing that technological change precedes
and drives high environmental stringency (e.g., Meyer, 1993) as implied by environmental
Kuznets curve5. The environmental Kuznets curve is based on the observation that high level of
economic development, coupled with increased environmental awareness and better technology,
may result in an increase in demand for environmental quality, and a subsequently increase in the
stringency of environmental regulations. Or an increasing rate of technological change may
otherwise cause regulatory agencies to increase the stringency of environmental regulations. For
example, U.S. EPA’s technology-based standards are based on concepts like Best Conventional
Technology (BCT) or Best Available Technology (BAT), which change over time as new
technologies are developed and implemented. Thus, advances in technology could cause more
stringent environmental regulations.
Thus, when testing the Porter hypothesis it is critical to identify the direction of causality
between environmental regulations and advances in environmental technology, because causality
between regulations and innovation may go in either (or both) directions. These questions seek
empirical answers, and our study attempts to contribute to the literature, empirically and
methodologically.
III. Modeling
Data Envelopment Analysis
Production frontier analysis provides the Malmquist indexes (e.g., Malmquist, 1953;
Caves et al, 1982a, 1982b), which can be used to quantify productivity change and can be
5 Note environmental Kuznets curve test is also suggested to consider both directions using system of equation
(Stern 1998).
DRAFT
9
decomposed into various constituents, as described below. Malmquist Total Factor Productivity
(MTFP) is a specific output-based measure of TFP that includes all categories of productivity
change, measured as the ratio of two associated distance functions (e.g., Caves et al, 1982a,
1982b). Under Variable Returns to Scale (VRS) following Ray and Desli (1997), the Malmquist
index can be decomposed into measures associated with technological change, efficiency change
and scale change:
MTFPVRS = TCVRS · ECVRS · SCVRS.
where TCVRS is technological change under VRS, ECVRS is efficiency change under VRS and
SCVRS is scale change6. Technological change measures shifts in the production frontier.
Efficiency change measures changes the position of a production unit relative to the frontier, so-
called “catching up” (Färe et al. 1994). Scale change measures shifts in productivity due to
changes in the scale of operations relative to the optimal scale.
In the endogenous growth theory framework, technological change is decomposed into
two categories: innovation and learning-by-doing (e.g., Young 1993). This relates to the two
models of technological change—innovation (e.g., Romer, 1990), that focuses on the creation of
distinct new technologies, and learning-by-doing (e.g., Arrow, 1962), that looks at incremental
improvements in productivity with existing technologies.
We use DEA to calculate these component productivity measures. DEA is a methodology
for organizing and analyzing data, and for identifying best practice frontiers. It represents a set of
nonparametric mathematical programming techniques developed to estimate the relative
efficiency of production units. DEA is not conditioned on the assumption of optimizing behavior
on the part of every individual observation, nor does DEA impose any particular functional form
on production technology. Avoiding these maintained hypotheses may be an advantage,
DRAFT
10
particularly for analyses with micro data that extend over a long time series, where assumptions
of technological efficiency of each and every production unit in all time periods might be
suspect.
We use several different versions of the model to measure and decompose productivity
changes (Table 1). First, a base model is used to calculate total productivity change, which
measures the total effect of increases in productivity due to improvements in technology for the
multi-production production of market and non-market (environmental) goods. Next we
decompose total TFP change into improvements in productivity for market outputs and
improvements in productivity for environmental outputs:
TFPTotal = TFPMarket · TFPEnvironment,
where TFPTotal is the total measure of TFP, which is a measure of the total change in
productivity, including both increases in productivity of oil and gas (TFPMarket) and increases in
productivity of environmental outputs (TFPEnvironment).
To carry out this decomposition, the first model includes variables that measure
environmental effects: measures of historic pollution outputs from the Gulf of Mexico, and the
second model excludes variables that measure environmental effects. When these environmental
input/output variables are included in DEA, DEA calculates technological change after
accounting for changes in these environmental effects. The DEA results with this model provide
our measure of TFP change for production, which measures increases in productivity after
accounting for environment effects. The environment TFP effect is then calculated as:
TFPEnvironment = TFPTotal / TFPMarket.
6 Unless otherwise indicated, we assume VRS throughout.
DRAFT
11
Thus, dividing the total measure of productivity change from Model 1 by the productivity
measure from Model 2 provides the measure of the improvement (or decrease) in productivity
due to productivity change in the environmental sector.
Next we decompose these measures of productivity change into indexes that represent
specific technological innovations and a residual, which we generally term learning-by-doing, as
defined in Managi, Opaluch, Jin, and Grigalunas (2001). Thus, the indexes of market and
environmental technological change (Model 1) are decomposed as7:
TCj = TCjInnov · TCj
LBD
where TCj is the index of technological change in sector j (j=market, environmental), TCjinnov is
the technological change associated with identifiable new technologies and TCLBD is the total
index of technological change that cannot be explained by specifically identifiable new
technologies, and includes such factors as learning-by-doing (Arrow, 1962) and other non-
structural factors. Following Managi, Opaluch, Jin, and Grigalunas (2001), we also decompose
the efficiency change (EC) into two indexes representing diffusion of identifiable new
technologies and a residual that is not explained by identifiable technologies.
When analyzing productive efficiency for extraction of non-renewable resources such as
the oil and gas industry, one face challenges not met in typical areas of production of goods and
services. Production from an oil field at some point in time depends upon past production from
the field due to depletion effects, in addition to the technology employed and other geologic
characteristics of the field (e.g., field size, porosity, field depth, etc). Holding inputs constant,
output from a given field follows a well known pattern of initially increasing output, obtaining a
peak after some years of production, then following a long path of declining output. For this
7 Same method is used also for the index of technological change for oil and gas production (Model 4) (see
Managi, Opaluch, Jin, and Grigalunas, 2001).
DRAFT
12
reason we measure productivity change by looking at relative productivity across fields of
different vintages, thereby separating productivity effects associated with aging of the field from
effects due to differences in the state of technology.
Chung, Färe and Grosskopf (1997) introduced a directional distance function to include
improvements in environmental outputs under Constant Returns to Scale (CRS)8. In contrast to
the Shepard (1970) output distance function that measures efficiency by expanding all outputs
simultaneously, the directional distance function measures efficiency due to increasing desirable
outputs (market goods) while decreasing undesirable outputs (e.g., pollution emissions). Using
the directional distance function specification, our problem can be formulated as follows. Let
x = (x1,...,xM)∈RM+, b = (b1,...,bL)∈RK
+, y = (y1,...,yN)∈RN+ be vectors of inputs, environmental
output (undesirable output) and market outputs, respectively. Define the technology set (Q) by
Qt = {(xt, bt, yt): xt can produce (yt, bt) }.
Qt represents the set of all output vectors, y and b, which can be produced using the input vector,
x. The directional distance function is defined at t as
},),,(:sup{);,,( ttttttttttod Qgbxygbxy ∈+= φφ
where g is the vector of directions which outputs are scaled. For this output oriented distance
function, we define g=(y, 0, -b), i.e. desirable outputs are proportionately increased, inputs are
held fixed and environmental outputs (pollution) are proportionately decreased.
Since we use a vintage model, the DEA formulation differs from that in Chung, Färe and
Grosskopf (1997). Our DEA formulation is as follows. Let k be field index, t be time (i.e.,
8 Other environmental performance analysis in DEA include Färe, Grosskopf, Lovell and Pasurka (1989), Färe,
Grosskopf, Lovell and Yaisawarng (1993), Coggins and Swinton (1996), Tyteca, (1997), Boyd and McClelland(1999) and Hailu and Veeman (2000).
DRAFT
13
year), ik be time of discovery for field k (i.e., discovery year), and jk be the number of years since
discovery of field k (i.e., field year). Thus, for each field, jk = t - ik.
Chung, Färe and Grosskopf (1997) define an output-oriented Malmquist-Luenberger
productivity index that is comparable to the Malmquist productivity index, but that includes
productivity changes with respect to both desirable and undesirable outputs. In the vintage
model, we consider all fields discovered in the same year a vintage group, and we calculate the
distance function for each discovery year (i), or vintage group. We then calculate the Malmquist
productivity index by comparing distance functions in two different vintages (i and i+1).
Consider discovery year i, the distance function for field k’ in field year j’ is calculated as
',''''''''''' max)]|g;,,,([ jki
jki
jki
jki
jki
jk
to VRSd φ=baxy
subject to( )
', '' '
( ) 0(1 ) 0, 1,..., ,
J kk j i i
k j n kj kjnk K i j
y y n Nφ λ∈ =
− + + ≥ =∑ ∑
,,...,1,0)1()(
)(
0''
',' Llbb i
kjliKk
kJ
jkj
i
ljk
jk ==−− ∑ ∑∈ =
λφ
,,...,1,0)1()(
)(
0''
',' Mmxx i
kjmiKk
kJ
jkj
i
mjk
jk =≥−− ∑ ∑∈ =
λφ
,,...,1,0)1()(
)(
0''
',' Ggaa i
kjgiKk
kJ
jkj
i
gjk
jk =≥−− ∑ ∑∈ =
λφ
),(,...,1),(,1)(
)(
0
kJjiKkiKk
kJ
jkj =∈=∑ ∑
∈ =
λ
).(,...,1),(,0 kJjiKkkj =∈≥λ
where a is attributes. K(i) includes all fields discovered in i and J(k) is the last field year for field
k. For the mixed period distance function, we have two vintage years i and i+1. For example,
the output constraint is
DRAFT
14
( )', ' 1
' '( 1) 0
(1 ) 0, 1,..., ,J k
k j i ik j n kj kjn
k K i jy y n Nφ λ +
∈ + =
− + + ≥ =∑ ∑
Our vintage model differs from conventional DEA models in two ways. First, for each
year (t), we consider only fields discovered in that year (i = t). Fields discovered in other years
are excluded from calculation of the distance function, although they may be producing at t. In
addition, in the constraints, we summarize across both fields (k) and field year (j), although field
year is not in the discovery year (except j = 0). Thus, we calculate the distance function for each
field k’ in field year j’.
In our study, t and i extend from 1968 through 1995; the vectors of outputs (y and b),
inputs (x) and attributes (a) are listed in Table 1. The weighted innovation index at t is assigned
to vintage group i = t, and held constant for all field years (j) in that group (i). Besides the two
depletion variables, other attribute variables (e.g., water depth) vary across fields, but are
constant over time for a given field. We use cumulative values for inputs (x) and outputs (y),
because for the above technology definition (i.e., x can produce y), it is more appropriate to
express the production relationship on cumulative terms for a nonrenewable industry. For
example, the production in a field at time t is determined by the total number of production wells
previously drilled in the field and total stock depletion in the field to date.
Assessment of the Porter Hypothesis
As discussed above, the Porter hypothesis states that well designed environmental
regulations can spur technological change, in the long run leading to increased productivity and
ultimately to increased profits (Porter, 1991; Porter and van der Linde, 1995). Technological
progress, however, is quite complex and still poorly understood. Contemporaneous analysis of
regulation is needed to find the immediate cost of implementing the regulation. But the ultimate
DRAFT
15
impact of environmental regulations will be felt several years later when the induced innovation
process has been completed.
This study tests the Porter hypothesis by examining whether lagged levels of the
stringency of environmental regulations are associated with increases in technological change.
We use two sets of hypothesis tests, one set based on an Almon distributed lag model (Almon,
1965) and a second set based the Granger causality tests (Granger, 1969).
The Almon lag model relates the technological innovation to prior environmental
regulations using the functional specification:
t
N
iitit EP εβα ++= ∑ −
where Pt denotes the productivity index at time t, α is a constant term, Et-i denotes lagged
environmental regulation index, βi is the coefficient of the ith lag and εt is a stochastic term. The
expected dynamic lagged effects of independent variables can be examined by imposing
theoretical restrictions on the coefficients of the lagged values of these variables. In general, the
Almon polynomial distributed lag model is an estimation procedure for distributed lags that
allows the coefficients of the lagged independent variables to follow a variety of patterns as the
length of the lags increases. The degree of these polynomials may vary in order to capture the
different lag distributions. An inverted “U” pattern is expected, and a second-degree polynomial
is considered appropriate to characterize the lag structure.
The use of Almon lags requires the determination of the appropriate lag periods.
Following common practice (e.g., Harvey, 1990), we choose the lag length is to minimize the
DRAFT
16
Akaike Information Criteria (AIC)9. We use this model as an empirical framework to test the
possible causal linkages between technological innovation and environmental regulations.
We also apply the Granger causality test (Granger, 1969) to examine the direction of
causality between environmental regulations and innovation. The Granger test is related to the
concept of precedence. A time series y1 is said to Granger cause y2 if the prediction of y2 can be
improved upon by the inclusion of lagged values of y1 in the information set used to predict y2.
In our case, the Porter hypothesis is consistent with a finding that lags on the stringency of
environmental regulations have positive and statistically significant coefficients in the equation
for technological change.
Our Granger test proceeds as follows. First, data vectors for productivity change (PC)
and environmental stringency (ES) are partitioned into current values and a series of time lags
[PCt, ESt] and [PC-, ES-], where the superscript indicate PC- and ES- are vectors of lags on
productivity change and environmental stringency. Then the Granger causality test is employed
in the multivariate setting using a vector autoregression (VAR) (Johansen 1988; Johansen and
Juselius 1990). The VAR with this partitioning is represented as follows:
+
=
−
−
2
1
2221
1211
εε
ESPC
∆∆∆∆
ESPC
We test the null hypotheses is that stringency of environmental regulations does not
“Granger cause” productivity change (∆12 =0) and the hypothesis that productivity change does
not “Granger cause” environmental stringency (∆21=0) using Wald statistics. The results of these
tests are used to identify the causal influence between environmental compliance and each
component of productivity.
9 Note that this same lag also maximizes adjusted R2, which is an alternative recommended criterion.
DRAFT
17
The Porter hypothesis states that well designed environmental regulations could provide a
net positive contribution to future productivity, which is consistent with the sum of the lagged
coefficients being positive. Thus, the Porter hypothesis implies that the net effect of past levels
of environmental regulation has a positive influence on future technological change. In contrast,
the traditional view that environmental regulation can only decrease productivity is consistent
with the result ∆12 < 0.
Our second hypothesis is based on the observation that past levels of technological
change induce future increases in the stringency of environmental regulations (∆21 > 0). This
could be the case, for example, because environmental regulations are technology-based, and
therefore technological advances could result in a subsequent increase in the stringency of
environmental regulations.
ApplicationWe apply the above methods to oil and gas production in the Gulf of Mexico. The Gulf
of Mexico is one of the first areas in the world to begin large scale offshore oil and gas
production. Since then, offshore operations in the Gulf of Mexico have played an important role
in production and stabilization of energy supply in United States. Federal offshore oil and gas
production accounted for 26.3 and 24.3 percent of total U.S. production, respectively (U.S.
Department of Interior, 2001), and the offshore fraction of production has been increasing over
time. Oil and gas production in Gulf of Mexico accounted for 88 and 99 percent, respectively, of
total U.S. offshore oil and gas production through 1997 (U.S. Department of Interior, 2001).
Reducing the environmental impact of offshore operations is among the most pressing
challenges facing the oil and gas industry in U.S. today. In recent decades, environmental
concerns led to numerous new regulations imposed on oil and gas operations. These regulations
provide a basis for many environmental improvements by industry, and compliance has become
DRAFT
18
costly and increasingly complex. In 1996, the petroleum industry, including refining, spent as
much on environmental protection as it spent searching for new domestic supplies: an estimated
$8.2 billion (American Petroleum Institute, 2001). Jin and Grigalunas (1993) examine the impact
of environmental regulation on firms in the oil and gas industry using optimal control theory
assuming a constant technology. Their results indicate that rising environmental compliance
costs lead to reductions in investment and production, implying that fewer resources will be
developed over time, and the associated economic benefits will decline.
The command and control method of regulation is used in offshore oil and gas
operations. There has been growing recognition for the need for more cost-effective approaches
to environmental protection. For example, API recently called for “common sense” in
regulations (API, 1996), including a recommendation for more a flexible regulatory framework
in determining how best to meet standards, potentially yielding the same environmental benefits
at lower costs.
Data used in this analysis are obtained from the U.S. Department of the Interior, Minerals
Management Service (MMS), Gulf of Mexico OCS Regional Office. Specifically, we develop
our project database using five MMS data sets:
(1) Production data, including monthly oil, gas, and produced water outputs from every well
in the Gulf of Mexico over the period from 1947 to 1998. The data include a total of
5,064,843 observations for 28,946 production wells.
(2) Borehole data describing drilling activity of each of 37,075 wells drilled from 1947 to
1998.
(3) Platform data with information on each of 5,997 platforms, including substructures, from
1947 to 1998.
DRAFT
19
(4) Field reserve data including oil and gas reserve sizes and discovery year of each of 957
fields from 1947 to 1997.
(5) Reservoir-level porosity information from 1974-2000. This data includes a total of
15,939 porosity measurements from 390 fields.
Because the early data did not include environmental reporting, we use the data for 1968-1998
from above data set. Thus, the project database is comprised of well-level data for oil output, gas
output, produced water output, and field-level data for the number of exploration wells drilled,
total drilling distance of exploration wells, total vertical distance of exploration wells, number of
development wells drilled, total drilling distance of development wells, total vertical distance of
development wells, number of platforms, total number of slots, total number of slots drilled,
water depth, oil reserves, gas reserves, original proved oil and gas combined reserves in BOE,
discovery year, and porosity.
Although we have well-level production data, the well level is not a good unit for
measuring technological efficiency due to spillover effects across wells within a given field.
Rather, the field level is a more appropriate unit for measuring technological efficiency. For this
reason the relevant variables were extracted from these MMS data files and merged by year and
field, so that the final data set was comprised of annual data at the field level over a 28-year time
horizon. On average there are 406 fields operating in any particular year, and a total of 10,964
observations.
Our environmental output data set is composed of 33 different types of water pollutants
in the four EPA categories, and oil spill volume from Coast Guard. The four categories are
conventional pollutants, non-conventional organic pollutants, non-conventional metal pollutants
and radionuclides. Conventional pollutants include oil, grease and total suspended solids (TSS).
DRAFT
20
Non-conventional organic pollutants include benzene, benzo(a)pyrene, chlorobenzene, di-n-
butylphthalate, ethylbenzene, n-alkanes, naphthalene, P-chloro-M-cresol, phenol, steranes,
toluene, triterpanes, total xylenes, 2-butanone and 2,4-dimethylphenol. Non-conventional metal
pollutants include alminum, arsenic, barium, boron, cadmium, copper, iron, lead, manganese,
mercury, nickel, silver, titanium and zinc. Radionuclides include radium 226 and radium 228.
To measure a tendency towards stringent environmental regulation, we use the environmental
compliance cost for preventing water pollution and oil spills. Our environmental compliance cost
is based on ex-ante estimates since we do not have the ex post cost studies10. We compiled a data
file for water pollution and oil spill prevention costs from Federal Register and EPA documents
which contains the ex-ante capital cost and operation and maintenance cost estimates by EPA for
each set of regulations. These environmental regulations require phased implementation over a
period of years and regulations are occasionally revised, which implies a variation in stringency
over time.
Output variables in our model are oil production, gas production, the vector of 33 water
pollution parameters described above and oil spills. Our input variables include number of
platforms, platform size, number of development wells, number of exploration wells, average
distance drilled for exploratory wells, average distance drilled for development wells, untreated
produced water and environmental compliance cost. Field attributes are water depth, initial oil
reserves, initial gas reserves, field porosity, and an aggregate measure of resource depletion,
based on total extraction of oil and gas reserves in the Gulf of Mexico to date for each time
period. Further description of the data is provided in Appendix.
10 Harrington, Morgenstern and Nelson (2000) looked at ex ante cost estimates of environmental regulations to the
ex post cost estimates and compared the accuracy of estimates of the direct costs of more than two dozenregulations. They conclude, at least for EPA and OSHA rules, unit pollution reduction costs estimates are oftenaccurate.
DRAFT
21
One goal of the study is to measure productivity effects and decompose the effects into
those associated with specifically identifiable new technologies versus less structural effects,
such as learning-by-doing. First, we adapt Moss’ (1993) methodology to focus on technological
innovation, rather than diffusion, and we extend the index for our full study period, from 1968 to
1998. We modify the Moss index to reflect innovation by counting only the first time a
particular technology is reported, so our index measures technological innovation rather than
diffusion11.
We further refined the Moss innovation count to consider the relative importance of
particular innovations using a study by the National Petroleum Council (NPC). The NPC carried
out an industry survey to analyze the technological needs of the industry and to identify the
expected impact of specific technological innovations, both in the short term and the long term
(National Petroleum Council, 1995). The 89 companies who responded to the survey account for
about 50 percent of total U.S. reserves. We use the results of this industry survey to construct a
cumulative weighted technology innovation index at time t, calculated as12:
.0 1
,,∑∑= =
×=t
tt
I
iti
NWtit
W InnovwInnov
where InnovWt is the cumulative weighted technology innovation index at time t; wi,t is the
weight for technology in category i at time t; InnovNWi,t is the non-weighted technology
innovation count adapted from Moss in category i at time t.
In addition to this weighted innovation index, one important innovation of the recent
decades is the extent of horizontal and directional drilling. Horizontal drilling refers to the ability
to guide a drillstring to deviate at all angles from vertical, which allows the wellbore to intersect
the reservoir from the side rather from above. This allows a much more efficient extraction of
11 Moss (1993) constructs a technology diffusion index that counts technology diffusion as it is reported in
DRAFT
22
resources from thin or partly depleted formations. Horizontal drilling is also advantagous for
formations with certain types of natural fractures, low permiability, a gap cap, bottom water, and
for some layered formations. A measure of horizontal & directional drilling13 and our weighted
innovation variable are used in the DEA framework to partition impacts of technological change
into components associated with specific technological innovations and more routine learning by
doing in the following.
IV. Empirical Results
Data Envelopment Analysis
The DEA framework was used to measure productivity change and to carry out the
various decompositions described above, thereby contributing to a better understanding of nature
of technological change for our application. Figure 1 presents the results for total TFP change,
and for TFP change decomposed into the oil & gas, and environmental sectors. Overall, total
TFP increases by about 65% from 1968 through 1995, or a geometric mean of about 1.9% per
year. Over the first 16 years of the time period (1968 through 1984), total TFP increases by
about 17%, or a rate of about 1.0% per year. In the next 10 years (1985 through 1995) total TFP
increases by about 34%, or a rate of about 3.2% per year. This is consistent with the increasing
rate of technological progress that has been observed in the industry (e.g., Bohi, 1997).
The results of TC for total, oil & gas, and environmental sector are presented in Figure 2.
Total, oil & gas production, environmental TC increase by about 48%, 35% and 10% from 1968
through 1995, respectively. Over the 28 year study period, TC change can be partitioned into
81.5 % due to oil and gas production sector, and 18.5% due to environmental sector. Thus, while
there has been significant improvement environmental technology over the study period, it has
industry trade journals.
12 See Managi, Opaluch, Jin and Grigalunas (2001) for precise description.
DRAFT
23
lagged far behind that for market outputs. This should, perhaps, not be surprising, since the
technology-based standards employed allow little latitude for innovation. Indeed, there could be
a significant disincentive to industry developing new environmental technologies, since
innovation could result in more stringent regulations in the future.
Figure 4 shows the trends for innovation, learning by doing and diffusion for
environmental technologies over period from 1968 through 1995. As discussed above, we find a
much more modest increase in TFP for the environmental sector, as compared to the market
sector (see Figure 3). Innovation, LBD and diffusion increase by about 3%, 7% and 10% for
the study period, respectively. Diffusion plays an important role in environmental TFP over the
whole study period. Ignoring the rapid increase in LBD during the last 3 periods' rapid,
cumulative diffusion effect is about 4.7 and 2.9 times as large as the affect associated with LBD
and innovations on 1993, respectively.
The trend in LBD suggests that despite the fact that environmental regulations are of the
command-and-control variety, to some extent industry is able to moderate the unit costs of
compliance over time. Policy makers can not estimate this LBD compliance cost reduction
effects ex ante, when regulations are being developed, since this LBD effect comes from the
engineer’s and manager's experience after regulations are implemented. This tends to lead to ex
ante estimates that overestimate ex post pollution control costs. Note that our LBD effects are
small compared to diffusion effects, and are much smaller than estimated LBD effects for market
outputs (Manage, Opaluch, Jin and Grigalunas, 2001), suggesting that technology-based
regulations allow relatively little flexibility for cost saving, and ex ante unit cost estimates are
relatively accurate. This is consistent with the results in Harrington, Morgenstern and Nelson
(2000), who conclude that unit pollution reduction costs estimates are often accurate for EPA
13 Appendix A describes the method used for calculating the measure of horizontal and directional drilling.
DRAFT
24
regulations. The technology-based methods used for offshore operations, limit innovative
approaches to pollution abatement. For example, BAT may lead to the adoption of new
technologies in the short-run, but as Jaffe et al. (2002) point out the long-term effect may be to
discourage innovation simply because emission standards could be tightened each time the firm
identifies a lower cost approach. These findings of modest levels of productivity increases in the
environmental sector are consistent with the conceptual literature, which demonstrates that
incentive-based environmental policies have far greater potential to encourage innovation than
command and control-based policies.
Overall effects on joint production of oil & gas and environmental outputs are presented
in Figure 5. Innovation, LBD and diffusion lead to TFP increases of about 20%, 25% and 30%,
respectively, over the study period. Note that diffusion plays the most important role until the
end of the time horizon, when there is a clear trend towards LBD and innovation. LBD and
innovation increase by about 2.2% and 1.4% per year, respectively. Increase in innovation
mainly comes from oil & gas production sector, rather than the environmental sector. The
increase productivity due to LBD came from both of oil & gas production and the environmental
sector.
Additional insights into the nature of technological change can be obtained by identifying
the extent to which it conforms to Hicks neutrality. Hicks neutrality of technological change for
joint production implies parallel shifts in isoquants on the input side and parallel shifts in the
production possibility frontiers on the output side. In contrast, biases in technological change
imply relative changes in productivity across inputs and/or outputs, which imply non-parallel
shifts of isoquants and/or production frontiers.
DRAFT
25
DEA decomposes productivity change into output biased technological change (OBTC),
input biased technological change (IBTC) and magnitude change (MC). However, DEA only
provides overall, absolute measures of input and output biases (Färe and Grosskopf, 1996), in
contrast to the parametric measurements of bias (e.g., Antle and Capalbo, 1988), which provide
relative measures, such as input using or saving for each factor. The DEA measure of OBTC
identifies the extent to which technological change increases productivity of each output relative
to all other outputs. Similarly, IBTC identifies whether technological change increases
productivity of each input relative to all other inputs. When the DEA measures of OBTC and
IBTC simultaneously equal 1, productivity change is Hicks neutral, and productivity change
equals the magnitude change component of the DEA decomposition.
With Model 2 we find an IBTC measure of 1. 81 and OBTC measure of 1. 29. Therefore,
the overall technological change bias index, which is the product of IBTC and OBTC, is 2.32.
Unfortunately, DEA is not a statistical technique, and therefore does not allow one to test for
statistical significance. However, the overall bias index is sufficiently far from one to suggest
that Hicks neutral technological change probably does not hold in Model 2, which includes
market outputs only. In comparison, we find larger biases on both the input and output sides
(IBTC=1.96 OBTC=1.60) for the joint production model, which includes market and
environmental outputs. These higher biases in technological change are consistent with
observation that technological change with respect to market goods is much larger than
technological change with respect to environmental goods, suggesting a bias in technological
change towards market outputs relative to environmental outputs.
Testing the Porter Hypothesis
DRAFT
26
As discussed above, the Porter hypothesis is based on the notion that well defined
environmental regulations spur technological change, which could lead to increased productivity
and, ultimately, increased profits for the regulated industry. Below we provide two tests of the
Porter hypothesis. First, we use Almon Distributed lag models to test whether increases in the
stringency of environmental regulations affect various measures of productivity for production of
market goods and for joint production of market and environmental goods.
Next we use Granger causality tests to explore causal directions between changes in
productivity and environmental stringency. To do so, we use technological change (TC) and
total factor productivity (TFP) as dependent variables, and we use lags on environmental
stringency as explanatory variables. When TC is the dependent variable, the model provides a
test of changes in the efficiency frontier, and when TFP is the dependent variable, the model
provides a test for overall changes in productivity, including efficiency change (EF) or “catching
up” by inefficient production units. We also test causality in the other direction, where we test
whether productivity change affects stringency of environmental regulations. One might expect
a positive finding of causality, since new regulations are based on the state of technology. Thus,
improvements in technology could result in subsequent increases in the stringency of
environmental regulations. Additionally, a finding of this sort would provide support for the
environmental Kuznets curve, i.e. a high level of economic development may result in an
increase in demand for environmental quality, thus supporting subsequent increases in the
stringency of environmental regulations.
The standard Porter hypothesis states that productivity of market outputs, hence
profitability, may increase with the stringency of environmental regulations. However, we are
also interested in testing our re-cast version Porter hypothesis, where efficiency is measured with
DRAFT
27
respect to the joint production of market and environmental outputs. Thus, we carry out these
tests using productivity measures for market outputs (oil and gas production) only and for joint
production of market and environmental outputs.
The results for the Almon lag model are reported in Table 214. The results for the
relationship between the stringency of environmental regulations and joint production of market
and non-market outputs show a high level of statistical significance. Most individual lag
coefficients are statistically significant at the 5% or 1% level, and the aggregate effects are both
significant at better than the 1% level. The initial lags are negative in sign, indicating that the
immediate effect of environmental regulation is to reduce overall productivity. But the longer
term lags and the sum of all lags are both positive, indicating that increases in the stringency of
environmental regulations lead to long term increases in joint productivity of environmental and
market outputs. The sum of all lags is positive and statistically significant, indicating a net
positive relationship between joint productivity and lags in the stringency of environmental
regulations.
However, the examining lags on market outputs only indicate that changes in
environmental stringency has no significant effect of productivity of market goods in the short
term or in the long term. The individual lags and the sum of all lags are not statistically
significant at standard levels. Hence, our results do not support the standard version of the Porter
hypothesis, which states that increases in the stringency of environmental regulations spur
innovation that leads to increases in productivity of market outputs, thereby leading to increased
profits in the long term.
Next, we apply Granger causality tests to further explore the causal relationships between
the stringency of environmental regulations and various productivity measures. The Granger
DRAFT
28
causality test is a more rigorous test than that based on the Almon lag model, discussed above, in
two ways. First, the Granger test identifies whether the lags on the independent variable
(stringency of environmental regulations) adds explanatory power, relative to the model based on
lagged dependent variable (productivity change). Secondly, the Granger test examines causality
in both directions, so that it can potentially distinguish between models where causality goes in
either or both directions.
The first set of hypotheses for the Granger tests relate to the Porter hypothesis. Thus, the
Porter hypothesis is consistent with positive and statistically significant coefficients on lags of
environmental regulations in the regression equation for productivity. The second set of
hypotheses in the Granger tests relate to whether productivity is a determining factor in
determining the stringency of future environmental regulations.
Both sets of null hypotheses are based on non-causality, and are tested using Wald
statistics. So that the Granger causality test is that 'X does not Granger cause Y', and rejecting a
null hypothesis is consistent with a finding of causality. The optimal number of lags is also a
critical issue in Granger causality test. We use a group of information criteria: AIC, Schwarz
Bayesian criteria (SC), and Akaike final prediction error criterion (FPE)15, and we find identical
results for the appropriate lags for each of the three criteria.
As indicated in Table 3, we find that environmental stringency causes TC in the joint
production model, which is consistent with the re-stated version of the Porter hypothesis. In
contrast, we find no significant causality between stringency of environmental regulations and
productivity of market goods, thus rejecting the standard form for Porter hypothesis. Of course,
it should be emphasized that this result is for our application only, and that special circumstances
14 Second degree polynomial with end point restrictions is used. 15 See Giles and Mirza (1999) for discussion about information criteria.
DRAFT
29
might lead to this result. For example, environmental regulations in this industry are command-
and-control oriented, so that there is not much flexibility to develop new technologies to comply
with environmental regulations. The results could differ in a context where regulations were
more flexible, such as when financial incentives are employed.
Finally, we test whether higher productivity leads to more stringent environmental
regulations. We also find a causal link from technological change of market outputs to
environmental stringency, but not for joint production. This finding is more consistent with the
environmental Kuznets curve than with technology-based standards.
V. Discussion and Conclusion
Technological progress plays an important role in addressing environmental problems,
while simultaneously maintaining standards of living. Over the past 50 years our profession has
greatly improved our understanding the process of technological innovation. We have
progressed from “confessions of ignorance”, where time is the only “explanatory” variable in
technological progress, towards an understanding of the mechanisms that drive productivity
change and measurement of various components of productivity change.
This paper contributes to the literature on productivity change in several ways. First, we
apply Data Envelopment Analysis to a unique field-level data set to measure various components
of total factor productivity within a joint production model, which considers both market and
environmental outputs. This contributes to our understanding of the impact environment
controls have had on various components of total factor productivity in this industry, and thereby
the potential for technological change to maintain productivity in the face of increasingly
stringent environmental regulations.
The results show an upward trend in productivity in the Gulf of Mexico offshore oil and
gas industry, despite depletion and increasingly stringent environmental regulations. Our
DRAFT
30
findings indicate improved productivity of environmental technologies, but environmental
productivity change has lagged behind that for market outputs. Over the 28 year study period,
TFP change can be partitioned into nearly 80% due to oil and gas production sector, and about
20% due to environmental sector. This may be due in part to the command-and-control nature of
most environmental regulations, which allow much less flexibility for innovative solutions, as
compared to that for market outputs.
We also analyzed the contribution of technological change and efficiency change in TFP
both for oil & gas production and for environmental technologies. The former comprises
technological innovation and learning by doing, and the latter technology diffusion and other
factors. We developed an index for decomposing technological change into technological
innovation which is associated with discovery of identifiable new technologies, and learning by
doing, which embodies the less structural components of productivity change. We use this index
to estimate the relative importance of technological innovation, learning by doing and
technology diffusion on TFP. Similarly, we isolated technology diffusion from that the rest of
the factors that impact on efficiency change and subsequently TFP. We compared the relative
impact of these technology indicators on TFP in the industry. The results indicate that diffusion
of technological (especially environmental technologies) had a significantly larger impact on
TFP than technological innovation and learning by doing. This is important for providing an
improved understanding of the process of technological change, and could contribute to design
of effective policy.
Next we apply two models to understand the dynamic relationship between the stringency
of environmental regulations and productivity, and thereby test the Porter hypothesis. The
Porter hypothesis states that well designed environmental regulations could spur innovation,
DRAFT
31
leading to long run increases in productivity and hence increased profits for the regulated
industry. We recast the Porter hypothesis to explore the relationship between environmental
regulations and productivity more fully. Specifically, we test whether environmental regulations
enhance joint productivity of environmental and market outputs, in addition to the standard
Porter hypothesis which applies to productivity of market outputs only.
Our results support the recast version of Porter hypothesis, which examines productivity
of joint production of market and environmental outputs. But we find no evidence for the
standard formulation of the Porter hypothesis, which states that increased stringency of
environmental regulation leads to increased productivity of market outputs. This finding could
be due in part to the command-and-control design of environmental regulations in offshore oil
and gas, which historically has not provided much latitude for innovation in achieving
environmental goals. Flexible environmental regulations are required to provide incentives for
innovation, which could result in a net positive effect on productivity. In contrast, inflexible
technology-based environmental regulations are less likely to contribute to innovation.
This suggests that we must be careful to maintain a realistic view of the potential for
environmental regulations. An overly naïve conviction that there exists a near universal potential
for win-win solutions in environmental problems could be used to justify poorly conceived
environmental policies.
DRAFT
32
Appendix. Data Construction
Data used in this analysis are obtained from the U.S. Department of the Interior,
Minerials Management Service (MMS), Gulf of Mexico OCS Regional Office. Relevant
variables are extracted from these MMS data files and merged by year and field. Using these
variables, the Gulf of Mexico regional new resource discovery and depletion over time are
constructed. The variable for horizontal & directional drilling is defined as the ratio of total
drilling length to vertical drilling length. Larger values imply a higher ratio of horizontal &
directional drilling relative to vertical drilling. We take cumulative input value data, as literature in
the oil and gas supply suggest, for well, drilling, horizontal & directional drilling.
Drilling is assumed to affect output starting the following year, since current drilling does
not affect current production. So well inputs in period t is determined by cumulative drilling
through period t-1. At any point in time ultimately recoverable resources are not known. Rather, the
remaining resource is estimated using current economically (not physically) known reserve stock
minus resources produced to date. We also take cumulative value for output variables including oil,
gas production to take account of the technological characteristics. We measured depletion effect on
the remaining stocks of oil and gas in the Gulf of Mexico at period t-1 by expressing it as a ratio of
terminal production. This captures the effects how difficult the new discovery is in the Gulf of
Mexico.
DEA requires the data on input usage and on characteristics that determine output.
Therefore, we created the water depth input as maximum water depth in the Gulf of Mexico minus
water depth in each field since production is more difficult in deeper water depth in given
technology. Units of oil and gas production are barrel and thousand cubic feet, respectively.
Platform size is defined as average number of slots per platform for the field. Units of oil and gas
DRAFT
33
reserves are million barrel and billion cubic feet, respectively. Units of remaining oil and gas
reserves in the Gulf of Mexico, and porosity are measured in percent terms. Missing porosity is
estimated using two-step estimation procedure to correct for this omitted variable problem
(Heckman 1979; Greene 1981) (see precise statistics in Managi, Opaluch, Jin and Grigalunas,
2001). The data source of water pollution includes EPA regulations. All of water pollution data
are concentration units and are multiplied to untreated produced water, where MMS is the data
source, in field level as a pollution output, assuming pollution concentration is constant over
fields each year since we do not have field base pollution concentration data. More precise
description of data is in Managi (2002).
DRAFT
34
Reference
Almon, Shirley. 1965. The Distributed Lag Between Capital Appropriations and Expenditures.Econometrica 33: 178-196.
American Petroleum Institute. 1996. Achieving Common Sense Environmental Regulation: Oiland Gas Exploration and Production. Washington, D.C.: American Petroleum Institute.
American Petroleum Institute. 2001. U.S. Petroleum Industry's Environmental Expenditures1990-1999. Washington D.C.: American Petroleum Institute.
Antle, John M., and Susan M. Capalbo. An Introduction to Recent Developments in ProductionTheory and Productivity Measurement, in Capalbo, Susan M.; Antle, John M., eds., 1988,Agricultural Productivity: Measurement and Explanation. Washington, D.C.: Resourcesfor the Future; Distributed by the Johns Hopkins University Press.
Arrow, Kenneth J. 1962. The Economic Implications of Learning by Doing. Review of EconomicStudies 29:155-73.
Barbera, Anthony J and Virginia McConnell. 1990. The Impact of Environmental Regulationson Industrial Productivity: Direct and Indirect Effects. Journal of EnvironmentalEconomics and Management 18 (1): 50-65.
Berman, Eli and Linda T. M. Bui. 2001. Environmental Regulation and Productivity: Evidencefrom Oil Refineries. Review of Economics and Statistics 83 (3): 498-510.
Bovenberg, A. Lans, Sjak A. Smulders. 1996. Transitional Impacts of Environmental Policy inan Endogenous Growth Model. International Economic Review 37 (4): 861-93.
Boyd, Gale A. and John D. McClelland. 1999. The Impact of Environmental Constraints onProductivity Improvement in Integrated Paper Plants. Journal of EnvironmentalEconomics and Management 38 (2): 121-42.
Caves, Douglas W, Laurits R. Christensen, and W. Erwin Diewert, 1982a. MultilateralComparisons of Output, Input and Productivity Using Superlative Index Numbers.Economic Journal 92 (365): 73-86.
Caves, Douglas W, Laurits R. Christensen, and W. Erwin Diewert, 1982b. The EconomicTheory of Index Numbers and the Measurement of Input, Output and Productivity.Econometrica 50(6): 1393-1414.
Charnes, A., W. W. Cooper, and E. Rhodes. 1978. Measuring the Efficiency of Decision MakingUnits. European Journal of Operational Research 2(6): 429-444.
Chung, Y. Rolf Färe, and Shawna Grosskopf. 1997. Productivity and Undesirable Outputs: ADirectional Function Approach. Journal of Environmental Management 51: 229-240.
DRAFT
35
Coggins, Jay S. and John R. Swinton. 1996. The Price of Pollution: A Dual Approach to ValuingSO2 Allowances. Journal of Environmental Economics and Management 30 (1): 58-72.
Cuddington, J.T., and D. L. Moss. 2001. Technical change, depletion and the U.S. petroleumindustry: a new approach to measurement and estimation. American Economic Review91(4): 1135-1148.
Färe, R., S. Grosskopf, and C.A. Knox Lovell. 1985. The Measurement of Efficiency ofProduction. Boston: Kluwer-Nijhoff.
Färe, R., S. Grosskopf, C.A.K. Lovell and C. Pasurka. 1989. Multilateral ProductivityComparisons When Some Outputs are Undesirable: a Nonparametric Approach. TheReview of Economics and Statistics 71 (1): 90-98.
Färe, R., S. Grosskopf, C. A. K. Lovell, and S. Yaisawarng. 1993. Derivation of Shadow Pricesfor Undesirable Outputs: a Distance Function Approach. The Review of Economics andStatistics 75 (2): 374-80.
Färe, R., S. Grosskopf, M. Norris, and Z. Zhang. 1994. Productivity Growth, Technical Progress,and Efficiency Change in Industrialized Countries. American Economic Review 84(1):66–83.
Giles, J.A. and S. Mirza. 1999. Some Pretesting Issues on Testing for Granger Noncausality.Department of Economics. University of Victoria, Working paper.
Granger, C. W. J. 1969. Investigating Causal Relations by Econometric Models and Cross-Spectral Methods. Econometrica 37 (3): 424-38.
Greene, William. 1981. Sample Selection Bias as a Specification Error: Comment.Econometrica 49(3) 795-98.
Griliches, Zvi, 1994. Productivity, R&D and the Data Constraint American Economic Review84(1): 1-23.
Hailu, Atakelty, and Terrence S. Veeman. 2000. Environmentally Sensitive ProductivityAnalysis of the Canadian Pulp and Paper Industry, 1959-1994: An Input DistanceFunction Approach. Journal of Environmental Economics and Management 40 (3): 251-274.
Hall, Bronwyn H., Adam B. Jaffe, and M. Trajtenberg. 2001. The NBER Patent Citations DataFile: Lessons, Insights and Methodological Tools. NBER Working Paper No. W8498.
Harrington, Winston, Richard D. Morgenstern, and Peter Nelson. 2000. On the Accuracy ofRegulatory Cost Estimates. Journal of Policy Analysis and Management 19 (2): 297-322.
Harvey, A. C. 1990. The Economics Analysis of Time Series New York; London: Philip Allan.
DRAFT
36
Heckman, James. 1979. Sample Selection Bias as a Specification Error. Econometrica 47(1):153-161.
Jaffe, Adam, Steven Peterson, Paul Portney, and Robert Stavins. 1995. EnvironmentalRegulation and the Competitiveness of U.S. Manufacturing: What Does the EvidenceTell Us? Journal of Economic Literature 33: 132-163.
Jaffe, Adam B., and Karen Palmer. 1997. Environmental Regulation and Innovation: A PanelData Study. Review of Economics and Statistics 79 (4): 610-19.
Jaffe, A. B., R. G. Newell, and R. N. Stavins. 2002. Technological Change and the Environment.in Karl-Göran Mäler and Jeffrey Vincent, eds. Handbook of Environmental EconomicsNorth-Holland. Forthcoming
Jin, Di and Thomas A. Grigalunas. 1993. Environmental Compliance and Optimal Oil and GasExploitation. Natural Resource Modeling. 7 (4): 331-52.
Johansen, Soren. 1988. Statistical Analysis of Cointegration Vectors. Journal of EconomicDynamics and Control 12 (2/3): 231-54.
Johansen, Soren, and Juselius Katarina. 1990. Maximum Likelihood Estimation and Inference onCointegration--With Applications to the Demand for Money. Oxford Bulletin ofEconomics and Statistics 52 (2): 169-210.
Jorgenson, Dale W., and Peter J. Wilcoxen. 1990. Environmental Regulation and U.S. EconomicGrowth. RAND Journal of Economics 21 (2): 314-40.
Kemp, R. 1997. Environmental Policy and Technical Change: A Comparison of theTechnological Impact of Policy Instruments. Edward Elgar, Cheltenham, UK.
Kneese, A.V., and C.L. Schultze 1978. Pollution, Prices and Public Policy BrookingsInstitution, Washington, DC.
Lanjouw, J. O. and Ashoka Mody. 1996. Innovation and the International Diffusion ofEnvironmentally Responsive Technology. Research Policy 25: 549-71.
Malmquist, S. 1953. Index Numbers and Indifference Curves. Trabajos de Estatistica, 4 (1):209-42.
Managi, Shunsuke. 2002. Technological Change, Depletion and Environmental Policy inOffshore Oil and Gas Industry. Ph.D. dissertation, University of Rhode Island.forthcoming.
Managi, S. J.J. Opaluch, D. Jin, and T.A. Grigalunas. 2001. Technological Change and Depletionin Offshore Oil and Gas, draft, University of Rhode Island.
Meyer, Stephen M. 1993. Environmentalism and Economic Prosperity: Testing theEnvironmental Impact Hypothesis. MIT Project on Environmental Politics and Policy.
DRAFT
37
Mohr, Robert. 2002. Technical Change, External Economies, and the Porter Hypothesis. Journalof Environmental Economics and Management 43 (1): 158-168.
Moss D. L. 1993. Measuring Technical Change in the Petroleum Industry: A New Approach toAssessing its Effect on Exploration and Development. National Economic ResearchAssociations, Working Paper #20.
Palmer, Karen; Wallace E. Oates, and Paul R. Portney. 1995. Tightening EnvironmentalStandards: The Benefit-Cost or the No-Cost Paradigm? Journal of EconomicPerspectives 9 (4): 119-32.
Portney, P. R. 1994. Does environmental policy conflict with economic growth? Resources 115,21-23.
Porter, Michael. E. 1991. America’s Greening Strategy. Scientific American 264, 168.
Porter, Michael E. and C. van der Linde. 1995. Toward a New Conception of the Environment-Competitiveness Relationship. Journal of Economic Perspectives 9 (4): 97-118.
Ray, Subhash C. and Evangelia Desli. 1997. Productivity Growth, Technical Progress, andEfficiency Change in Industrialized Countries: Comment. American Economic Review 87(5): 1033-39.
Repetto, Robert. 1996. Has Environmental Protection Really Reduced Productivity Growth: WeNeed Unbiased Measures Washington D.C.: World Resources Institute.
Romer, Paul M. 1990. Endogenous Technological Change. Journal of Political Economy 98 (5,pt. 2): S71-S102.
Simpson, D. R. and R. L. Bradford, III. 1996. Taxing Variable Cost: Environmental Regulationas Industrial Policy Journal of Environmental Economics and Management 30: 282-300.
Solow, Robert. 1957. Technical Change and the Aggregate Production Function. Review ofEconomics and Statistics 39:312-320.
Stern D. I. 1998. Progress on the environmental Kuznets curve? Environment and DevelopmentEconomics 3: 175-198.
Tyteca, D. 1997. Linear Programming Models for the Measurement of EnvironmentalPerformance of Firms--Concepts and Empirical Results. Journal of Productivity Analysis8 (2): 183-198.
Ulph, Alistair. 1996. Environmental Policy and International Trade When Governments andProducers Act Strategically. Journal of Environmental Economics and Management 30(3): 265-81.
U.S. Department of Interior, 2001. Technical Information Management System (TIMS)Database, U.S. Mineral Management Service (MMS).
DRAFT
38
Xepapadeas, A. and A. de Zeeuw. 1999. Environmental Policy and Competitiveness: the PorterHypothesis and the Composition of Capital. Journal of Environmental Economics andManagement 37: 165- 182.
Young, Alwyn. 1993. Invention and Bounded Learning by Doing. Journal of Political Economy101 (3): 443-72.
DRAFT
39
Table 1. Model Specifications
Model 1 Model 216 Model 3 Model 4
Index calculated:
BaseModel:TotalTFP
ProductionTFP
InnovationLBD &
Diffusion(Total)
InnovationLBD &
Diffusion(Oil &Gas)
Output VariablesOil production (bbl) X X X X
Gas production (Mcf) X X X XWater Pollution X X
Oil Spill X XInput Variables
Number of platforms X X X XAve. Platform size (#slot /
#platform) X X X X
Number of exploration wells X X X XNumber of development wells X X X XAverage Drilling Distance for
Exploratory Wells X X X X
Average Drilling Distance forDevelopment wells X X X X
Produced Water X X X XWeighted Innovation Index X X
Horizontal & Directional Drilling(Exploratory) X X
Horizontal & Directional Drilling(Development) X X
Environmental Compliance Cost X XAttribute Variables
Water Depth X X X XDepletion Effects (Oil) X X X XDepletion Effects (Gas) X X X XOil Reserves in the Field X X X XGas Reserves in the Field X X X X
Porosity (Field Type) X X X X
16 Model2 and model 4, is analyzed in Managi, Opaluch, Jin and Grigalunas (2001) over 1947-1995.
DRAFT
40
0.91
1.11.21.31.41.51.61.7
1965 1970 1975 1980 1985 1990 1995Year
TFP
TFP(Total)TFP(Market)TFP(Environment)
0.91
1.11.21.31.41.51.61.7
1965 1970 1975 1980 1985 1990 1995Year
TC
TCv(total)TC(Market)TC(Environment)
Figure 1. Total Factor Productivity
Figure 2. Technological Change
DRAFT
41
0.9
1
1.1
1.2
1.3
1965 1970 1975 1980 1985 1990 1995Year
TFP
Effe
ct
LBD(Market)INNOV(Market)DIFF(Market)
0.9
1
1.1
1.2
1.3
1965 1970 1975 1980 1985 1990 1995
Year
TFP
Effe
ct
LBD(Environment)INNOV(Environment)DIFF(Environment)
0.9
1
1.1
1.2
1.3
1965 1970 1975 1980 1985 1990 1995
Year
TFP
Effe
ct
LBD(Total)INNOV(Total)DIFF(Total)
Figure 3. Market (Oil and Gas Production) Sector
Figure 5. Total Sector
Figure 4. Environmental Sector
DRAFT
42
Table 2. Almon Distributed Lags for Impact of Environmental Regulations on AlternativeProductivity Measures
Joint Production ofEnvironmental & Market Outputs
Market Outputs Only
TechnologicalChange
Total FactorProductivity
TechnologicalChange
Total FactorProductivity
TimeLag
0
1
2
3
4
5
6
7
8
-0.185*(-2.04)
0.022***(3.55)
0.122**(2.59)
0.115**(2.46)
-0.376***(-3.52)
-0.110*(1.88)
0.100***(4.94)
0.254***(18.72)
0.352***(10.60)
0.394***(8.79)
0.38***(8.02)
0.31***(7.59)
0.18***(7.32)
0.008(0.06)-0.030(-0.39)-0.030(-0.23)0.008(0.11)
-0.003(-0.02)-0.013(-0.13)0.081(0.08)0.060(0.48)
Sum ofLags
0.075***(3.86)
1.484***(28.46)
0.040(0.42)
0.053(0.84)
Adj. R2 0.361 0.976 0.111 0.112
AIC 88.4 95.1 200.8 172.9
Note: * Significant at 10%, ** Significant at 5%, *** Significant at 1%. t-statistics are reported in parentheses.
DRAFT
43
Table 3. Granger Causality Test for Productivity and Environmental Stringency
Null Hypothesis 1 Null Hypothesis 2
Environmental Stringency Productivity Productivity Environmental Stringency
Productivity Measure χ2 Prob. > χ2 χ2 Prob. > χ2
TC 11.72 0.0388 4.27 0.5132Joint Production ofEnvironmental and
Market Outputs TFP 2.51 0.2855 2.00 0.3528
TC 0.97 0.6142 0.55 0.0378Market OutputsOnly TFP 0.23 0.8912 1.07 0.5848