The impact of state government subsidies and tax credits in an emerging industry: ethanol production 1980-2007.
Cotti, Chad ; Skidmore, Mark
1. Introduction
In 1980, ethanol production in the United States was virtually
nonexistent, but by 2007, production had expanded to 6500 million
gallons. Federal subsidies and mandates requiring ethanol to be mixed
with gasoline are notable driving factors in the emergence of the
biofuel industry in the United States. (1) Currently, the federal
government subsidizes ethanol blended with gasoline at a rate of 51
cents per gallon. (2) Federal subsidies have received much attention
from supporters as well as critics. Much less focus, however, has been
on the subsidies and tax credits provided at subnational levels of
government. Many states offer subsidies and/or tax breaks for the
production and/or consumption of ethanol, and in a number of cases,
these are substantial: Some states offer 20 cents per gallon, and in a
few cases, as much as 40 cents per gallon are offered to ethanol
producers. To illustrate, consider Wisconsin, which offers an annual
subsidy at the rate of 20 cents per gallon up to 15 million gallons.
According to the National Corn Growers Association, a 40 million gallon
plant employs 32 full-time workers.
On a per worker basis for the average-size plant in Wisconsin (53
million gallons), this subsidy is equivalent to about $71,400 per
worker. This is a sizable and costly subsidy if the goal is to generate
jobs, but clearly the goals of such subsidies are much broader than job
creation.
Beginning in the 1980s, state and local governments increasingly
began to rely on subsidies and tax breaks to stimulate growth in
employment and business activity. Between 1984 and 1993, the number of
states offering various incentives increased from 27 to 44 (Chi 1994). A
significant focus of within-state incentive packages is generating
economic activity in economically distressed rural areas (Greenberg and
Reeder 1998). More recently, it has been argued that one of the benefits
of ethanol production and consumption is job creation in rural areas.
(3) What is less clear is whether such incentives actually stimulate
economic development. Much of the academic research evaluating the
impacts of economic development incentives suggests that such incentives
are costly in terms of direct payments or forgone taxes, and the
ultimate gains in development are limited (Bartik 1991). Nearly all of
this work has sought to evaluate the impacts of incentives on general
business activity or activity in well-established markets.
The present study makes several contributions to the body of work
on the emerging biofuel industry and the research that has sought to
evaluate the effectiveness of development incentives. First, while we
have detailed and accurate national data on ethanol production dating
back to 1980, state-level data on ethanol production are not
systematically collected. Using information from multiple sources, we
constructed annual data on ethanol production capacity dating back to
1980 for each state. These data may be of interest to researchers
seeking to understand the evolution and emergence of the ethanol
industry in the United States. Second, we compiled a detailed list of
all state-level tax credits and subsidies with imposition dates over the
1980-2007 period. We use these data to evaluate the role of incentives
offered by subnational governments on the location of ethanol production
plants. To date, little is known about the role of state subsidies in
the emerging biofuel industry, particularly in terms of influencing
location decisions. This research also contributes more generally to the
literature evaluating the importance of incentives in fostering the
development of an emerging industry.
As a prelude to our results, we find that incentives directed at
ethanol production (particularly per gallon credits) have a significant
influence on plant location/production. However, we also show that the
magnitudes of our estimates are sensitive to estimation technique. Next,
we provide a brief review of the most relevant research on the emerging
biofuel industry and the research on effectiveness of development
incentives. In section 3, we present our data and empirical analysis,
and section 4 concludes.
2. Literature Review and Theoretical Discussion
In this section, we present a review of the recent research in the
emerging biofuel industry, as well as the extensive body of work that
has sought to evaluate the effectiveness of state and local government
economic development incentive programs.
Ethanol Production
Much of the research on ethanol production is only peripherally
related to state government incentives. For example, Gardner (2007, p.
19) employs a cost-benefit analysis to evaluate the net societal
benefits of relying more heavily on ethanol as a renewable fuel source.
His analysis evaluates the market impacts and deadweight loss resulting
from the 51 cents per gallon federal ethanol subsidy. He concludes that
"ethanol subsidies and mandates are unlikely to generate social
gains." While such subsidies clearly have an appeal for certain
farm interests (for example, corn producers), the net societal gains of
such policies are questionable. Similarly, Hahn and Cecot (2007) also
employ a cost-benefit method to evaluate current federal biofuel
policies. Consistent with Gardner (2007), they conclude that under
current policies, the costs of increased production are likely to exceed
the benefits.
Hahn and Cecot (2007) regard federal as well as state and local
government subsidies and regulations as the driving force of the ethanol
industry. The following are some of the incentives that "corn
states" (4) are using: Illinois grants up to $5.5 million for the
production of new plants; Indiana offers a 0.125 cents per gallon
production tax credit; Iowa offers 0% interest loans up to half the cost
of the production project; and Missouri offers tax incentives of 20
cents per gallon of ethanol produced. Even many states that are not
typically thought of as corn states are using such measures in an
attempt to attract the industry. Some of these states and the provisions
they offer include: Hawaii, which has a tax credit equal to 30% of
nameplate capacity; Maine, which has a tax credit of 0.05 cents per
gallon; and Vermont, which provides loans to assist research and
planning for the production of biofuels. To our knowledge, no studies
have evaluated the importance of state government subsidies and tax
credits in ethanol plant location decisions. While it is clear that
state officials are using incentives as tools to compete for ethanol
plants, the effectiveness of such policies remains open to debate. There
is, however, a substantial literature evaluating the effectiveness of
general economic development incentives, and we turn to this research
next.
Subnational Incentives and Economic Development
Given that this literature is now very extensive, it is not within
the scope of this article to review this entire body of work. Rather,
our goal is to summarize some of the key findings of the existing
research in the context of the objectives of the present study.
Perhaps it is appropriate to begin with the text by Fisher (2007),
who provides an excellent summary of the existing research. Generally,
state and local governments use three basic types of fiscal incentives
to attract business: (i) financial incentives, including loans below
market-level interest rates, direct grants, and loan guarantees; (ii)
tax reductions, including the use of credits, deductions, abatements,
and specialized rates; and (iii) direct grants of goods or services,
including land, labor training, and infrastructure. According to Fisher,
"Most states offer all these incentives in one way or another,
developing a package of specific incentives from the general list for
each potential investment project" (Fisher, 2007, p. 647). The
primary goal behind these provisions is to offset real or perceived
business cost differences among states.
Regarding the implementation of subsidies and other fiscal
provisions, the evidence as to whether such policies are effective is
mixed. Citing many references, Edwards (2007) argues that policy
instruments and financial assistance payments are largely ineffective.
Rather, the more important driving forces in firm location decisions are
as follows: (i) proximity to the product market, (ii) the quality of
labor, and (iii) the quality of transportation networks. Edward
concludes that government incentives distort markets and lead to
inefficient outcomes.
There is, however, some work showing that incentives may influence
some firm location decisions. For example, Bartik (1991) shows that in
the long run, the elasticity of business activity with respect to state
and local taxes lies in the range of -0.1 to -0.6 for business location
decisions. Wasylenko (1997) furthers the argument by noting that
significant differences among states' incentives must exist in
order for an impact to be felt as a result of the incentives. He argues
that an impact substantial enough to be measured will occur only when a
state's policies are significantly different from those offered by
other states. Bartik reviewed 30 studies, of which 60% find
statistically significant positive effects on business activity. The
work of Bartik (1989), Garcia-Mila and McGuire (1992), and Tannenwald
(1996) suggests that the success of the incentives may depend on the
type of business the region is trying to attract. Finally, Gabe and
Kraybill (2002) show that establishments receiving incentives tend to
overstate announced employment targets.
Our work adds to both these lines of research. In terms of biofuel
research, little is known about the ways in which state incentives
influence location decisions of ethanol plants. With regard to the
general research evaluating the effectiveness of economic development
incentives, our work makes two contributions. First, we evaluate the
role of incentives in an emerging industry as opposed to already
existing industries. Second, as we discuss in greater detail in section
4, we address important endogeneity and time trend issues.
3. Empirical Analysis
Data
Our goal is to evaluate the role of state subsidies/tax credits in
the location of ethanol production. Unfortunately, ethanol production
data has only been systematically collected at the regional or national
level. These aggregated data prevent state-level policy analysis. To
overcome this data constraint, we collected detailed information on
ethanol plant capacity from the Annual Industry Outlook reports produced
by the Renewable Fuels Association (RFA), which collected information on
plant capacity, along with their locations, for the entire United States
for the years 2001-2007. We utilized information from all RFA annual
reports available, which allowed us to construct state-level capacity
levels from detailed plant capacity data for years 2001-2007. (5)
However, data on plant capacity for each state prior to 2001 were not
available in any single public or private document. Fortunately, the
2001 RFA annual report provided a list of all existing plants (with
information on location and capacity) as of 2001. We used this
information to conduct a search to identify the start dates of each
pre-2001 plant. The search was exhaustive and required contacting
numerous sources. For a number of plants, we were able to obtain start
dates from companies directly, but some cases required a more extensive
search through newspapers, contacting local governments, etc. (6) Figure
1 presents national actual production data collected by the U.S.
Department of Energy along with our production capacity data aggregated
to the national level for years 1980-2007. While the match is quite
good, note also that our capacity data systematically overstate actual
production. The use of production capacity as a proxy for actual
production is a potential concern, but through discussions with the
Energy Information Administration (EIA), we have learned that the vast
majority of ethanol plants in the country operate close to capacity and
that most changes in production are the result of increases in capacity.
While these claims are impossible to verify at the state level (due to
the lack of state-level production data), we are able to measure the
relationship between production and capacity over time at the national
level, and a very strong correlation (r = 0.9945) exists between
national ethanol production and capacity. Hence, we conclude that the
capacity data should accurately capture variation in production during
the sample time frame studied.
[FIGURE 1 OMITTED]
We matched the production capacity data to detailed information on
state-level ethanol subsidies and tax breaks, which were obtained from
the U.S. Department of Energy (7). However, a thorough search through
state statutes was required to identify specific dates of
implementation. (8) Our research focuses on 1980-2007, a period which
covers the complete emergence of the biofuel industry in the United
States. Importantly, this time period was characterized by the passage
of a multitude of state-level subsidies and tax breaks aimed at
increasing ethanol production and capacity. Hence, this panel provides
us with a substantial amount of within-state variation in policies from
which to identify an effect. Table 1 provides an overview of key ethanol
policies that were implemented for all states during the time period
that we study and some details about those policies for each state.
Specifically, one can see that there is a significant amount of
variation in the timing and location of policy adoption. For example,
per gallon tax credits range from zero in a number of states to nearly
50 cents a gallon in Wyoming by the end of 2007. Moreover, there is also
significant variation in the production capacity measures, both within
and across states. As an illustration, ethanol production was almost
nonexistent in the United States in 1980, but by 2007, two states (Iowa
and Nebraska) each had the ability to produce well over 1 billion
gallons of ethanol a year.
We also collected other variables at the state level that are used
at various points in the analysis as controls or as a basis for
restricting the sample. Specifically, measures of the most important
input costs that vary by state over time, state-level corn prices,
average hourly wages, natural gas prices, and gasoline prices were
collected from the U.S. Department of Agriculture (USDA), the Current
Population Survey (CPS), and U.S. Energy Information Administration
(EIA). Information on the level of corn production in each state was
also obtained from the USDA. Summary statistics, definitions, and data
sources of all nonpolicy data for all states as well as treatment and
control groups are provided in Table 2 and Appendix A, respectively.
Methodology
Several methods and approaches could be utilized to estimate the
impact of different ethanol policies on production capacity. We begin by
pooling the states passing ethanol-based policies (the treatment group)
and the remaining states in the United States (the control group). We
will, however, also utilize different empirical approaches with narrower
samples later to examine the sensitivity of our findings, which prove
robust. (9)
Our core analysis begins with the following fixed-effects
regression model:
[E.sub.st] = [[alpha].sub.s] + [[tau].sub.t] +
[[beta].sub.1][TC.sub.st] + [[beta].sub.2][PGC.sub.st] +
[[gamma]'[X.sub.st] + [[epsilon.sub.st], (1)
where subscript s denotes state, and subscript t denotes year. The
terms [[alpha].sub.s] and [[tau].sub.t] are the state and time fixed
effects. The inclusion of state and time fixed effects is imperative in
this context and helps alleviate one major concern of this
analysis--specifically, that differences in production across states
that are largely time-invariant, such as corn yield, and differences in
production across time that are common in all states, such as variation
generated due to changes in petroleum prices, economic shocks, or
changes in federal policies, will not bias estimates.
E is defined in ordinary least squares (OLS) estimates as the log
of the ethanol capacity in a given state-year. (10) The logarithmic
specification would seem the most appropriate measure of the dependent
variable because the median ethanol capacity for the state-years in the
sample is less than the mean. Estimation of Equation 1 will therefore
initially be by OLS, and standard errors are corrected to allow for
nonindependence of observations from the same state through clustering
(Arellano 1987; Pepper 2002; Bertrand, Duflo, and Mullainathan 2004).
(11)
Our specification contains two distinct ethanol-based policy
measures: TC, which is a dummy variable equal to one if a state has any
type of tax credit or subsidy for ethanol-producing firms (except per
gallon credit) in effect for a given year and equal to zero otherwise,
and PGC, which represents the amount of per gallon tax credit and/or
subsidy provided by a given state in a given year (in 2007 dollars).
(12) Thus, the estimates of [[beta].sub.1] can be interpreted as the
percent change in production after the passage of a fixed state tax
credit relative to a control group of states that did not experience a
change in the tax credit or subsidy status. Similarly, [[beta].sub.2]
reflects the percentage change in production given an increase in the
per gallon production tax credit/subsidy.
Even though the use of fixed effects deals with many of the
problems that may be caused by differences in state characteristics that
are unchanging over the sample period, there may exist some correlation
between policy passage and changes in other important factors. We deal
with this possibility by including state-specific changes in important
production factors in the X vector, which fluctuate between states and
over time with some regularity. (13) Specifically, we add as controls
the natural logarithm of corn prices, lagged corn production, gasoline
prices, hourly wages, and natural gas prices in a state. (14) The
within-state variations in the most important production factors across
years are captured by these variables.
Results
Primary Findings
To provide the reader with some insight into the time trends
present in the capacity data, with respect to our policy measures, we
have undertaken a simple before-and-after analysis. Figures 2 and 3
present the time-series patterns in mean production levels in states
that have adopted incentives and those that have not. Since production
incentives are imposed at different times, the horizontal axis is years
pre- or postadoption of the production incentive. For comparison, these
figures also report a simple average of ethanol production capacity
levels for all unaffected states for the relevant years. (15) In looking
at the figures, it is apparent that in both cases, the treatment states
produce more ethanol to begin with, but it is also true that these
states see large increases in production capacity following the passage
of incentives. A similar corresponding change is not apparent in the
nontreated states. This said, these figures do not control for the
potentially important influences of relevant covariates (or the
different incentives), nor do they contain the degree of detail or
sophistication that is necessary to draw any definitive conclusions.
Nevertheless, Figures 2 and 3 clearly provide a basis for hypothesizing
that these subsidies may affect production capacity over time and, as
such, warrant more rigorous empirical investigation.
[FIGURE 2 OMITTED]
To begin, we estimate Equation 1 by OLS for our primary sample,
which consists of all 50 U.S. states. Column 1 of Table 3 provides the
result using fixed effects with no other controls included, and the
regression shows the positive effects of both the production tax credit
and per gallon subsidies on state ethanol production. While the
production tax credit is too imprecisely estimated to conclude that the
coefficient is statistically different from zero, the magnitude of the
coefficient suggests that some effect may be present. On the other hand,
the coefficient on the per gallon tax credit variable is highly
significant and indicates that a $0.10 increase in a state's per
gallon tax credit increases production by approximately 40%. Overall,
these estimates indicate that per gallon incentives seem to play a very
important role in stimulating production capacity, whereas other tax
breaks, grants, or subsidized loans as characterized by the TC indicator
variable may influence production in important ways, but statistical
evidence prohibits strong inference about the degree or magnitude.
[FIGURE 3 OMITTED]
In the second column, we add controls for natural gas and hourly
wages, which are both important input costs in the production process.
The results suggest that there is very little effect on state ethanol
production capacity as a result of changes in these covariates. This is
not entirely surprising since changes in input costs, such as corn
prices and natural gas prices, may effect short-run production
decisions, but they are not likely to affect production capacity, which
is much less variable in the short run. Regardless, the estimated effect
of the subsidies remains relatively unchanged with the inclusion of
these variables. Capturing changes in the corn market is also very
important, given the vital nature of this input to the ethanol
production process. To capture the impacts of variations in the corn
market over time, in columns 3 and 4, we add one-year-lagged state corn
production and state corn prices, respectively. In both cases, these
variables take on signs that seem reasonable, although only the lagged
corn production variable is statistically significant. (16)
Nevertheless, the addition of these potentially relevant covariates does
not qualitatively affect the estimates on the two policy variables in
question. (17)
Given the fixed-effects approach utilized here, there may be some
concern over the comparability of the treatment states and the control
states. Specifically, there may be some concern that the inclusion of
non-ethanol-producing states in the control group could bias our results
toward finding a positive impact of these policies. So to test the
sensitivity of our results to this concern, we restrict the analysis to
include only states that have positive ethanol production capacity for a
minimum of one year during the sample period. The results, presented in
columns 5 and 6 of Table 3, are nearly identical to those from the
expanded sample, which again suggests that per gallon tax credits play a
significant role in the location of ethanol plants. (18)
Test of Endogeneity
Research on the impact of policy changes on economic activity is
often hampered by endogeneity concerns, and this study is no exception.
In this case, it is possible that subsidies are more likely to be
introduced because ethanol producers who plan to locate a plant in a
particular state may also successfully lobby for subsidies. In this
sense, the timing of subsidy adoption may very well depend on plans for
new plant location. We examine the possible endogeneity of subsidies
directed at ethanol production more rigorously by undertaking an
instrumental variable approach, which requires that we identify at least
three variables that determine subsidy adoption but that do not directly
determine ethanol production capacity in the state. Importantly, given
that we are using a fixed-effects framework, we must use instruments
that vary over time. Based on these criteria, we identified five
instruments that prove effective for these purposes. Specifically, the
instruments are two indicator variables indicating Democrat and
Republican rule, a variable indicating the length that the National Corn
Growers Association (NCGA) has been present in a state, the dollar
amount of check-off revenue generated in each state-year, and a state
gubernatorial election year indicator variable. The Democratic rule
variable (DEM) is equal to one when the governor is a Democrat, and the
Democratic Party has majority control in both the Senate and House, and
equals zero otherwise. Republican rule (REP) is equal to one when the
governor is a Republican, and the Republican Party has majority control
in both the Senate and House, and equals zero otherwise. (19) We
hypothesize that political control by one party or another may influence
the propensity to which a state imposes ethanol subsidy legislation. The
NCGA variable indicates the number of years that a state has had an
affiliation with the corn growers association. This variable may reflect
the political "clout" or "sway" that corn farmers
may hold in their state, thus influencing the likelihood of the
introduction of incentives. The check-off revenue instrument (CKOFF)
requires a brief explanation. The check-off is something like a tax on
corn, but the revenues from the check-off are used to promote, market,
and lobby on behalf of corn growers. The size of the check-off ranges
from 0.25 cents per bushel to 1 cent per bushel. To obtain the revenues
generated from the check-off, we multiplied the per bushel check-off by
the total number of bushels produced annually in the state. We
hypothesize that the more revenue generated by the check-off, the
greater will be the influence of corn growers in the political process
of that state. We note that over our sample period, numerous states
became affiliated with the National Corn Growers Association and
introduced check-offs. (20) Lastly, the election year variable (ELECT)
indicates each year that a particular state held a gubernatorial
election during the sample period. State elections may lead to an
environment that is more conducive to new policy implementation.
To move forward with our instrumental variables (IV) approach, we
must first demonstrate that these variables are valid instruments.
Utilizing a two-stage least squares approach, the first stage estimates
the following model:
[POLICY.sub.st] = DEM [[mu]sub.1] + [REP.sub.st][[mu].sub.2] +
[NCGA.sub.st] [[mu].sub.3] + [CKOFE.sub.st] [[mu].sub.4] +
[ELECT.sub.st] [[mu].sub.5] + [V.sub.st][mu]' + [S.sub.s] [T.sub.t]
+ [[epsilon].sub.st] (2)
in state s in period t. POLICY represents the per gallon tax credit
in one first-stage regression and the tax credit dummy in the other.
[DEM.sub.st] is an n x 1 vector that indicates Democratic Party control
in the n state-years in our data set, [[mu].sub.1] measures the effect
of this measure on the probability of an ethanol subsidy being in place
in a particular state-year, and [REP.sub.st], [NCGA.sub.st],
[CKOFF.sub.st], and [ELECT.sub.st] are the comparable counterparts.
[V.sub.st] is an nxk set of control variables (k is the number is
controls), and [[mu.sub.3] is a k x 1 vector of parameters. [S.sub.s]
represents the statespecific effects, [T.sub.t] is the set of time
indicator variables, and [[epsilon].sub.st] is the residual. The results
of the two first-stage regressions are presented in Appendix B. In
column 1, estimates of [[mu].sub.4] and [[mu].sub.5] are positive and
statistically significant, with t-statistics equaling 2.02 and 2.01,
respectively, suggesting that states holding elections or with high
levels of check-off revenues are more likely to adopt production tax
credit/subsidy legislation. More importantly, the p-value for the test
of excluded instruments was equal to 0.0423, indicating that the
instruments are jointly significant. In the second first-stage
regression (column 2), only the REP instrument is individually
significant (t-statistic = 1.85), indicating that states coming under
Republican control are more likely to adopt ethanol per gallon tax
credit/subsidies, and again the excluded instruments are jointly
significant (p-value = 0.0253).
Overall, this IV model successfully rejects the underidentification
test (p-value = 0.0674), which suggests that the excluded instruments
are relevant and correlated with the endogenous regressors. (21) While
the underidentification test examines the relevance and correlation of
the excluded instruments, it is also valuable to evaluate the strength
of this correlation. A test of the joint strength of the instruments
only provides marginal evidence of strength (Cragg-Donald F statistic =
5.603), indicating that the results may possibly suffer from issues
associated with weak instruments, such as tests for significance having
incorrect size and incorrectly estimated confidence intervals. (22)
Lastly, it is also important to conduct a Sargan-Hansen test of
overidentifying restrictions. (23) This examination tests the joint null
hypothesis that the instruments are valid, that is, uncorrelated with
the error term in the second stage. A rejection of this hypothesis would
suggest that one or more of the instruments are correlated with the
disturbance process and, hence, would cast doubt on the suitability of
the instrument set. Fortunately, we are unable to reject the null
hypotheses (p-value = 0.3747), indicating that the instruments are
jointly valid.
The second-stage results are presented in the last column of Table
3. These estimates, while qualitatively similar to the estimates that do
not correct for endogeneity, are substantially larger, and the
coefficient on the production tax credit/subsidy indicator variable
becomes significant. These results suggest that the OLS estimates are
not biased upward, and in fact suggest that impacts may be even more
substantial. (24) While the IV estimation does not provide evidence in
opposition to the OLS findings, the magnitude of the results is
dramatically larger, and the precision of the estimates is significantly
lower (standard errors are much larger). (25) This lost efficiency may
suggest that the OLS estimates are preferable when attempting to
prescribe the potential magnitude of the effect of these policies. As a
further examination, we use a Hausman test to evaluate OLS versus IV
estimators, where the null hypothesis states that the OLS estimator is
consistent and fully efficient. In our case, we reject the null
hypothesis at the 10% level (p-value = 0.0637), suggesting that there is
some endogeneity present. Nevertheless, due to the potential weakness of
the instruments, the statistical significance and marginal effects
estimated by the IV approach may be too large, and so we tend to favor
the more conservative and precisely measured OLS estimates.
Intertemporal Analysis
Earlier we presented evidence in Figures 2 and 3 that suggested a
positive effect of ethanol subsidies/credits on production over time. In
Table 4, we present an intertemporal analysis of the effect of ethanol
subsidies/credits over time by introducing 2 year lead effects and 2
year lagged effects, as well as a contemporaneous effect of each policy.
(26) These estimates, presented in Table 4, include all control
variables used in the third and fourth columns of Table 3.
The lead effects are informative because we can determine whether
the positive effects from Table 3 do indeed stem from the per gallon
credit or if they are the result of a preexisting trend. In both columns
of Table 4, the two credit lead effects are insignificant and generally
negative for both policies measures, suggesting that the estimates in
Table 3 are not the result of trending differences between the treatment
and control states. Turning our attention to the contemporaneous
measures and the lagged measures of the production tax credit dummy
variable, due to a lack of precision, we see no statistically
significant coefficients, although the pattern of results is consistent
with those observed in Figure 3. For the per gallon tax credit, the
results are also consistent with Figure 2 in showing that it may take a
couple of years before significant and noticeable increases in
production capacity are seen. We find that the pattern presented in
Table 4 is consistent with intuition about the ways in which businesses
may react to new incentives. Namely, businesses respond to new
incentives and increase production, but it may take some time before new
plants can be made operational.
Overall, our analysis reveals a consistent positive relationship
between the adoption of per gallon production subsidies and changes in
ethanol production capacity. These findings prove robust to alternative
samples, the inclusion (or exclusion) of important control variables,
appropriate alternative estimation techniques, and the presence of
preexisting trends.
4. Conclusion
In this study, we present evidence that state-level per gallon tax
credits do indeed influence ethanol plant location patterns. Existing
research on the emerging ethanol industry suggests that the social
benefits of ethanol production and consumption are minimal, perhaps even
negative. According the analysis by Gardner (2007), it is likely that
federal subsidies and mandates for ethanol actually generate a
substantial long-run deadweight loss in the range of $3.5 to $4 billion
annually. Taking into account state subsidies only increases the
magnitude of this deadweight loss measure. From the point of view of a
particular state, however, state policymakers should consider offering
certain subsidies if the objective is to be a leader in the emerging
biofuel industry. In the context of the very recent and growing concern
about the tradeoff between food and fuel, the broader implications of
increasing subsidies for biofuel production, at least in the context of
existing corn-based technologies, (27) should be carefully considered.
While it is difficult to directly measure the importance of federal
subsidies in the emergence of the ethanol industry in the United States,
our work is consistent with the notion that federal subsidies are very
important. We cannot, however, disentangle the difference between the
effect of state subsidies on the growth of national production capacity
and the location of ethanol production. It seems likely that the
underlying driving forces in national production are the federal
mandates and subsidies, whereas state subsidies primarily influence
location patterns. On the margin, however, state subsidies may have, to
some degree, also increased total national ethanol production capacity.
More generally, our work contributes to the literature that has
sought to evaluate the effectiveness of economic development incentives
by examining the issue in the context of an emerging industry. Over the
period 1980-2007, we observed the complete emergence of the ethanol
industry, and we observed numerous state-level policy changes aimed at
subsidizing the industry. Our work demonstrates that incentives appear
to have played an important role in determining the location of ethanol
plants, although a general investigation of the data and the location of
policy activity suggest that subsidies need to be substantial in order
for them to be effective in states without strong potential to produce
ethanol. For example, Oklahoma and Montana each have $0.40 (in nominal
values) per gallon subsidies, but it seems unlikely that either state
will ever be significant producers of ethanol, at least under corn-based
ethanol technologies. This is because the conditions are not amenable to
producing corn (too dry or too cold), and the costs of transporting corn
are prohibitive. Nevertheless, it appears that subsidies are important
in terms of attracting new plants in states where there is the potential
for corn production.
Appendix A
Definitions and Sources of Variables
Variables Definitions Source
Hourly wage Average hourly wage for all CPS
nonsalary workers in each
state-year collected from the
CPS-ORG.
Natural gas price Price of natural gas sold to DOE, EIA
commercial consumers in each
state-year in dollars per thousand
cubic feet
Per gallon credit Per gallon tax credit and/or direct DOE, C&S
subsidy in real 2007 dollars. A
full listing of existing subsidies
is available at http://www/afdc/
energy.gov/afdc/
incentives-laws.html.
Production tax Indicator variable equal to one if DOE, C&S
credit/subsidy the state has a production tax
credit or subsidy, and zero
otherwise. This variable includes
an array of subsidies including:
grants, loans, per plant subsidies,
production tax credits, and
infrastructure tax credits. A full
listing of existing subsidies is
available at http://www.afdc.
energy.gov/afdc/incentives
laws.html.
State corn price Average corn price per bushel in USDA
each state-year
Ethanol production Ethanol production capacity in each RFA, C&S
capacity state-year
Corn production State corn production in millions USDA
of bushels per year
Sources: CPS (Current Population Survey: www.census.gov/cps)
(Accessed 17 April 2008), USDA (U.S. Department of Agriculture:
www.usda.gov), DOE (U.S. Department of Energy: http://www.doe.gov).
EIA (Energy Information Administration [part of DOE]:
http://www.eia.doe.gov), REA (Renewable Fuels Association:
www.ethanolrfa.org), C&S (Chad Cotti and Mark Skidmore: compiled
by the authors in this study).
Appendix B
First Stage IV Results from Table 3
Production Tax Per Gallon
Credit/Subsidy (1) Credit (2)
Republican legislative
control -0.0055 (0.0283) 0.0214 * (0.0115)
Democrat legislative
control 0.0455 (0.0385) 0.0055 (0.0058)
Gubernatorial election year 0.0100 ** (0.0050) -0.0010 (0.0027)
NCGA years -0.0014 (0.0046) 0.0017 (0.0014)
Check-off revenue (in
millions of 2007 dollars) 0.4580 ** (0.2270) 0.0002 (0.0554)
Log state natural gas price -0.0970 (0.0698) -0.0104 (0.0242)
Log state hourly wage 0.0205 (0.2691) 0.0644 (0.0667)
p-value: test of excluded
instruments 0.0423 0.0253
Sample size
(number of states) 1350 (50) 1350 (50)
Each column is from a separate regression that includes both state
and year fixed effects. The standard errors in parentheses are
corrected to allow for nonindependence of observations within a
state through clustering. ** and denote statistical significance
at the 0.05 and 0.10 levels, respectively. Prices are in 2007
dollars.
References
Arellano, M. 1987. Computing robust standard errors for
within-groups estimators. Oxford Bulletin of Economics and Statistics
49(4):431-34.
Bartik, Timothy J. 1989. Small business start-ups in the United
States: Estimates of the effects of characteristics of states. Southern
Economic Journal 55(4): 1004-18.
Bartik, Timothy J. 1991. Who benefits from state and local economic
development policies?. Kalamazoo, MI: W. E. Upjohn Institute for
Employment Research.
Bertrand, M., E. Duflo, and S. Mullainathan. 2004. How much should
we trust differences-in-differences estimates? Quarterly Journal of
Economics 119(1):249-75.
Buchanan, Patrick. 1999. "A Family Farm Bill of Rights."
Accessed 20 September 2007. Available http://www.
buchanan.org/pa-99-0809-familyfarms.html.
Chi, K. 1994. State business incentives: Options for the future. In
State Trends and Forecasts. Lexington, KY: Council of State Governments.
Edwards, Mary E. 2007. Regional and urban economics and economic
development. Boca Raton, FL: Auerbach Publications.
Fisher, Ronald C. 2007. State & local public finance. 3rd
edition. Mason, OH: Thomson South-Western.
Gabe, Todd, and David S. Kraybill. 2002. The effect of state
economic development incentives on employment growth of establishments.
Journal of Regional Science 42(4):703-30.
Garcia-Mila, Teresa, and Therese McGuire. 1992. The contribution of
publicly provided inputs to state's economies. Regional Science and
Urban Economics 22:229-41.
Gardner, Bruce. 2007. Fuel ethanol subsidies and farm price
support. Journal of Agricultural & Food Industrial Organization
5(2):Article 4.
Greenberg, Elizabeth, and Richard Reeder. 1998. Who benefits from
business assistance programs? Results of the ERS rural manufacturing
survey. Agriculture Information Bulletin No. 736-04. Washington, DC:
Economic Research Services, U.S. Department of Agriculture.
Hahn, Robert, and Caroline Cecot. 2007. The benefits and costs of
ethanol. Journal of Regulator), Economics 35(3):275-95.
Hausman, J., B. H. Hall, and Z. Griliches. 1984. Econometric models
for count data with an application to the patents R&D relationship.
Econometrica 52:909-38.
Kleibergen, Frank, and Richard Paap. 2006. Generalized reduced rank
tests using the singular value decomposition. Journal of Econometrics
133:97-126.
Pepper, John. 2002. Robust inferences for method of moments
estimators from random clustered samples: An application using data from
the panel study of income dynamics. Economics Letters 75(3):341-45.
Renewable Fuels Association. 2007. "Annual Industry Outlook
(2002 2008)." Accessed 19 November 2007. Available
http://www.ethanolrfa.org/industry/outlook/.
Sargan, John. 1958. The estimation of economic relationships using
instrumental variables. Econometrica 26:393-415.
Tannenwald, Robert. 1996. Business tax climate: How should it be
measured and how important is it? State Tax Notes May: 1459-71.
U.S. Department of Energy. 2007. "Alternative Fuels Data
Center." Accessed August 2007. Available
http://www.eere.energy.gove/afde/progs/ind_fed_incentives.cgi.
Wasylenko, Michael. 1997. Taxation and economic development: The
state of the economic literature. New England Economic Review
March:37-52.
Wooldridge, Jeffery. 2006. Introductory econometrics: A modern
approach. 3rd edition. New York: Thomson.
Chad Cotti * and Mark Skidmore ([dagger])
* Department of Economics, University of Wisconsin Oshkosh, 800
Algoma Boulevard, Oshkosh, WI 54901, USA; E-mail cottic@uwosh.edu.
([dagger]) Department of Agriculture, Resource and Food Economics
and Department of Economics, 208 Agriculture Hall, Michigan State
University, East Lansing, MI 48824-1039; USA; E-mail mskidmor@msu.edu.
Financial support for this research was provided by the U.S.
Department of Agriculture/CSREES administered through the Michigan
Agriculture Experiment Station.
Received July 2008; accepted March 2009.
(1) See the U.S. Department of Energy (2007) for a summary of
federal incentives encouraging alternative fuel production and use.
(2) It is important to note that a 54 cent per gallon tariff on
imported ethanol protects U.S. producers from international competition.
(3) For example, even conservative political commentator Patrick
Buchanan (1999, p. 1) stated that "... just as 1 support the
independence of the family farm, I support a policy of U.S. energy
independence that includes a strong stand for ethanol. This industry
creates 40,000 jobs, adds $12 billion in net farm income each
year...."
(4) A "corn state" is defined by Hahn and Cecot (2007) as
a state that grows greater than 1 million bushels of corn per year.
(5) These data were collected directly from the Renewable Fuels
Association's (2007) website (http://www.ethanolrfa.org/).
(6) There are a few small ethanol production plants for which we
were not able to find start dates. We also note that we were able to
identify several plant closings, and we have incorporated these closings
into our data. These are as follows: Florida--4 million gallons per year
(mgy) plant closed in 2002; Idaho--3 mgy plant closed in 2003; Idaho--3
mgy plant closed in 2004: Louisiana--35 mgy plant closed in 1990:
Louisiana a plant opened and then closed in the same year (1987);
Washington--a small plant (actual production capacity unknown) closed in
2004. Unfortunately, no source exists from which we can determine
whether we have omitted any other significant plant closings.
(7) http://www.eere.energy.gov/afdc/progs/all_state_summary.php/afdc/0.
(8) Several states also offer subsidies for the consumption (as
opposed to production) of ethanol. For example, Illinois exempts
gasoline blended with ethanol from sales taxation. While consumption
subsidies may increase the demand for ethanol, they are not location
specific, and so it is difficult to identify their overall impacts in
the state-level analysis we employ.
(9) Specifically, we will impose the restriction that there was at
least some capacity to produce ethanol in the state during the sample
time frame. It seems appropriate to test the robustness of the result
under a sample restricted to these ethanol-producing states, since they
may provide a useful alternative sample for testing the effect of
ethanol subsidies on ethanol production.
(10) Because there are many cases in which a 0 is recorded for
ethanol production capacity, we add 1 to all observations to avoid
arithmetic error when taking logarithms.
(11) The nonzero and mostly integer nature of our dependent
variable, in addition to significant overdispersion, would also suggest
that a negative binomial regression (Hausman, Hall, and Griliches 1984)
could be utilized. For these reasons, we will also test the robustness
of the OLS estimates to this alternative econometric approach.
(12) Other subsidies such as low interest loans, grants, etc., are
included as part of the tax credit dummy variable. It should be pointed
out that these make up a very small proportion of the total amount of
subsidies/credits that fall under this variable.
(13) Due to the presence of fixed effects, it is only necessary to
include relevant variables if they fluctuate differently between states,
and even then, it would only affect our outcomes if the fluctuations
were correlated with being in the treatment group or control group.
(14) All dollar values were adjusted for inflation into 2007
dollars.
(15) For example, period t-1 in Figure 2 corresponds to 1981, 1982,
1985, 1995, 1997, 2000, 2000, 2001, 2001, 2001, 2003, 2003, 2003, 2003,
2004, 2004, 2005, and 2006 in the per gallon credit states. Thus, the
value constructed for the non-per-gallon-credit states is the average
ethanol production capacity in those years. It should be noted that some
years appear more than once, and therefore carry a heavier weight than
other years because there is more than one relevant observation in that
year.
(16) The number of states analyzed falls when the corn price
variable is included because nine states do not produce any corn, and
hence, no price data are available.
(17) The results from columns 3 and 4 were duplicated with the
addition of a state price of gasoline variable, but this inclusion had
little impact on the earlier result and so was left out of Table 3.
These estimates are available upon request.
(18) As mentioned earlier, due to the non-negative and mostly
integer nature of the dependent variable, we could also have utilized a
negative binomial regression approach. To test the robustness of our
results to the OLS approach, all estimates reported in columns 1-7 of
Table 3 were duplicated using negative binomial regression. These
results were qualitatively very similar across all variables, although
the magnitudes of the effects were in some cases larger. Details of
these regressions are available from the authors upon request.
(19) The omitted category is state-years in which neither the
Democrats nor the Republicans have full control.
(20) We list the states and years they joined the NCGA and
introduced check-offs, respectively: Arkansas--1997, 1998;
Colorado--1979, 1987; Georgia--1984, 1996; Illinois--1971, 1982;
Indiana--1971, 2007; Iowa--1967, 1977: Kansas--1975, 1977;
Kentucky--1982, 1990; Louisiana--1985, 1985; Maryland--1977, 1991;
Michigan--1973, 1993; Minnesota--1978, 1990; Mississippi--1993, 2006;
Missouri--1978, 1984; Nebraska--1973, 1978; New York--1988, no
check-off; North Carolina 1978, 1979; North Dakota--1987, 1991;
Ohio--1977, 1989; Oklahoma 1996, no check-off; Pennsylvania--1973, no
check-off; South Carolina--1991, no check-off; South Dakota--1986, 1988;
Tennessee--1986, no check-off; Texas--1989, 1990; Virginia, 1979, 1980;
Wisconsin--1975, 1982.
(21) The test is essentially the test of the rank of a matrix:
under the null hypothesis that the equation is underidentified. A
rejection of the null indicates that the matrix is full-column rank;
that is, the model is identified. See Kleibergen and Paap (2006) for
details.
(22) While a Cragg-Donald F statistic of 5.603 is not trivial, it
is also not large enough to conclude with any certainty that the
excluded instruments will not suffer from some of the potentially
problematic issues associated with weak instruments.
(23) See Sargan (1958) for details.
(24) Given the growth patterns presented in Figures 2 and 3, the
large estimated coefficients on both subsidy variables generated from
the IV estimation procedure may not be entirely implausible.
(25) This outcome is not unusual, as is pointed out by Wooldridge,
who states "[there is an] important cost to performing IV
estimation ... the asymptotic variance of the IV estimator is always
larger, and sometimes much larger, than the asymptotic variance of the
OLS estimator" (Wooldridge 2006, p. 516: emphasis added).
(26) For the intertemporal analysis, a dummy variable was used to
represent the per gallon credit variable rather than the actual per
gallon credit itself (as in Table 3). A dummy variable approach is often
used to allow for easy interpretation of the lagged effects.
(27) It is possible to produce ethanol in states where climatic and
soil conditions are suboptimal. It is, however, muchmore expensive to
produce corn in these states. Scientists are working on methods of
producing ethanol from cellulosic materials such as wood or switch
grass. Under current technologies, it is much more expensive to produce
ethanol from such materials. It is conceivable that at some point a
breakthrough will enable ethanol producers to use a wider range of
inputs in the production process, and this may change and/or expand the
location of ethanol production nationwide. Currently, corn is the least
costly approach given existing technological capabilities.
Table 1. State Ethanol Production and Ethanol Subsidy Policies,
1980-2007
Tax Credit/ Per Gallon Tax
State Subsidy Grants/Loans Credit/Subsidy
Alabama -- -- --
Alaska -- -- --
Arizona -- -- --
Arkansas -- -- 2007
California -- 2006-2007 --
Colorado -- -- --
Connecticut -- -- --
Delaware -- -- --
Florida 2006-2007 -- --
Georgia -- -- --
Hawaii 2002-2007 -- --
Idaho -- -- --
Illinois -- 2003-2007 --
Indiana -- -- 1982-1986,
2004-2007
Iowa 2001-2007 1994-2007 --
1996-2007
Kansas -- -- 2001-2007
Kentucky -- -- --
Louisiana -- -- --
Maine -- 1999-2007 2004-2007
Maryland -- -- 2006-2007
Massachusetts -- -- --
Michigan 2003-2007 -- --
Minnesota -- -- 1986-2007
Mississippi -- -- 2002-2007
Missouri -- -- 2002-2007
Montana -- -- 1983-2007
Nebraska 1990-1999 -- 2000-2007
Nevada -- -- --
New Hampshire -- -- --
New Jersey -- -- --
New Mexico -- -- --
New York -- -- --
North Carolina 2000-2007 -- --
North Dakota (a) -- 2007 2005-2007
Ohio -- -- --
Oklahoma -- -- 2004-2007
Oregon 2006-2007 -- --
Pennsylvania -- 2006-2007 2005-2007
Rhode Island -- -- --
South Carolina -- -- 2007
South Dakota -- -- 1996-2007
Tennessee -- --
Texas -- -- 2004-2007
Utah -- --
Vermont -- -- --
Virginia -- -- 2007
Washington 2003-2007 -- --
West Virginia -- -- --
Wisconsin -- -- 2001-2006
Wyoming -- -- 1998-2007
Maximum Per
Gallon Tax Credit Ethanol Production
Offered (in 2007 Capacity, 2007 (in
State Dollars) Millions of Gallons)
Alabama -- 0
Alaska -- 0
Arizona -- 55
Arkansas 0.200 0
California -- 69
Colorado -- 125
Connecticut -- 0
Delaware -- 0
Florida -- 0
Georgia -- 0.4
Hawaii -- 0
Idaho -- 4
Illinois -- 813
Indiana 0.286 392
Iowa -- 1976.5
Kansas 0.088 398.5
Kentucky -- 37
Louisiana -- 0
Maine 0.055 0
Maryland 0.051 0
Massachusetts -- 0
Michigan -- 262
Minnesota 0.336 684.6
Mississippi 0.230 0
Missouri 0.230 195
Montana 0.367 0
Nebraska 0.207 1313.5
Nevada -- 0
New Hampshire -- 0
New Jersey -- 0
New Mexico -- 30
New York -- 0
North Carolina -- 0
North Dakota (a) 0.424 140.5
Ohio -- 177
Oklahoma 0.437 2
Oregon -- 148
Pennsylvania 0.053 0
Rhode Island -- 0
South Carolina 0.400 0
South Dakota 0.255 657
Tennessee -- 67
Texas 0.219 115
Utah -- 0
Vermont -- 0
Virginia 0.100 0
Washington -- 0
West Virginia -- 0
Wisconsin 0.234 458
Wyoming 0.496 10.7
(a) North Dakota also has a per plant credit.
Table 2. Annual State Means for Nonpolicy Variables in the Analysis
States States
with without
Ethanol Ethanol
All States Subsidies Subsidies
Ethanol production capacity (a) 42.76 72.71 4.64
Natural gas price (b) 8.69 8.68 8.72
Hourly wage (c) 12.81 12.59 13.09
Corn price per bushel (c) 3.78 3.70 3.91
Corn production (d) 173,456 275,784 43,220
Number of observations (states) 1400 (50) 784 (28) 616 (22)
(a) In millions of gallons.
(b) Per million British Thermal Units (BTUs), in 2007 dollars.
(c) In 2007 dollars.
(d) In thousands of bushels.
Table 3. Effects of Ethanol Subsidies on Ethanol Production Capacity
OLS
1 2
Production tax credit/subsidy 0.517 0.5180
-0.4151 (0.4168)
Per gallon credit 4.0182 ** 4.0189 **
-1.4365 (1.4301)
Log state natural gas price -- 0.0372
(0.4687)
Log state hourly wage -- -0.0447
(1.3320)
Lagged corn production -- --
(in millions of bushels)
Log state corn price -- --
Adjusted [R.sup.2] 0.7873 0.7870
Sample size (number of states) 1400 1400
(50) (50)
Sample period 1980-2007 1980-2007
OLS
3 4
Production tax credit/subsidy 0.2920 0.6709
(0.3746) (0.4696)
Per gallon credit 3.6928 ** 3.6281 **
(1.2658) (1.3774)
Log state natural gas price -0.4769 -0.2815
(0.4289) (0.5849)
Log state hourly wage 0.2283 0.8816
(1.1519) (2.1008)
Lagged corn production 0.0003 ** --
(in millions of bushels) (0.0001)
Log state corn price -- -0.0704
(0.6754)
Adjusted [R.sup.2] 0.8180 0.7819
Sample size (number of states) 1350 1148
(50) (4l)
Sample period 1981-2007 1980-2007
OLS
5 6
Production tax credit/subsidy 0.3297 0.5732
(0.5361) (0.5409)
Per gallon credit 3.3428 ** 3.4672 **
(1.3980) (1.5202)
Log state natural gas price -1.2218 -0.5966
(0.7406) (0.8009)
Log state hourly wage 4.1148 4.4094
(3.0938) (3.4747)
Lagged corn production 0.0002 * --
(in millions of bushels) (0.0001)
Log state corn price -- -0.2127
(1.3471)
Adjusted [R.sup.2] 0.7901 0.7691
Sample size (number of states) 729 756
(27) (27)
Sample period 1981-2007 1980-2007
IV
7
Production tax credit/subsidy 6.6930 *
(3.6316)
Per gallon credit 18.0111 *
(9.9275)
Log state natural gas price 0.4603
(0.7695)
Log state hourly wage -0.3271
(2.2386)
Lagged corn production --
(in millions of bushels)
Log state corn price --
Adjusted [R.sup.2] --
Sample size (number of states) 1350
(50)
Sample period 1981-2007
Each column is from a separate regression that includes both state
and year fixed effects. The dependent variable is the log of
ethanol production capacity +1 in a state-year. The standard errors
in parentheses are corrected to allow for nonindependence of
observations within a state through clustering. ** and * denote
statistical significance at the 0.05 and 0.10 levels, respectively.
Samples sizes in columns 1-3 and 7 incorporate all fifty U.S.
states. Column 4 results are estimated from only 41 states, since
corn price data are only from the states that produce corn.
Estimates in columns 5 and 6 utilize a sample of 27 states that
have positive ethanol production capacity for a minimum of 1 year
during the sample period. Prices are in 2007 dollars.
Table 4. Intertemporal Analysis
Coefficient Coefficient
Estimates Estimates
Lead and Lagged Policy
Variables 1 2
Per gallon credit lead 2 0.0742 (0.2379) -0.1214 (0.3246)
Per gallon credit lead 1 -0.0120 (0.2921) -0.1175 (0.3437)
Per gallon credit
contemporaneous period -0.0998 (0.3219) -0.2248 (0.3507)
Per gallon credit lag 1 0.2921 (0.4978) 0.2223 (0.4361)
Per gallon credit lag 2+ 1.2743 ** (0.4447) 1.1678 ** (0.4635)
Production tax
credit/subsidy lead 2 -0.2315 (0.2232) -0.3440 (0.2961)
Production tax
credit/subsidy lead 1 0.0765 (0.3775) 0.0907 (0.4227)
Production tax credit/
subsidy contemporaneous
period 0.3125 (0.3999) 0.1010 (0.4447)
Production tax
credit/subsidy lag 1 0.2357 (0.5739) 0.3611 (0.6676)
Production tax
credit/subsidy lag 2+ 0.0036 (0.4026) 0.3280 (0.4942)
Log state natural gas
price -0.5894 (0.4445) -0.4448 (0.5899)
Log state hourly wage 0.1638 (1.1315) 0.4777 (2.0232)
Lagged corn production
(in millions of bushels) 0.0003 ** (0.0001) --
Log state corn price -- -0.1892 (0.6813)
Adjusted [R.sup.2] 0.8227 0.7839
Sample size
(number of states) 1350 (50) 1148 (41)
Each column is from a separate regression that includes both state
and year fixed effects. The dependent variable is the log of
ethanol production capacity +1 in a state-year. The standard errors
in parentheses are corrected to allow for nonindependence of
observations within a state through clustering. ** and * denote
statistical significance at the 0.05 and 0.10 levels, respectively.
Prices are in 2007 dollars.
