Development of Reduced-Form Models to Evaluate Macroeconomic Impacts of Greenhouse Gas Mitigation

Abstract

Since 2000, more than thirty-fi ve states have or are developing comprehensive plans to mitigate greenhouse gas (GHG) emissions and to achieve related public policy goals. Most State Climate Action Plans include detailed micro-level analyses of the mitigation policy options focusing
on the direct costs/savings associated with the implementation of the options. Estimation of the macroeconomic impacts of a policy on future employment and income typically requires the use of sophisticated modeling tools, whose application is often costly and time-consuming, and thus is often
prohibitive at an early phase of the policy evaluation process. In this paper, we develop reduced form statistical models that can be used to quickly and relatively inexpensively predict the likely macroeconomic impacts of various climate mitigation options. The regression models are built
based on the macroeconomic modeling results of 92 GHG mitigation policy options across four major states in the U.S.

Full text

J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015 377
*Corresponding author: adamzros@price.usc.edu
DOI: 10.7569/JSEE.2015.629503
Development of Reduced-Form Models to
Evaluate Macroeconomic Impacts of Greenhouse
Gas Mitigation
Dan Wei1, Noah Dormady2, and Adam Rose*,3
1Research Assistant Professor, Sol Price School of Public Policy, University of Southern California
(USC), Los Angeles, CA.
2Assistant Professor, John Glenn College of Public Affairs, Ohio State University, Columbus, OH.
3Research Professor, Price School, USC, Los Angeles, CA.
Received March 14, 2015; Accepted March 24, 2015
Abstract: Since 2000, more than thirty-fi ve states have or are developing
comprehensive plans to mitigate greenhouse gas (GHG) emissions and to
achieve related public policy goals. Most State Climate Action Plans include
detailed micro-level analyses of the mitigation policy options focusing on
the direct costs/savings associated with the implementation of the options.
Estimation of the macroeconomic impacts of a policy on future employment
and income typically requires the use of sophisticated modeling tools,
whose application is often costly and time-consuming, and thus is often
prohibitive at an early phase of the policy evaluation process. In this paper,
we develop reduced form statistical models that can be used to quickly and
relatively inexpensively predict the likely macroeconomic impacts of various
climate mitigation options. The regression models are built based on the
macroeconomic modeling results of 92 GHG mitigation policy options
across four major states in the U.S.
Keywords: Climate action plans, GHG mitigation options, macroeconomic impacts,
reduced-form model
1 Introduction
Given the lack of signifi cant progress in comprehensive climate policy formation
at the federal level, major climate initiatives have been undertaken at sub-national
levels of government in the U.S. in the past decade. Since 2000, more than thirty-
fi ve states and several hundred municipalities have or are developing compre-
hensive plans to mitigate greenhouse gas (GHG) emissions and to achieve related

Dan Wei et al.: Macroeconomics of GHG Mitigation
378 J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
public policy goals, such as health, energy, and economic improvement. Although
the process used in formulating climate action plans (CAPs) varied to some extent
from state to state, most engaged in a comprehensive, multi-objective, stakeholder-
based planning process that developed and analyzed a range of sector-specifi c
policy actions and mechanisms.
The analysis and evaluation of the GHG mitigation and sequestration policy
options in the state CAPs are usually grouped into four broad sectoral categories:
1) Energy Supply (ES), which focuses on fossil fuel extraction, processing and
transportation, and electricity generation, transmission, and distribution (major
options include renewable portfolio standards, combined heat and power, power
plant effi ciency improvements); 2) Residential, Commercial, and Industrial
(RCI), which focuses on emissions from stationary sources such as industrial
processes and fuel and electricity use in residential and commercial buildings
(major options include demand-side management, building codes, appliance
standards, customer-sited renewable energy); 3) Transportation and Land Use
(TLU), which focuses on mobile sources of GHG emissions and related drivers
associated with land use (e.g., alternative fuel strategies, alternative vehicles,
transit, land use); 4) Agriculture, Forestry and Waste Management (AFW) that
examines emissions and carbon sequestration opportunities in AFW manage-
ment (e.g., soil carbon management, on-farm energy effi ciency and renewable,
afforestation/reforestation).1
In most of the state CAPs, microeconomic impacts of individual policy
options or bundles, such as GHG reduction potentials and direct net costs, are
quantifi ed in a Delphi-type process (expert elicitation) by a group of experts
comprising a Technical Working Group (TWG) for each of the four sectoral
groupings associated with mitigation as mentioned above. The analysis of the
options are based on the policy-specifi c defi nition of the baseline, the techni-
cal policy design in terms of level of effort, timing, coverage of parties, and
the choices of data sources, methods, key assumptions, and uncertainty tech-
niques throughout the planning process that is led by the TWGs with high level
engagement of stakeholders [1–4].
However, all of the micro-level analyses of these policy options focus on the
direct (on-site) costs or savings associated with the implementation of the options.
When policymakers and stakeholders consider the impacts of potential options
to mitigate GHG emissions or sequester carbon, a major question often asked is:
“how will these options affect the local, state, or national economy?” Calculation
of the microeconomic (direct) costs or cost savings of policy options is a generally
1 Other than the above widely adopted options, some were less commonly recommended but were
carefully evaluated in some states depending on the economic structure, energy production and con-
sumption mix, and other special features and policy priorities. Examples include carbon capture and
sequestration, power distribution system upgrades, industrial process incentives, active transportation
programs that encourage bike/walk trips, and others.

Dan Wei et al.: Macroeconomics of GHG Mitigation
J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015 379
DOI: 10.7569/JSEE.2015.629503
straightforward application of accounting and cost-engineering. However, the
analysis of the macroeconomic impacts of a policy—the effects of the policy on
future employment and income, for example—typically requires the application
of sophisticated modeling tools. Moreover, the cost and time involved in perform-
ing a full macroeconomic study is often prohibitive at an earlier phase of the policy
evaluation process. The objective of this study is to develop reduced form statisti-
cal models that can be used to quickly and relatively inexpensively predict the
likely macroeconomic impacts of various climate mitigation options. To the extent
that most of these options are related to energy, the models can also be used to
evaluate some major aspects of energy policy.
The models we have developed are based on multivariate analyses of the rela-
tionships between macroeconomic impacts and various microeconomic costs,
structural linkages within the economy, and the characteristics of the mitigation
options evaluated. We accomplish this by utilizing the results of the application
of a macroeconomic modeling approach that has been widely used to analyze the
broader economic impacts of CAPs. This is the Regional Economic Models, Inc.
Policy Insight Plus (REMI PI+) Model [5]. We regress the results of the applica-
tions of the model in 4 major US states against several explanatory variables. This
is not the typical application of regression analysis to explain variations in random
processes. Rather, it is more of a curve-fi tting approach to develop an expedient
model that can provide rapid and inexpensive estimates of the macroeconomic
impacts of various GHG mitigation options, in contrast to having to run a com-
plicated and expensive model. The model will yield crude estimates of macroeco-
nomic impacts, but does provide insight into their margin of error. The “reduced
form” model is intended for use at the early scoping, or screening stages of the
CAP process to identify individual mitigation options for further study with more
advanced models. Note that the analysis is not intended as an assessment of the
REMI model results themselves, or a vindication of the REMI model in general
but simply to develop a quick turn-around and inexpensive tool to facilitate the
Climate Action Planning process.
The REMI Model is well-documented [6–8] and widely used at the state and
local level for policy analysis [9]. The reduced form modeling approach is well-
documented for understanding, testing, and simplifying the results of large-scale
economic models, including REMI [10–13].The paper is structured as follows.
Section 2 introduces the basic data we use in the regression analyses. The devel-
opment and summary results of regression models for GDP and employment
impacts are described in Sections 3 and 4, respectively. Section 5 briefl y summa-
rizes the strengths and weaknesses of these regression models, describes how they
might be applied to results of the evaluation of direct costs of mitigation options
to prepare estimates of the macroeconomic impacts of those options, and identi-
fi es key “next steps” in the development of these “reduced form” macroeconomic
modeling tools.

Dan Wei et al.: Macroeconomics of GHG Mitigation
380 J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
2 Basic Data
The basic data utilized for the regression analyses are taken from a set of macro-
economic analyses undertaken by the authors in conjunction with the Center for
Climate Strategies2 for the states of Florida, Pennsylvania, Michigan, and New
York. These state-based analyses evaluated the macroeconomic impacts of a com-
prehensive set of GHG emission mitigation options, the critical features of which
were specifi ed in each respective state’s CAP [14–17]. Appendix A presents the list
of major GHG mitigation and sequestration policy options that are recommended
in the CAPs in the four states. The variables analyzed by the regression tool spec-
ifi ed below are the estimated microeconomic and macroeconomic impacts of a
pooled cross-section of mitigation options. The mitigation options were identifi ed
and the microeconomic impacts were analyzed by sets of sector-specifi c technical
working groups (TWGs) in each state, with each group comprised of a broad set
of stakeholders [1,15,18,19].The dependent variables to be explained by the statis-
tical regression analyses are the Net Present Value (NPV) of Gross State Product
(GSP) impacts (in million 2005$) and employment impacts (in thousand person-
years) of each individual mitigation option. Estimates of these impacts are derived
from the results generated by the REMI PI+ macroeconometric model [5,7]. These
results in turn are shaped by the values and interactions of many independent
variables, the most relevant of which are carried over in the reduced form model
[12]. Given the diversity of the four states from which modeling results were
taken, there is also a great deal of variation in the macroeconomic impacts across
the states. For this reason, the data analyzed here are “noisy,” and some adjust-
ments must be made in order for the analysis to attain the required inferential
asymptotic qualities (i.e., to be able to provide mathematically reliable results).
The planning horizon used for Florida and Michigan was 17 years (from 2009 to
2025), for New York 20 years (from 2011 to 2030) and for Pennsylvania 12 years
(from 2009 to 2020). Given the differences in planning horizons, and nonlinearities
presented in the macroeconomic impacts across years (e.g., some policy options
may have relatively more long-run benefi ts), in the regression model for GSP
impacts, our dependent variable considers GSP impacts on an annualized basis;
i.e., the NPV of GSP impacts across a planning horizon is divided by the number
of years of its planning horizon. In the regression model for employment impacts,
the annualized employment impact is used. We fi rst compute the total employ-
ment impact in terms of person-years of a policy option as the simple sum of each
year’s employment impacts over the planning horizon. The average employment
impact is then computed by dividing the total employment impact by the number
of years in each state’s planning horizon.
2 The Center for Climate Strategies (CCS) is a non-profi t organization headquartered in Washington,
DC. Since 2000, CCS has facilitated the comprehensive, multi-objective, stakeholder-based climate
action planning process for 22 states.

Dan Wei et al.: Macroeconomics of GHG Mitigation
J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015 381
DOI: 10.7569/JSEE.2015.629503
The two main explanatory variables are the NPV of the direct net cost (“DNC”)
of a GHG mitigation option over the entire planning horizon and the NPV of the
investment requirements (“INV”) over the same time period, which are obtained
from the microeconomic analyses of the individual policy options in the respective
state CAPs. Analogous to the dependent variable, the annualized direct net cost
and investment requirements are calculated by dividing the NPVs of the direct
net cost and investment requirements, respectively, by the number of years in the
planning horizon. For the direct net cost variable, a positive value indicates that
the option has been estimated to be cost incurring, and a negative value indicates
that the direct effect of the option will be cost saving.
The regression model also includes eight binary (“categorical”) variables to
help explain the option-specifi c characteristics. The variables ES, RCI, TLU, and
AFW indicate the sector in which the mitigation policy is implemented (Energy
Supply; Residential, Commercial and Industrial; Transportation and Land Use;
and Agriculture, Forestry and Waste Management Sectors, respectively). These
variables have a value of 1 when the policy option is applied to the respective sec-
tor, and zero when the option is applied to other sectors. These sectoral dummy
variables are also used in interaction terms in some regression models to assign
the direct costs (or net savings) and the investment requirements of each option to
the sector that implements the option. “Construction” (CONST) is a binary vari-
able that indicates whether or not the mitigation option involves a capital invest-
ment in construction (e.g., building a new power plant). “Manufacturing” (MFG)
is a binary variable that indicates that the option involves a capital investment in
equipment or appliance manufacturing. “Government Subsidy” (GS) is a binary
variable indicating whether or not the mitigation option receives state govern-
ment aid. And fi nally, “Consumption Reallocation” (CR) indicates that the mitiga-
tion option results in a shift in the composition of consumer expenditures, such as
reducing spending on electricity, gas, and other fuels, and increasing consumption
in energy-effi cient appliances and other consumption categories.
Table 1 provides the descriptive statistics of all of the independent variables
used in our regression models. Here statistics for interaction terms, such as “DNC*
TLU” or “INV*TLU”, describe the annualized NPV of the direct net cost (or invest-
ment requirement) of policy options in each sector. The references to Model 1
through 4 in Table 1 pertain to the different regression models discussed below.
3 Regression Model for GSP Impacts
The functional form of the regression model for the GSP impacts is given by equa-
tion 1. The fi rst four terms of the model are the interaction terms of sectoral binary
variables and the direct net cost of an option. These interaction terms describe
the direct net cost impacts of the options applied to different sectors on GSP. The
following four terms are the interaction terms of sectoral binary variables and
the investment requirement associated with an option. These interaction terms

Dan Wei et al.: Macroeconomics of GHG Mitigation
382 J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
Table 1 Descriptive Statistics
Mean Standard
Deviation
Minimum
Value
Maximum
Value
D.V.: Annual Gross State
Product Impact (y)
(in Models 1 and 2)
-23.30 194.39 -886.00 532.74
D.V.: Annual
Employment Impact
(y)(in Models 3 and 4)
2.20 4.81 -5.57 22.59
Direct Net Cost (DNC) 60.13 165.53 -279.12 1,075.39
Investment Requirement
(INV) 114.97 233.51 0.00 1,420.13
DNC ¥ ES -0.21 65.55 -528.23 259.59
DNC ¥ RCI -22.41 81.99 -488.34 79.46
DNC ¥ TLU -15.06 150.40 -886.00 532.74
DNC ¥ AFW 14.39 61.23 -30.39 423.38
INV ¥ ES 44.64 158.59 0.00 1,268.71
INV ¥ RCI 26.42 151.41 0.00 1,420.13
INV ¥ TLU 24.35 98.85 0.00 666.98
INV ¥ AFW 19.55 79.58 0.00 541.28
ES 0.17 0.38 0 1
RCI 0.24 0.43 0 1
TLU 0.24 0.43 0 1
AFW 0.35 0.48 0 1
CONST 0.38 0.49 0 1
MFG 0.57 0.50 0 1
GS 0.22 0.41 0 1
CR 0.35 0.48 0 1
describe the impact of investment requirement of the options coming from differ-
ent sectors on GSP. The next four terms describe sectoral impacts (we assume that
options from different sectors have inherent differences in addition to the direct
net cost and investment requirement impacts captured by the interaction terms)
of the policy option on GSP. The fi nal four terms describe the GSP impacts of the
option related to whether or not the option involves construction investment,
manufacturing investment, government subsidies, and consumption reallocation.

Dan Wei et al.: Macroeconomics of GHG Mitigation
J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015 383
DOI: 10.7569/JSEE.2015.629503
12 3 4
56 7 8
910 11 12 13 14 1516
*
*
y = DNC ES+ DNC RCI + DNC TLU+ DNC AFW +
INV ES+ INV RCI + INV TLU + INV AFW
ES + RCI + TLU AFW + CONST MFG + GS CR
bb b b
bb b b
bb b b b b b b e
∗∗ ∗
∗∗ ∗+
++++
(1)
where
y:Annualized NPV of the GSP impacts of a policy option
DNC:Annualized NPV of the direct net cost of a policy option
INV: Annualized NPV of investment requirement of a policy option
ES:Energy Supply policy option binary variable
RCI:Residential, Commercial, Industrial policy option binary variable
TLU:Transportation and Land Use policy option binary variable
AFW:Agriculture, Forestry, and Waste Management policy option
binary variable
CONST:Capital investment on building constructions, which has stimulus
impacts to the construction sector (binary variable)
MFG:Capital investment on equipment, which has stimulus impacts
to the machinery and equipment manufacturing sectors (binary
variable)
GS:Policy option that receives state government subsidy (assuming
government spending decreases by the same amount elsewhere)
(binary variable)
CR:Policy option that results in consumer consumption realloca-
tion and increased purchasing power of the consumers (binary
variable)
b1 to b16:Regression coeffi cients
e:Stochastic error term
Tables 2 and 3 provide the results of our multivariate statistical analysis. We ran
both a basic model (Model 1) and an interactive model (Model 2), which includes
interaction terms to evaluate the individual sectoral impacts of the direct net costs
and investment requirements associated with GHG mitigation policy options. The
functional form of the regression model, as specifi ed in equation 1, provides the
full interactive model.
In each model the intercept term is suppressed. This is warranted on theoreti-
cal grounds, due to the fact that in the absence of a policy change there would
be no incremental change in the GSP of a state economy. This also enables us to

Dan Wei et al.: Macroeconomics of GHG Mitigation
384 J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
explicitly display the effects of our four binary sectoral variables.3 Our analysis
also implicitly assumes that the extant economies, as described by the coeffi cients
and equations in the REMI models for each state, are in equilibrium. To account
for potential heteroskedasticity (a violation of one of the basic regression model-
ing assumptions, which requires that the modeling errors have a constant variance
across the observations), we used the White’s robust standard errors [20], which
provides a correction that penalizes the model for any heteroskedasticity [21].
Both models 1 and 2 have strong fi tness and summary statistics, as indicated
by a statistically signifi cant F-statistic and the R-squared values. These measures
indicate that the model fi ts the data relatively strongly and that more than half
(almost three fourths) of the error variance is explained by model 1 (and model 2).
Given that the sample size is relatively small compared to other large sample
analyses (N=92), the relatively strong fi tness measures indicate that the sample
is large enough to have predictive power and thus remain externally valid. We
also conducted tests to ensure that the explanatory variables were not correlated
(i.e., multicollinearity). These tests indicate that the only collinearity present in the
models that would infl ate the variance comes from sectoral indicator variables
3 Inclusion of the intercept would force us to exclude one sectoral category from the regression model
to use it as the reference sector for the other sectoral binary variables, and in such a case, the coeffi cients
of the sectoral binary variables included in the regression model need to be interpreted as the differen-
tial impact of the modeled sector with respect to the reference sector).
Table 2 Results of the Regression Analysis for GSP Impact – Model 1
Coeffi cient Robust Std. Error
Direct Net Cost (DNC) –0.51*** 0.15
Investment Requirement (INV) 0.31*** 0.08
ES –15.27 38.26
RCI –18.64 45.62
TLU –45.66 36.25
AFW 6.83 20.97
Construction Inv. (CONST) 40.91 30.38
Manufacturing Inv. (MFG) 25.13 25.34
Government Subsidy (GS) 21.59 34.99
Consumption Reallocation (CR) –17.49 38.01
N92
R-squared 0.52
F-Statistic 4.13***
Ordinary Least Squares (OLS) Regression with White’s Robust Standard Errors.
*** p<0.01, **p<0.05, *p<0.1.

Dan Wei et al.: Macroeconomics of GHG Mitigation
J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015 385
DOI: 10.7569/JSEE.2015.629503
Table 3 Results of the Regression Analysis for GSP Impact – Model 2
Coeffi cient Robust Std. Error
DNC ¥ ES –1.35*** 0.25
DNC ¥ RCI –0.42** 0.20
DNC ¥ TLU –0.32*** 0.07
DNC ¥ AFW –0.56* 0.29
INV ¥ ES 0.57*** 0.09
INV ¥ RCI 0.22*** 0.07
INV ¥ TLU 0.12* 0.05
INV ¥ AFW 0.63* 0.36
ES –67.18** 32.44
RCI –10.09 44.22
TLU –27.19 26.21
AFW –19.65 21.14
Construction Inv. (CONST) 39.41 25.61
Manufacturing Inv. (MFG) 42.31 26.07
Government Subsidy (GS) 27.22 39.57
Consumption Reallocation (CR) –9.34 32.11
N92
R-squared 0.71
F-Statistic 11.07***
Ordinary Least Squares (OLS) Regression with White’s Robust Standard Errors.
*** p<0.01, **p<0.05, *p<0.1.
(i.e., dummies), which does not present a problem, as policy options cannot simul-
taneously be in two sectors. Furthermore, a moderately high degree of collinearity
exists between Government Subsidy and Manufacturing Investment (rho = -0.49).
However, as only 2 of the 20 policy options that receive a government subsidy are
in the manufacturing sector, we do not believe that the degree of collinearity is
high enough to justify exclusion from the models.
Model 1 indicates that the direct costs of mitigation options constitute a sig-
nifi cant determinant of the overall macroeconomic impacts on GSP. Based on th e
results of Model 1, when the other variables are held constant at their mean values,
when the annualized direct net cost of an average mitigation option decreases by
one million dollars, the annualized GSP impact is expected to increase by about
$0.51 million.

Dan Wei et al.: Macroeconomics of GHG Mitigation
386 J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
Looking at the sectoral decomposition of the direct cost effects, the coeffi cients
of the interaction terms of direct net cost with the four sector dummy variables (in
Model 2) are all negative, which indicates that options with higher direct net cost
are expected to result in less favorable GSP impacts. All of the interaction terms
with respect to direct net cost are statistically signifi cant in Model 2. Based on the
results of Model 2, when all the other variables are held constant at their mean val-
ues, a one million dollar decrease in direct net costs for average mitigation options
in the ES, RCI, TLU, and AFW sectors is expected to increase the annualized GSP
impact by $1.35, $0.42, $0.32, and $0.56 million, respectively.
Model 1 also indicates the statistically signifi cant role of a policy option’s
investment requirement on GSP. If all of the other variables are held constant at
their mean values, when the annualized investment requirement of an average
mitigation option is increased by one million dollars, the annualized GSP impact
is expected to increase by about $0.31 million. All of the interaction terms related
to investment requirement are statistically signifi cant in Model 2. If we hold all of
the other variables constant at their mean values, a one million dollar increase in
investment requirements for an average mitigation option in the ES, RCI, TLU,
and AFW sectors is expected to increase the annualized GSP impact by $0.57,
$0.22, $0.12, and $0.63 million, respectively.
The sectoral binary variables, which try to capture the inherent difference (other
than direct net cost and investment requirements) of options from different sectors,
however, lack statistical signifi cance across the board in both Model 1 and Model
2, except for the Energy Supply sector. It is important to control for differences in
each sector’s mitigation options, but our models show there to be no statistically
signifi cant difference between sectors (other than the impacts of direct net cost
and investment requirement that are captured in the interaction terms). The only
exception is the ES sector. Holding all of the other variables constant at their mean,
an average ES option tends to have a lower stimulus effect on GSP compared with
an average option from the other sectors.
The coeffi cient estimate of the variable pertaining to the capital investment to
the construction sector is positive and just shy of signifi cant in Model 2. The posi-
tive sign of the coeffi cient means that those mitigation options that involve a capi-
tal investment expenditure in the construction sector (for example, investments
in building industrial plants, electricity generation facilities, highways, or other
infrastructure) have an overall positive impact on a state’s macroeconomy. Based
on the results of Model 2, holding all the other variables fi xed at their mean values,
if a mitigation option involves capital investment in construction (i.e., the value of
the CONST dummy variable changes from zero to one), the overall impact on the
annualized GSP is expected to be an increase of $39 million. Simulating the macro-
economic impact of construction capital investment increases in the REMI Model
results in two types of effects: 1) increases in capital costs in the sectors that under-
take the mitigation actions, and 2) increases in the fi nal demand for goods and ser-
vices in the construction sector. In general, the former yields negative impacts on

Dan Wei et al.: Macroeconomics of GHG Mitigation
J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015 387
DOI: 10.7569/JSEE.2015.629503
the economy, while the latter yields positive impacts. The positive sign of the con-
struction investment binary variable indicates that the positive effects are expected
to exceed the negative effects in the four states to which the model was applied.
The coeffi cient estimate of the variable pertaining to the capital investment in
the equipment manufacturing sector is positive as well, but just shy of signifi cance
due to the variability of impacts of those policy options. The positive sign of the
coeffi cient means that at the mean, those mitigation options that involve invest-
ments in manufactured equipment also tend to have a strong positive infl uence on
a state’s overall macroeconomy. Based on the results of Model 2, holding all the
other variables fi xed, if a mitigation option involves capital investment in equip-
ment and machinery (for example, energy-effi cient appliances, vehicles, equip-
ment, and other manufactured devices), that is, the value of the MFG dummy
variable changes from zero to one, the overall average impact on the annualized
GSP is expected to be an increase of $42 million.
Those options that include subsidies from a state government have an overall
positive, but insignifi cant, effect on GSP. In REMI, the state government subsidy
is simulated in two ways: 1) stimulus effects arise from increased spending by
households or increased investment in sectors that receive the subsidies, 2) while
dampening effects stem from the decrease of the same amount of government
spending elsewhere. The positive sign of this variable indicates it is expected that
the stimulus effects of directing government subsidies to mitigation options, in
general, can more than offset the dampening effects associated with decreased
government spending in other areas.
Mitigation options that include consumption reallocation have only a minimal
infl uence on a state’s GSP, on the average. Whereas some mitigation options that
include a consumption reallocation have overall positive effects on a state’s GSP
and others have overall negative effects, based on the results of Model 2, an aver-
age mitigation option that includes a consumption reallocation has a $9 million
lower positive effect on GSP if all the other variables are held constant at their
mean values. Again, however, this relationship is not statistically signifi cant.
4 Regression Model for Employment Impacts
We developed similar regression models to that shown in equation 1, to estimate
the employment impacts of climate mitigation options. The dependent variable in
this case is the annualized employment impact over the entire planning horizon
in terms of person-years. All of the explanatory variables included in the employ-
ment impact regression models are the same as those included in the correspond-
ing GSP impact regression models.
Tables 4 and 5 provide the results of the regression analyses for employment
impacts. Similar to the modeling of GSP impacts, we ran both a basic model
(Model 3) and an interactive model (Model 4). The former model includes one
independent variable each pertaining to the direct net costs and investment

Dan Wei et al.: Macroeconomics of GHG Mitigation
388 J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
Table 4 Results of the Regression Analysis for Employment Impact -- Model 3
Coeffi cient Robust Std. Error
Direct Net Cost (DNC) –0.0080*** 0.00
Investment Requirement (INV) 0.0126*** 0.00
ES –0.2187 1.17
RCI –1.7634* 1.04
TLU –2.4386*** 0.80
AFW 0.2195 0.57
Construction Inv. (CONST) 1.7896** 0.79
Manufacturing Inv. (MFG) 0.5216 0.67
Government Subsidy (GS) 1.5744 1.10
Consumption Reallocation (CR) 0.6010 0.89
N92
R-squared 0.57
F-Statistic 9.32***
Ordinary Least Squares (OLS) Regression with White’s Robust Standard Errors. ***p<0.01, **p<0.05,
*p<0.1. Additional decimal places provided for coeffi cients due to the magnitude of employment
impacts.
requirements, respectively, associated with the implementation of the GHG miti-
gation options, while in the latter model we include interaction terms to evaluate
the individual sectoral impacts of the direct net costs and of investment require-
ments associated with the options implemented in respective sectors.
The direct net cost of an option provides a signifi cant determinant of the overall
employment impact of this option. Based on the results of Model 3, holding all of
the other variables constant at their mean values, decreasing the annualized direct
net cost of an average mitigation option by one million dollars yields an annual-
ized employment impact increase of about 8.0 person-years.
Model 4, which includes the interaction terms of direct net costs in each sector
with sectoral dummy variables, provides a sectoral decomposition of the effects
stemming from changes in direct net cost. The coeffi cients of the four interaction
terms of direct net cost with the four sector dummies are all negative, which indi-
cate that options with higher direct net cost are expected to result in less favorable
employment impacts. According to Model 4, the coeffi cient estimates show that
the most statistically signifi cant variation across the direct cost variable occurs in
the ES, RCI, and TLU sectors. Holding the non-sectoral binary variables constant
at their mean values, a decrease of one million dollars in direct net cost of an aver-
age mitigation option in the ES, RCI, and TLU sector is expected to increase the
annualized employment impacts by 6.9, 13.9, and 6.8 person-years, respectively.

Dan Wei et al.: Macroeconomics of GHG Mitigation
J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015 389
DOI: 10.7569/JSEE.2015.629503
Table 5 Results of the Regression Analysis for Employment Impact – Model 4
Coeffi cient Robust Std. Error
DNC ¥ ES –0.0069* 0.00
DNC ¥ RCI –0.0139** 0.01
DNC ¥ TLU –0.0068*** 0.00
DNC ¥ AFW –0.0004 0.01
INV ¥ ES 0.0344*** 0.00
INV ¥ RCI 0.0065*** 0.00
INV ¥ TLU 0.0056*** 0.00
INV ¥ AFW 0.0312*** 0.01
ES –2.4217** 0.93
RCI –0.1769 0.94
TLU –0.2919 0.61
AFW –0.1210 0.43
Construction Inv. (CONST) 1.0762* 0.60
Manufacturing Inv. (MFG) 0.3516 0.57
Government Subsidy (GS) –0.2325 0.67
Consumption Reallocation (CR) –0.2268 0.77
N92
R-squared 0.02
F-Statistic 16.78***
Ordinary Least Squares (OLS) Regression with White’s Robust Standard Errors. ***p<0.01, **p<0.05,
*p<0.1. Additional decimal places provided for coeffi cients due to the magnitude of employment
impacts.
Model 3 also indicates that the impact of a policy option’s investment require-
ment on employment is statistically signifi cant. If all the other variables are held
constant at their mean values, when the annualized investment requirement of
an average mitigation option is increased by one million dollars, the annualized
employment impact is expected to increase by about 12.6 person-years. In Model
4, all of the sector-specifi c interaction terms for investment requirement are sta-
tistically signifi cant at the signifi cance level of 0.01. If we hold all the other vari-
ables constant at their means, an increase of one million dollars in the investment
requirement for average mitigation options in each of the ES, RCI, TLU, and AFW
sectors is expected to increase the annualized employment impacts by 34.3, 6.5,
5.6, and 31.2 person-years, respectively.
Based on the results of Model 4, the sectoral binary variables again lack sta-
tistical signifi cance except for the ES sector. That means, across our sample, the

Dan Wei et al.: Macroeconomics of GHG Mitigation
390 J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
sectoral impact has no statistically discernible difference (other than the impacts
of direct net cost and investment requirement that are captured in the interaction
terms) on employment impacts results except for the ES sector mitigation options.
Holding all the other variables constant, an average ES option tends to have lower
stimulus effects to the economy in terms of employment impact compared with an
average option from other sectors.
The coeffi cient estimate of the variable pertaining to capital investment in miti-
gation options directed to the construction sector is positive and signifi cant in both
models. This means that, holding all the other variables constant at their mean
values, those mitigation options that involve a capital investment expenditure
in the construction sector are expected to result in more employment gains than
those options that do not. The coeffi cient of the binary variable pertaining to the
capital investment in equipment is also positive but not statistically signifi cant.
The positive sign of the coeffi cient means those mitigation options that involve
investments in equipment are also expected to lead to a stronger positive effects on
job creation. The higher value of the coeffi cient of CONST (the construction sector
investment binary variable) than the coeffi cient of MFG (the equipment manu-
facturing sector investment binary variable) comes about for two reasons. First,
in most states, the construction sector has a higher Regional Purchase Coeffi cient
(RPC) than the equipment manufacturing sector. This indicates that, dollar for dol-
lar, capital investments in the construction sector are more stimulating to the in-
state job market than investments in equipment manufacturing, whose demand is
satisfi ed by a greater proportion of imports of equipment and related items from
out of state. Second, compared with the equipment manufacturing sectors, the
construction sector is relatively more labor-intensive.
The coeffi cients of the binary variables pertaining to the state government sub-
sidy and consumption reallocation are positive in Model 3, but negative in Model
4. These two variables, however, are not statistically signifi cant in either model.
5 Model Applications and “Next Steps” in Model Development
In response to the need for an affordable and rapid use policy screening tool to
evaluate the likely macroeconomic impacts of GHG mitigation policy options
at an earlier phase of their design process, we developed reduced-form statisti-
cal models that can be used to quickly predict the likely GDP and employment
impacts of these various climate mitigation options. The reduced-form models are
developed based on microeconomic impact assessment results from state stake-
holder processes and REMI macroeconometric modeling results of climate action
plans for four states (Florida, Pennsylvania, Michigan, and New York), which
include the analyses of 92 mitigation policy options across these states.
The reduced-form models presented above have been developed based on
REMI modeling results of 92 individual GHG mitigation options at the state level.
Therefore, the direct application of the regression models should be for individual

Dan Wei et al.: Macroeconomics of GHG Mitigation
J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015 391
DOI: 10.7569/JSEE.2015.629503
options at the state level in order to appropriately capture the impacts of the
dummy variables included in the models. If these regression models are applied
to evaluate the likely macroeconomic impacts of policy options at different scales,
such as policy bundles that aggregate options from one sector together, or miti-
gation options implemented at different geographical levels, the direct net cost
and investment requirement values of the options need to be scaled-up or scaled-
down to the individual option level, as well as to the appropriate geographic
level, before applying the regression equations. For example, when we apply the
models to evaluate the potential macroeconomic impacts of mitigation options at
the national level, the estimated direct net cost and investment requirement of
an option both need to be fi rst divided by a factor of 25 to scale down from the
national level to the state level before applying the regression models to the input
data. (Note that a factor of 25 rather than a factor of 50—as in 50 states––is used
for the national-state scale-down is because the four states from which microeco-
nomic and macroeconomic results were used as the basis for the reduced-form
regression models are larger, in terms of the size of their economies, than average
states in the U.S.). Then the regression application results need to be multiplied by
25 to get back to the national level estimations of GDP and employment impacts.
To the extent that the reduced-form model is applied to regions, states, or sub-state
areas that have economies different in size from the average of the four states upon
which model results are based, macroeconomic results pertaining to the average
impacts of individual options may need to be scaled up or down to account for
those size differences. Similarly, the macroeconomic impacts of options depend to
some extent on the particular, locale-specifi c design of the option, including how
aggressive the option’s goals are, relative to those of the average option by sector
among the 92 options now used as the basis for the reduced-form models. These
issues of model scale and option design need to be considered when interpreting
the results of the reduced-form model in particular applications.
The key next step to further refi ne the reduced-form modeling tool is to expand
the underlying database as more REMI macroeconomic impact analyses on miti-
gation options for additional states and regions are performed. As the underlying
database expands, specifi c regression models that are tailored to specifi c types of
economies or are designed for specifi c sectors in the economy can be developed
as well. We also plan to develop a documented, easily-applied spreadsheet-based
version of the reduced-form model that is convenient for application in the early
option screening phase. The function of the spreadsheet-based application would
include functions such as scale-up/down factors based on state GSP/employ-
ment, and key drivers of policy stringency for at least some key policies.
Note also that the results pertain to conditions in which we assume that all
investment in mitigation options does not displace investment in ordinary plant
and equipment. This requires that additional investment funds become avail-
able by attracting investors from outside the state, attracting federal subsidies, or
using in-region business retained earnings. State governments can take actions to

Dan Wei et al.: Macroeconomics of GHG Mitigation
392 J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
promote the fi rst two of these conditions, while the third is likely in times other
than economic recession years. As such, the estimates yielded by our reduced form
equations should be considered reasonable upper bounds in terms of the availabil-
ity of additional investment funds to support the GHG mitigation actions.
Acknowledgement
This paper is based on research supported by The Center for Climate Strategies
(CCS) through a grant to USC. We wish to offer special thanks to CCS members
and affi liated researchers for access to the microeconomic impact data for GHG
mitigation options in various state climate action plans. We are also grateful for
the helpful comments of the editors and one anonymous reviewer. Of course, any
remaining errors and omissions are solely those of the authors. Moreover, the
views expressed in this article represent those of the authors and not necessar-
ily any of the institutions with which they are affi liated nor the institutions that
funded the research.
References
1. Florida Climate Action Team (FCAT). Florida’s Energy and Climate Change Action Plan.
http://www.naseo.org/Data/Sites/1/documents/stateenergyplans/FL_2008.pdf
(2008).
2. Pennsylvania Department of Environmental Protection (PA DEP). Pennsylvania Final
Climate Change Action Plan. http://www.elibrary.dep.state.pa.us/dsweb/View/
Collection-10677 (2009).
3. Michigan Climate Action Council (MCAC). Michigan Climate Action Plan MCAC
Final Report. http://www.michigan.gov/documents/deq/deq-miclimateactionplan-
part1_276563_7.pdf (2009).
4. New York State Climate Action Council (NYSCAC). Climate Action Plan Interim Report.
http://www.dec.ny.gov/docs/administration_pdf/irpart1.pdf (2010).
5. Regional Economic Models, Inc. (REMI). REMI Website: http://www.remi.com/ (2015)
6. G. Treyz, M.J. Greenwood, G.L. Hunt, and D.S. Rickman. Migration, regional equilib-
rium, and the estimation of compensating differentials. Am. Econom. Rev. 81 (5): 1382–
1390 (1991).
7. G. Treyz. Regional Economic Modeling: A Systematic Approach to Economic Forecasting and
Policy Analysis, Boston, Kluwer (1993).
8. W. Fan, F. Treyz, and G. Treyz. An evolutionary new economic geography model. J. Reg.
Sci. 40 (4), 671–695 (2000).
9. F. Treyz, and G. Treyz. The Evaluation of Programs Aimed at Local and Regional
Development: Methodology and Twenty Years’ Experience Using REMI Policy Insight,
Evaluating Local Economic and Employment Development: How to Assess What Works
Among Programmes and Policies, (pp. 151–189). Paris, France: Organization for Economic
Co-operation and Development, ISBN: 9264017089 (2004).

Dan Wei et al.: Macroeconomics of GHG Mitigation
J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015 393
DOI: 10.7569/JSEE.2015.629503
10. P. Dixon, J. Giesecke, M. Rimmer, and A. Rose, The economic costs to the U.S. of clos-
ing its borders: A computable general equilibrium analysis. Defen. Peace Econom. 22 (1),
85–97 (2011).
11. P. Dixon and M. Rimmer. Validation in Computable General Equilibrium Modeling, in
Handbook of Computable General Equilibrium Modeling, P.B. Dixon, D.W. Jorgenson, (Eds.),
pp. 1271–1330, Amsterdam: North-Holland (2013).
12. A. Rose, D. Wei, and N. Dormady. Regional macroeconomic assessment of the pennsyl-
vania climate action plan. Reg. Sci. Policy Pract. 3 (4), 357–379 (2011).
13. J.A. Giesecke and J.R. Madden. Regional Computable General Equilibrium Modeling,
in Handbook of Computable General Equilibrium Modeling, P. D. , Vol. 1A, pp. 379–476,
Amsterdam: North-Holland (2013).
14. A. Rose and D. Wei. Macroeconomic impacts of the fl orida energy and climate change
action plan. Clim. Policy 12(1), 50–69 (2012).
15. A. Rose, D. Wei, J. Wennberg, and T. Peterson. Climate change policy formation in
Michigan: The case for integrated regional policies. Intl. Reg. Sci. Rev. 32 (4), 445–465
(2009).
16. S. Miller, D. Wei, and A. Rose. The Macroeconomic Impact of the Michigan Climate Action
Council Climate Action Plan on the State’s Economy. Report to Michigan Department
of Environmental Quality. http://www.climatestrategies.us/ewebeditpro/items/
O25F22416.pdf (2010).
17. D. Wei and A. Rose. The Macroeconomic Impact of the New York Climate Action Plan: A
Screening Analysis. Report to New York State Energy Research and Development
Authority (2011).
18. T. Peterson, R. McKinstry, and J. Dernbach. Developing a comprehensive approach to
climate change policy in the United States that fully integrates levels of Government
and economic sectors. Virginia Environ. Law J. 26 (219), 227-269 (2008).
19. M. Doria, E. Boyd, E.L. Tompkins, and W.N. Adger. Using expert elicitation to defi ne
successful adaptation to climate change. Environ. Sci. Pol. 12 (7), 810–819 (2009).
20. H. White. A heteroskedasticity-consistent covariance matrix estimator and a direct test
for heteroskedasticity. Econometrica 48 (4), 817–838. (1980).
21. J.M. Woolridge. Introductory Econometrics: A Modern Approach. Cengage (2013).

Dan Wei et al.: Macroeconomics of GHG Mitigation
394 J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
Appendix A. List of Major GHG Mitigation Policy Options
Table A1 lists the major GHG mitigation options that are recommended in the four
state CAPs (those that are capable of reducing emissions by more than 1% of base-
line in at least one of the four states analyzed). Overall they represent two-thirds
of the options and around 93% of total emission reductions across the four states.

Dan Wei et al.: Macroeconomics of GHG Mitigation
J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
395
Appendix Table A1. List of Policy Options Applying to More than 1% of GHG Emission Reduction in Various States
Policy Option Policy Name
Florida Pennsylvania Michigan New York
% GHG
Reduction
Cost-
Effectiveness
% GHG
Reduction
Cost-
Effectiveness
% GHG
Reduction
Cost-
Effectiveness
% GHG
Reduction
Cost-
Effectiveness
Energy Supply Options
FL_ESD-5; MI_ES1,
NY_PSD-2/6
Renewable
Portfolio
Standard (RPS)
7.45% –$32.3 3.73% $21.0 4.42% $47.1 14.56% $30.0
ESD-6, PA_E-10,
MI_ES2 Nuclear Power 1.58% $40.1 4.96% $8.2 2.59% $24.8
FL_ESD-8, PA_E-
9, MI_ES4,
NY_RCI-2b,
Combined Heat
and Power
(CHP) Systems
0.47% $5.6 1.47% $9.8 0.17% $5.1 0.43% $2.3
FL_ESD-9, PA_E-6,
MI_ES3
Power Plant
Effi ciency
Improvements
1.92% –$15.6 1.83% –$16.1 0.85% $3.1
FL_ESD-11
Landfi ll Gas-
To-Energy
(LFGTE)
1.88% $1.1
PA_E-5
Carbon
Capture and
Sequestration
1.70% $33.7
Residential, Commercial, and Industrial Options
FL_ESD-12, PA_
RC-10/11/13, MI_
RCI1, NY_RCI-2a
Demand-Side
Management
(DSM)
4.71% –$47.9 4.45% –$16.3 9.87% –$31.4 6.70% $0.0
FL_ESD-13a
Energy Effi ciency
in Existing
Residential
Buildings
1.17% –$31.2
(Continues)

Dan Wei et al.: Macroeconomics of GHG Mitigation
396 J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
Policy Option Policy Name
Florida Pennsylvania Michigan New York
% GHG
Reduction
Cost-
Effectiveness
% GHG
Reduction
Cost-
Effectiveness
% GHG
Reduction
Cost-
Effectiveness
% GHG
Reduction
Cost-
Effectiveness
PA_RC-6/8, MI_RCI2,
NY_RCI-7
Appliance
Standards 4.99% –$40.3 8.75% –$31.4 2.48% –$30.9
PA_RC-9,
NY_RCI-3a/3b/3c
Customer-Sited
Renewable
Energy
0.48% $67.8 4.42% $17.1
PA_Ind-2
Industrial NG &
Electricity Best
Management
Practices
1.74% –$41.7
NY_RCI-11 Industrial Process
Incentives 1.03% –$108.9
Transportation and Land Use Options
FL_TLU-1, MI_TLU4,
NY_TLU-4
Alternative Fuel
Strategies 2.72% –$158.2 2.02% $4.8 3.35% $90.5
FL_TLU-4
Improving Trans-
portation System
Management
(TSM)
1.51% –$89.1
PA_T-9, MI_TLU5,
NY_TLU-7/10 Transit 0.40% $61.0 0.15% $117.9 2.13% $286.9
NY_TLU-1
Vehicle Technol-
ogy and Opera-
tions
6.70% $71.1
Agriculture, Forestry, and Waste Management Options
FL_AFW-1, PA_F1/3,
NY_AFW-7a
Forest Protection/
Restoration 0.13% $29.0 0.78% $21.6 1.85% $6.9
FL_AFW-2A1/2A2,
PA_F-4, MI_AFW6,
NY_AFW-7c
Afforestation/
Reforestation 3.17% $5.6 1.35% $25.0 0.32% $52.1 0.95% $41.3
Appendix Table A1. (Continued)

Dan Wei et al.: Macroeconomics of GHG Mitigation
J. Sustainable Energy Eng., Vol. 2, No. 4, April 2015
DOI: 10.7569/JSEE.2015.629503
397
FL_AFW-2B, PA_F-7,
MI_AFW7, NY_AFW-
7b
Urban Forestry 1.88% $11.1 1.01% $95.4 1.04% $210.0 0.79% $160.4
FL_AFW-4, PA_F-
8/9a/9b/W-1/5/6,
NY AFW-6
Expanded Use of
Agriculture, For-
estry, and Waste
Management
(AFW) Biomass
Feedstocks for
Electricity, Heat,
and Steam Pro-
duction
8.63% $23.4 0.60% –$17.0 0.16% $1.1
FL_AFW-6, NY_AFW-
5
Reduce the Rate
of Conversion of
Agricultural Land
and Open Green
Space to Develop-
ment
0.11% $103.6 2.17% $18.3
FL_AFW-7
In-State Liquid/
Gaseous Biofuels
Production
1.77% -$8.9
FL_AFW-9B
WWTP Biosolids
Energy Produc-
tion & Other Bio-
mass Conversion
Technologies
1.08% $49.0
PA_W-2, MI_AFW5,
NY_AFW-3 Waste Recycling 1.84% –$8.2 7.03% $18.9 0.28% $40.1