Memo on Using Other (Non-CGE) Economy-Wide Models to Estimate

Social Cost of Air Regulation

September 22, 2015

Prepared for the U.S. EPA Science Advisory Board Panel on Economy-Wide Modeling of
the Benefits and Costs of Environmental Regulation

This paper has been developed to inform the deliberations of the SAB Panel on the technical merits and
challenges of economy-wide modeling for an air regulation. It is not an official EPA report nor does it
necessarily represent the official policies or views of the U.S. Environmental Protection Agency.

-------
[This page intentionally left blank.]

2

-------
Table of Contents

1	Introduction	4

2	Measuring Welfare in Economy-Wide Models	4

3	Dynamic Stochastic General Equilibrium Models	6

3.1	Environmental Policy Applications	8

3.2	Potential Advantages and Limitations	9

4	Concluding Remarks	10

5	Appendix	15

3

-------
1 Introduction

The white paper, Economy-wide Modeling: Social Cost and Welfare, (henceforth denoted as the social
cost white paper) that accompanies this memo focuses on the technical merits and challenges of using
computable general equilibrium (CGE) models to estimate the social cost of an air regulation based on
EPA's experience with that class of models. In contrast, this memo takes a preliminary look of the potential
use of other types of economy-wide models for estimating the social cost of air regulations. While EPA
has limited experience utilizing these other economy-wide modeling approaches, they may hold promise
for estimating certain aspects of social cost potentially missed by current CGE models.

As previously discussed in the social cost white paper, the concept of social cost encompasses direct,
indirect, and transitional costs (EPA, 2010). Transitional costs are short term costs incurred while the
economy is adjusting to a new equilibrium. Given that most state-of-the-art single-country CGE models
for the U.S. economy are long-run models with instantaneous (or near-instantaneous) market adjustment
in response to a policy shock, these models may miss transitional costs, and may therefore under-
represent social cost. This leads to the question of whether other types of economy-wide models, aside
from CGE models, could potentially add value with regard to the measurement of social cost by more
completely capturing transitional costs. A related question is whether or in what circumstances
transitional costs are expected to be substantial enough to potentially warrant investment in modeling
tools that more fully capture them, in addition to those tools that capture the first order impacts
associated with achieving the new equilibrium.

While these other (non-CGE) economy-wide models are used for a wide array of policy analysis, the focus
of this memo is on their potential for estimating the social cost of an air regulation. In particular, we focus
on the degree to which these models measure changes in consumer and producer surplus, the building
blocks for estimating changes in economic welfare, both in equilibrium and transition, in response to a
policy shock.1 Several types of economy-wide models are discussed: input-output models, large-scale
macro-econometric forecasting models, a hybrid between the two (input-output (l-O) macro-econometric
models), and dynamic stochastic general equilibrium models (DSGE). It is our conclusion that, aside from
CGE models, only DSGE models produce an estimate of changes in consumer and producer surplus
required for benefit-cost analysis of policies. For this reason, the main portion of the memo discusses
DSGE models while the other three types of economy-wide models are described in the appendix.

2 Measuring Welfare in Economy-Wide Models

There are a number of ways to characterize the main attributes of different types of economy-wide
models. Arora (2013b) distinguishes between economy-wide models that are based on optimization by
economic agents and those that are not. Optimization models build up to the macro-economy from
microeconomic foundations: consumers are assumed to maximize utility and firms maximize profits

1 Note that these other types of economy-wide models will also be discussed in the economic impacts white paper
with regard to their ability to shed light on other economic measures aside from social cost that are of potential
interest to policymakers.

4

-------
subject to resource constraints, preferences, and production possibilities; the models also examine the
ramifications of a policy shock in equilibrium (i.e., when supply re-equilibrates with demand in all
markets). In contrast, non-optimized models, while potentially based on consumer and/or firm level data,
are generally not predicated on the concept that economic agents optimize decisions. In addition to CGE
models, only DSGE models can be classified as an optimization model, while the other types of economy-
wide models discussed in this memo, input-output, macro-econometric, and 1-0 macro-econometric
models, are all examples of non-optimized models.2

Which of these models can be used to derive a welfare measure follows directly from whether agents
optimize. Optimization models, because of their microeconomic foundations and, in particular, their basis
in utility theory, can calculate equivalent variation (EV) or compensating variation (CV) to capture welfare
changes due to a policy shock. In the case of a more stringent regulation with positive net compliance
costs, EV measures what a consumer would be willing to pay to avoid an increase in prices (and thus, a
decline in real income) resulting from a regulation. CV measures how much a consumer would need to be
compensated to accept changes in prices and income such that they achieve the same level of utility
experienced prior to the policy shock (EPA 2010). In an economy-wide model that also includes benefits,
net willingness to pay would also factor in changes in environmental quality as well as how these interact
with price changes. Non-optimized models commonly report the effect of a policy shock on other
aggregate economic measures of interest to policymakers such as changes in GDP, consumption, and
employment, but as previously discussed in the social cost paperthese may not correlate well with welfare
measures appropriate for benefit-cost analysis.

CGE and DSGE models are both built on micro-theoretic foundations so that behavioral responses in the
models derive from the underlying structure of the models. These models also reduce the potential
applicability of the Lucas critique since they do not assume that past behavioral relationships of
consumers and firms govern future reactions.3 Instead, consumer and firm behavior is modeled directly,
which means that responses to policy shifts can be incorporated (Arora, 2013a; Arora, 2013b).

Another potentially useful distinction across model types is structural versus reduced-form specifications.
Woodford (2009) distinguishes between reduced-form methods used to characterize data under a priori
assumptions and structural models that explicitly specify underlying preferences, production, and

2	We do not discuss vector autoregressive (VAR) techniques separately. VAR econometric techniques specify multiple
endogenous variables in terms of other endogenous variables and error terms (Sims introduced VAR in 1980 as an
alternative to large-scale, macro-econometric models; he argued that all variables in these models could be
potentially endogenous and therefore decisions about which are exogenous for purposes of solving the model are
ad hoc and unjustified (Bj?)rnland, 2000)). VARs often are reduced-form approaches that strive to explain observed
empirical phenomena and relationships, including sources of business cycles. Recently, some have been depicted
VAR techniques as useful for describing data and for forecasting macroeconomic phenomena, but because they do
not differentiate between correlation and causation that may prove less useful for structural inference and policy
analysis unless explicitly derived from theory (Stock and Watson, 2001). See Arora (2013a) for further discussion.

3	Because DSGE and CGE models are based in microeconomic theory, they potentially avoid the Lucas critique.
Reduced-form macro-econometric models that rely on historical data cannot take into account the possibility that a
firm or consumer may modify its behavior with changes in policy. The Lucas critique points out that the inability to
account for this dependence invalidates these models for purposes of policy evaluation outside of short-term
forecasting (Schmidt and Wieland, 2013; Arora, 2013a; Heutel and Fischer, 2013).

5

-------
resource allocation in ways that are consistent with economic theory. The calibration of structural or
behavioral model parameters with actual data ensures that the model represents important economic
features while remaining in agreement with the underlying theory. In contrast, reduced-form models are
principally data driven and based on empirical observation. For example, many equations in a macro-
econometric model are specified in reduced form, and the parameters are econometrically estimated
using time-series data. The equations in such models form a system of simultaneous equations that are
then combined with national income account identities (Smith, 2012). Both CGE and DSGE models have
been classified as structural models.4 Input-output, macro-econometric, and l-O-macro-econometric
models are typically empirically driven exercises and as such are largely reduced-form, though macro-
econometric and 1-0 macro-econometric models are often described as guided by macroeconomic theory.

3 Dynamic Stochastic General Equilibrium Models

As only the DSGE model may potentially be used to estimate changes in consumer and producer surplus,
in addition to CGE models, we provide an initial overview of its capabilities and limitations based on
available assessments in the literature.

DSGE models are dynamic, inter-temporal models that attempt to explain aggregate macroeconomic
phenomena but are based in micro-economic theory (e.g., households maximize utility and firms
maximize profit subject to income constraints; prices are assumed to adjust until markets clear)
(Woodford, 2009). Similar to a CGE model, DSGE models often assume a representative consumer that
works, consumes, and saves (invests); a representative firm that hires labor, capital and other
intermediate commodities to supply a final product; and a government sector that is tasked to design and
implement monetary or fiscal policy, in addition to participating in the product and product markets (see
Figure 1). By combining these basic ingredients, DSGE models describe the interrelated movement or
evolution of basic macroeconomic variables.

It is worth noting that, at a high level of abstraction, "DSGE models in principal are not different from CGE
models" (Townsend, 2010). For instance, both types of models are built from microeconomic foundations
and assume agents optimize and that markets clear in the long run. However, since these two economy-
wide modeling approaches arise from different literatures, the specific implementations reflect different
modeling priorities and tradeoffs. While the number of sectors represented in a DSGE model is typically

4 Recall that CGE models are built around the assumption that for some discrete period of time, an economy can be
characterized by a set of conditions in which supply equals demand in all markets. When the imposition of a
regulation alters conditions in one market, the model will determine a new set of relative prices that re-equilibrate
supply and demand in all markets, accounting for interactions and feedbacks between commodity, input, household,
and government sectors (EPA, 2010). CGE models examine the medium to long run effects of policy shocks, can be
static or dynamic, are calibrated but may also rely on econometrically estimated behavioral parameters, and vary
widely with regard to sectoral detail and representation of trade. For purposes of examining the effects of an air
regulation, EPA has typically relied on single-country dynamic CGE models of the United States with a relatively
simple representation of trade and a moderate level of sectoral detail (35 to 40 sectors).

6

-------
small relative to a CGE model, they are able to model the dynamics of the economy overtime in ways that
a CGE model typically does not, making them well suited for studying the cyclical effects of policy.5

Figure 1 illustrates the emphasis in many DSGE models on forecasting macroeconomic aggregates.
Demand is a function of expectations about uncertain future real economic activity (Ye) and the nominal
interest rate minus expected inflation (i - ne). Supply is characterized with regard to key drivers of
inflation. The government sector determines monetary policy through its setting of nominal interest rates.

Figure 1: A DSGE Model's Basic Structure

Demand
shocks

Mark-up
shocks

Source: Sbordone, et al., 2010

Another difference from CGE models is that, in addition to resource constraints, consumers in DSGE
models maximize the present value of utility flows in the presence of processes and parameters that are
subject to random, unexpected shocks in each period (e.g., in productivity, tastes, technology), as shown
in Figure 1 (also see Romer, 2011; Faust and Gupta, 2012; De Grauwe, 2010; Heutel and Fischer, 2013).
Consumers and firms typically are assumed to respond rationally to current and potential future cyclical
changes when optimizing (i.e., on average, they correctly form expectations about the future). Since much
DSGE work in macroeconomics has focused on explaining differences in growth across countries, the
financial sector has often been less developed in these models, though this is an active area of current
research (Townsend, 2010; Sbordone et al., 2010).

Similar to efforts to modify some CGE models (see the social cost white paper), short-run market rigidities
may be introduced into DSGE models via a variety of mechanisms. For instance, some models assume

5 "Roughly, the more the model incorporates dynamics and uncertainty, the less it is able to handle disaggregation"
(Townsend, 2010).

7

-------
monopolistically competitive firms that cannot instantaneously and costlessly adjust. Others assume
production inputs are firm-specific (i.e., firms only accumulate capital for their own use), or only a certain
proportion of households can reset their nominal wage in any particular time period (Sbordone, et al.,
2010; Schmidt and Wieland, 2013).6 It is through the introduction of some type of short-run rigidity that
an analyst might capture transitional costs. We are unaware of any explicit comparisons of CGE and DSGE
models in this capacity, however.

3.1 Environmental Policy Applications

While DSGE models are often used by central banks or in academic research (Woodford, 2009; Sbordone,
et al., 2010), there are only a few applications of a DSGE model in an environmental policy context. One
set of papers compares how different economic instruments perform in an economy that faces some pre-
defined set of exogenous productivity, price, or wage markup shocks. For example, Fischer and Springborn
(2011) explore the macroeconomic performance of an emissions tax, cap-and-trade, and intensity targets
for an economy that faces uncertain future productivity. In their framework, a polluting input is a
necessary part of production and any emission abatement requires reducing output. Heutel (2012) models
emissions as a byproduct of production; firms are able to reduce emissions by installing a costly
abatement technology. This framework is used to assess optimal environmental policy when economic
fluctuations are caused by persistent productivity shocks. Another set of papers in the environmental
economics literature combines aspects of DSGE models with highly stylized global integrated assessment
models (lAMs) - Nordhaus' DICE or RICE model, in particular - for the analysis of climate policy (e.g.,
Hassler and Krussel, 2012; Cai et al., 2012; Lemoine and Traeger, 2014; and Barrage, 2014).7

In comparison to CGE models, the DSGE models used in environmental applications are highly aggregate,
stylized representations of the economy. Arora (2013a) points out that this is in part due to the
incorporation of uncertainty. For instance, Fischer and Springborn (2011), Heutel (2012), and
Angelopoulos et al. (2010) model a single representative consumer and a single representative firm.
Dissou and Karnizova (2012) model six production sectors (coal, electricity, natural, gas, services, energy-
intensive goods, and non-energy-intensive goods) and allow for more than one source of an exogenous
shock in the economy: each sector experiences its own autocorrelated production shock, which means
that contrary to previous studies they find that the performance of the economic instrument varies with
the source of the shock. Of the four papers that compare instruments for environmental policy, only
Fischer and Springborn (2011) and Dissou and Karnizova (2012) include labor as an input into production.

6	Arora (2013a) notes another difference in the calibration process for CGE and DSGE models. CGE models typically
calibrate structural or behavioral parameters so that the model is able to reproduce the social accounting matrix
(SAM), which "records all the transactions and transfers between production activities, factors of production, and
agents in an economy." DSGE models do not rely on a base year SAM to calibrate parameter values. Instead they
combine commonly accepted values from the literature with calibration of certain ratios to long run averages in the
macro data. DSGE parameters can also be estimated using historical data.

7	DICE "is an optimal growth model based on a global production function with an extra stock variable (atmospheric
carbon dioxide concentrations). Emission reductions are treated as analogous to investment in "natural capital." By
investing in natural capital today through reductions in emissions— implying reduced consumption—harmful effects
of climate change can be avoided and future consumption thereby increased (Interagency Working Group, 2010).

8

-------
3.2 Potential Advantages and Limitations

Bukowski (2014) identifies two possible advantages that DSGE models bring to policy analysis that are not
afforded by static or recursively dynamic CGE models. First, DSGE models are fully dynamic such that
investment and saving decisions are interdependent and endogenously determined by the future path of
the economy. The second advantage of DSGE models over CGE models lies in their ability to explicitly
model decision-making under uncertainty (also Arora, 2013a). While both types of models assume rational
expectations, forward-looking CGE models typically assume perfect foresight, while DSGE models allow
for imperfect foresight when forming expectations about the future. An analyst has the ability to vary
stochastic processes and agent's information sets within the model to explore the dynamic properties of
the model. For instance, in contrast to most CGE models, "not only long but also short to medium term
policy implications can differ depending on whether economic agents perceive a given policy as
transitional or permanent" (Bukowski, 2014). See the uncertainty white paper for more discussion.

When using DSGE models to analyze environmental policy, researchers have noted a further appeal over
CGE models in their ability to model the short- to medium-run within a tractable and consistent economic
framework (Heutel and Fischer, 2013; Woodford, 2009). For instance, DSGE models can potentially be
used to analyze the short run effects of policies under assumptions other than perfectly competitive
markets and fully flexible wages and prices (e.g., one could model imperfect competition in labor or
product markets, assume wages or prices that are fixed for some period of time before adjusting to reflect
current market conditions, or build in search and matching frictions that result in underutilized resources
(Woodford, 2009)). While CGE models also can accommodate some market rigidities and economies of
scale, they do not typically explicitly model the transition period as the economy moves from one
equilibria to another.

Despite their popularity with central banks, economists appear to disagree on the extent to which
applying DSGE models to policy making is practical.8 Cogley and Yagihashi (2010) caution that if a DSGE
model is misspecified (i.e., it fails to predict the "true" correctly specified model parameters), it may not
be policy invariant. While this is true of any model, the introduction of random, exogenous shocks may
complicate matters. For instance, correlated structural shocks are a potential indicator of misspecification
(Andrle, 2014). In an ideal setting, structural shocks are uncorrelated with the policy intervention (Faust,
2009). However, some of the shocks most commonly evaluated in DSGE models - for instance, wage and
price markup shocks or government spending shocks - may fail to meet this criterion (Chari, et al., 2008;
Faust, 2009). This concern may be less immediate when evaluating the effects of environmental policies
that do not explicitly raise government revenues or vary over time to mitigate exogenous fiscal or
monetary shocks (e.g., Heutel (2012) discusses the possibility of a carbon tax being pro-cyclical).

8 DSGE models have been criticized for assuming rational expectations. De Grauwe (2010) states that it imposes
higher level "cognitive capabilities on individuals" than they actually have. (Many CGE models are subject to the
same criticism.) DSGE models also are sometimes criticized for their parameter estimation approach. Tovar (2008),
for example, notes that maximum likelihood methods in DSGE models leads to "stochastic singularity" problems.

9

-------
A second recent criticism of DSGE models relates to their overall forecasting performance. De Grauwe
(2010) points out that the validity of a model is usually judged by its ability to make empirical predictions
that are supported by the data. In this regard, Morely (2010) points out that "just as older versions of
large-scale macro-econometric models failed to predict or even explain the 'stagflation' of the 1970s,
micro-founded 'dynamic stochastic general equilibrium' (DSGE) models inspired by the Lucas critique have
failed with the Great Recession of 2007-2009." Caballero (2010) echoes this sentiment:
"macroeconomics—by which I mainly mean the so-called dynamic stochastic general equilibrium
approach—has become so mesmerized with its own internal logic that it has begun to confuse the
precision it has achieved about its own world with the precision that it has about the real one." (See also
Robert Solow's 2010 testimony regarding DSGE models' predicative abilities to the House Committee on
Science and Technology.) Woodward (2009) points out that the macroeconomics literature offer little in
the way of guidance or agreement on how to best specify an empirical DSGE model. That said, it remains
an open question - to our knowledge, undiscussed in the literature - to what extent this criticism applies
when DSGE models are used to assess the potential welfare effects of a policy ex-ante rather than for
forecasting purposes. When conducting benefit-cost analysis of air regulations, the EPA focuses on
estimating changes in welfare relative to a baseline rather than correcting predicting any particular future.

4 Concluding Remarks

Given EPA's relative lack of experience with DSGE and other types of economy-wide modeling approaches
aside from CGE models, this memo should be viewed as an initial foray into characterizing the potential
value added of other economy-wide models for estimating the social cost of an air regulation. In
particular, this memo is designed to aid the SAB Panel in addressing the charge questions:

•	Are there other economy-wide modeling approaches beside CGE that EPA should consider for
estimating the social cost of air regulations?

•	What are the potential strengths and weaknesses of these alternative approaches in the
environmental regulatory context compared to using a CGE approach?

As in the social cost white paper, we have left aside the question of whether benefits should also be
included in an economy-wide model to fully characterize changes in economic welfare.

DSGE models are particularly intriguing to the EPA because of their ability to potentially characterize the
short-run implications of unexpected policy shocks, which could allow an analyst to characterize the
transition between the baseline and new equilibria orto formally characterize uncertainty in an economy-
wide framework. However, since the application of DSGE models to environmental policy questions in the
published literature is relatively recent, it is unclear how or to what extent this class of models could be
applied to the analysis of an air regulation, what features of a DSGE model would be particularly important
in a regulatory context, or how results from a DSGE model compare to those of a CGE model when used
to evaluate the same environmental policy. In addition, the relative high level of sector aggregation in
most DSGE models raises similar issues to those discussed in the social cost paper: when using CGE models
to measure social costs, there is the potential for missing heterogeneity in compliance technologies and
costs that may matter when estimating aggregate social cost.

10

-------
References

Andrle, M. 2014. "Estimating Structural Shocks with DSGE Models." Working Paper. International
Monetary Fund, Research Department.

Angelopoulos, K., Economides, G., and Philippopoulos, A. 2010. "What is the best environmental policy?
Taxes, permits and rules under economic and environmental uncertainty." Working paper.

Arora, V. 2013a. "An Evaluation of Macroeconomic Models for Use at EIA." Working Paper. U.S. Energy
Information Administration. December.

Arora, V. 2013b. "Alternative Measures of Welfare in Macroeconomic Models." Working Paper. U.S.
Energy Information Administration. December.

Barrage, L. 2014. "Optimal Dynamic Carbon Taxes in a Climate-Economy Model with Distortionary Fiscal
Policy." Working paper, University of Maryland. November.

Bjeirnland, H. 2000. "VAR Models in Macroeconomic Research." Statistics Norway, Research Department.
Paper 2000/14.

Burkowski, M. 2014. "On the Endogenous Directed Technological Change in Multi Sector DSGE Model:
The Case of Energy and Emissions efficiency." NEUJOBS working paper, D3.8.

Caballero, R. 2010. "Macroeconomics after the Crisis: Time to Deal with the Pretense-of-Knowledge
Syndrome." NBER working paper, No.16429.

Cai, Y., Judd, K., and Lontzek, T. 2012. "DSICE: A Dynamic Stochastic Integrated Model of Climate and the
Economy." RDCEP working paper, 12-02.

Chari V., Kehoe, P., and McGrattan, E. 2008. "New Keynesian Models: Not Yet Useful for Policy Analysis."
National Bureau of Economic Research Working Paper. No. wl4313.

Cogley, T., and Yagihashi, T. 2010. "Are DSGE Approximating Models Invariant to Shifts in Policy?" B.E.
Journal of Macroeconomics, Contributions 10 (1): Article 27.

De Grauwe, P. 2010. "The scientific foundation of dynamic stochastic general equilibrium (DSGE) models."
Public Choice 144: 413-443.

Dissou, Y., and Karnizova, L. 2012. "Emissions cap or emissions tax? A multi-sector business cycle analysis."
Working Paper, Univ. of Ottawa.

11

-------
Dwyer, L., Forsyth, P., and Spurr, R. 2005."Estimating the Impacts of Special Events on the Economy."
Journal of Travel Research 43: 351-359.

Dwyer, L., Forsyth, P., and Spurr, R. 2006. "Assessing the Economic Impacts of Events: A Computable
General Equilibrium Approach." Journal of Travel Research 45: 59-66.

European Commission. 2015. "Methods, Models, and Costs and Benefits." Better Regulation Guidelines,
Toolbox. See http://ec.europa.eu/smart-regulation/guidelines/toc guide en.htm .

Faust, J. 2009. "The New Macro Models: Washing Our Hands and Watching for Ice-Bergs." Economic
Review: 45-68.

Faust, J., and Gupta, A. 2012. "Posterior Predictive Analysis for Evaluating DSGE Models." NBER Working
Paper, No. 17906.

Fischer, C., and Springborn, M. 2011. "Emissions Targets and the Real Business Cycle: Intensity
Targets versus Caps or Taxes," Journal of Environmental Economics and Management 62: 352-366.

Hahn, R., and Hird, J. 1991. "The Costs and Benefits of Regulation: Review and Synthesis",

Yale Journal on Regulation 8: 233-278.

Hassler, J., and Krusell, P. 2012. "Economics and Climate Change: Integrated Assessment in a Multi-Region
World." Journal of the European Economic Association, European Economic Association 10(5): 974-1000.

Heutel, G. 2012. "How Should Environmental Policy Respond to Business Cycles? Optimal Policy under
Persistent Productivity Shocks." Review of Economic Dynamics 15: 244-264.

Heutel, G., and Fischer, C. 2013. "Environmental Macroeconomics: Environmental Policy, Business Cycles,
and Directed Technical Change." Working Paper.

Interagency Working Group. 2010. Technical Support Document: Social Cost of Carbon for Regulatory
Impact Analysis under Executive Order 12866. See

https://www.whitehouse.gov/sites/default/files/omb/inforeg/for-agencies/Social-Cost-of-Carbon-for-
RIA.pdf.

Lemoine, D., and Traeger, C. 2014. "Watching Your Step: Optimal Policy in a Tipping Climate." American
Economic Journal: Economic Policy 6(1): 137-166.

Morgenstern, R., Ho, M., and Shih, J. 2008. "Impact of Carbon Price Policies on U.S. Industry." RFF
Discussion Paper 08-37. Resources for the Future.

Morley, J. 2010. "The Emperor Has No Clothes." Macroeconomic Advisers Macro Focus 5, 4:1-14.
Macroeconomic Advisers, LLC.

12

-------
Office of Management and Budget (OMB). 1995. Guidance for Implementing Title II of S. 1. Memorandum.
March 31. https://www.whitehouse.gov/sites/default/files/omb/memoranda/m95-09.pdf

Office of Management and Budget (OMB). 2014. 2014 Draft Report to Congress on the Benefits and Costs
of Federal Regulations and unfunded Mandates on State, Local, and Tribal Entities. See

https://www.whitehouse.gov/sites/default/files/omb/inforeg/2014 cb/draft 2014 cost benefit report

-updated.pdf.

Portney, P. 1981. "The Macroeconomic Impacts of Federal Environmental Regulation." Natural Resources
Journal 459:459-488.

Romer, P. 2011. Advanced Macroeconomics. McGraw-Hill.

Sbordone, A., Tambalotti, A., Rao, K., and Walsh, K. 2010. "Policy Analysis Using DSGE Models: An
Introduction." Federal Reserve Bank of New York, Economic Policy Review.

Schmidt, S., and Wieland, V. 2013. "The New Keynesian Approach to Dynamic General Equilibrium
Modeling: Models, Methods and Macroeconomic Policy Evaluation" in The Handbook of CGE Modeling.
Vol. 1.

Smith, V. 2012. "Reflections—In Search of Crosswalks between Macroeconomics and Environmental
Economics." Review of Environmental Economics and Policy 6 (2): 298-317.

Solow, R. 2010. "Building a Science of Economics for the Real World" Testimony for the House Committee
on Science and Technology, July. See http://www.gpo.gov/fdsvs/pkg/CHRG-lllhhrg576Q4/pdf/CHRG-
lllhhrg576Q4.pdf.

Stock, J., and Watson, M. 2001. "Vector Autoregressions." Journal of Economic Perspectives 15(4): 101-
115.

Tovar, C. 2008. "DSGE Models and Central Banks," Bank of International Settlements, Working Paper, #
258, Basel, September.

Townsend, R. 2010. "Financial Structure and Economic Welfare: Applied General Equilibrium
Development Economics." Annual Review of Economics 2: 507-546.

West, G. 1995. "Comparison of Input-Output, Input-Output + Econometric and Computable General
Equilibrium Impact Models at the Regional Level." Economic Systems Research 7(2): 209-227

White Force Climate Change Task Force. 1997. Peer Review of Report "Estimating the Economic Effects of
Global Climate Change Policy." Interagency Analysis Team.

13

-------
Woodford, M. 2009. "Convergence in Macroeconomics: Elements of the New Synthesis." American
Economic Journal: Macroeconomics, 1:1: 267-269.

U.S. EPA. 2010. Guidelines for Preparing Economic Analyses. Report Number EPA 240-R-10-001. Available
at: http://vosemite.epa.gov/ee/epa/eerm.nsf/vwAN/EE-0568-50.pdf/$file/EE-0568-50.pdf.

14

-------
5 Appendix

As previously mentioned, the main limitation of input-output (l-O) models, macro-econometric models,
and hybrid 1-0 macro-econometric models for social cost estimation is that they are non-optimizing
models. Since they are not based on micro-economic foundations, they cannot be used to directly
estimate changes in consumer and producer surplus resulting from the introduction of a new policy.
However, because these types of models may be of interest for estimating economic impacts of air
regulations on affected input or goods markets in the short or longer-run, we briefly describe each of
these three economy-wide models and then summarize their main advantages and limitations compared
to a CGE model. Whether these models are potentially useful for estimating economic impacts will be
discussed in detail in the economic impacts white paper.

Similar to CGE models, input-output models are based on highly disaggregated national level input-output
tables that describes the interrelated flows of good and factors of production (in terms of values of
purchases) for a particular year. However, while the information base is similar, input-output models
differ from CGE models in that they impose a simple linear model that relates changes in final demand to
changes in the total amount of goods and services, including intermediate inputs, required to meet that
demand. Input-output models are static and assume fixed prices and technology. Thus, while they can
capture immediate-term direct and indirect effects of a new policy, they do not model interactions
between sectors or feedback effects that normally occur in response to price changes (EPA, 2010). Given
this description, it is not surprising that input-output models are sometimes used to characterize the very
short run effects (before prices can adjust) of national policy (e.g., Morgenstern, Ho, and Shih, 2008).9
They are also often used to examine local or regional policies that are anticipated to have relatively small
effects relative to the national economy.

Macro-econometric models are large-scale, highly detailed systems of equations designed to predict
quarterly or annual effects of mainly fiscal or monetary policy changes on macroeconomic aggregates (e.g.
GDP, interest rates, net employment growth).10 While not explicitly derived from microeconomic theory,
they are designed to be consistent with macroeconomic theory: the short-run structure is commonly
based in Keynesian theory such that variable outcomes are demand determined. In the long run, macro-
econometric models are consistent with neoclassical growth theory in that supply side effects dominate
model outcomes (Arora, 2013a). These models combine a series of accounting relationships (e.g., savings
equal investment) with econometrically estimated relationships that are based on historical time-series
data (Hahn and Hird, 1991). The validity of their predictions is premised on the assumption that historical
relations (as reflected in the data) are valid predictors of future effects (European Commission, 2015).

1-0 macro-econometric models are hybrid models that integrate the high level of detail from an input-
output model with the forecasting properties of an econometrically estimated macroeconomic model

9	Morgenstern et al. (2008) go further by characterizing the very short-run response to a national carbon price based
on an input-output model as "a partial equilibrium view of the effects."

10	These models provide the level of detail encompassed in the National Income and Product Accounts (e.g. U.S. GDP
and its components) (Portney, 1981).

15

-------
(EPA, 2010). Unlike 1-0 models, this hybrid approach closes the model using a system of endogenous
econometric relationships between primary factors and final demand (West, 1995). By combining the two
model types, it is possible to examine the broader macroeconomic implications of industry-specific
policies. However, an 1-0 macro-econometric model may have less sectoral resolution than an input-
output model by itself and is still not derived from microeconomic theory.

Table A-l summarizes the types of costs (beyond direct effects on the regulated sector) typically captured
by the three types of economy-wide models discussed in the appendix in comparison to CGE and DSGE
models.

Table A-l: Types of Costs Captured by Economy-Wide Models

Attributes	Input- Macro-	1-0	DSGE	CGE

Output Econometric Econometric

Can estimate welfare

V	V

effects

Can be used to measure
transitional costs

V

V

V

V



Can capture indirect
effects

V

V

V

V

V

Can capture feedback
and interaction effects





Some

V

V

Advantages and Limitations

We now turn to a discussion of the key advantages and limitations of these three types of economy-wide
models based on available literature.

Input-output models are described as transparent and relatively easy to use and interpret (EPA 2010,
West, 1995). These models are capable of having a high level of sectoral disaggregation. If used in the
appropriate context, they can also provide "considerable insight into short term supply chain issues and
how industries are related" (European Commission, 2015). As previously mentioned, fixed prices may be
a valid assumption when evaluating a policy in a local or regional context when local producers are not
expected to greatly affect supply or prices outside the region (West, 1995).

The limitations of input-output models are particularly relevant when contemplating their use for
analyzing the effects of national policy. 1-0 models are static and described as lacking realistic behavioral
reactions by producers and consumers. For example, they do not include supply constraints (which are
usually transmitted via price increases) (West, 1995; Dwyer et al., 2006). 1-0 models also assume linearity
in the production system (i.e., a fixed, strictly proportional relationship between input coefficients and
output). This assumption is unrealistic if changes in relative prices cause firms to substitute away from
more expensive inputs (Dwyer et al., 2005). In addition, the model does not account for feedback effects

16

-------
between final demand and input markets (EPA, 2010; Dwyer et al., 2006; West, 1995). The government
sector is also not explicitly modeled.

As previously mentioned, 1-0 models are not based in micro-economic theory and do not estimate
changes in consumer or producer surplus. The lack of resource constraints mean that 1-0 models
frequently overestimate the economic effects of a policy. While the degree of overestimation is likely not
too large when assessing local impacts, it may be of significant size when analyzing larger regions or the
national economy (Dwyer et al., 2005; Dwyer et al., 2006; West, 1995).

Macro-econometric models are recognized as having a number of appealing features. First, they are
comprehensive, and identify effects of policy on important aggregate measures "as well as price and
output effects for at least some individual economic sectors." In addition, the "predictions they generate
are integrated and simultaneously determined....price increases in one sector are translated into cost and
price increases in other sectors" (Portney, 1981). This is a key advantage over 1-0 models that assume
away these effects.

However, macro-econometric models are described as having a number of limitations compared to CGE
models. First, they are often subject to the Lucas critique. Model predictions are based on observed
historic correlations between macroeconomic variables. However, consumer and firm responses are not
policy invariant and may change when a new policy is introduced (EU Commission, 2015; Arora, 2013).
Thus, while these models may capture short-term responses, they are less likely to capture longer run
changes. Second, while the many "feedbacks inherent in econometric models ensure at least a crude
approximation to the simultaneous and interdependent decision-making characterizing a market
economy" (Portney, 1981), macro-econometric models are still based on a top-down approach. As such,
they are frequently unable to account for interactions between different sectors as the model does not
specify how the input of one sector might be related to the output of another sector (Arora, 2013).

Third, environmental regulations may be too small to have a notable effect in these models. While
environmental control expenditures "are large in absolute terms, [they] still account for a fairly small
fraction of gross national product" (Portney, 1981). The Office of Management and Budget (OMB, 1995)
notes that macroeconomic effects tend to show up in national level macro-econometric models "only if
the economic impact of the regulation reaches 0.25 percent to 0.5 percent of Gross Domestic Product...
A regulation with a smaller aggregate effect is highly unlikely to have any measurable impact in macro-
economic terms unless it is highly focused on a particular geographic region or economic sector."

To put this in perspective, in 2014 U.S. GDP was about $17 trillion, so a regulation would have to reach
$43 billion to $85 billion in effects to show up in a macro-econometric model. There were 24 major air
regulations promulgated by EPA between fiscal years 2003 and 2013 with total annual costs of $41 billion
to $49 billion (in 2014 dollars) (OMB, 2014). Thus, if EPA had reason to examine the effects of its air
regulations in aggregate it is possible that a macro-economic approach may prove useful. However, the
most expensive individual air regulation during this time period, the Utility MACT, had annualized costs of
about $11 billion (2014$), well below the OMB threshold for use of a macro-economic model.

17

-------
Portney (1981) notes that macro-econometric models rely on pollution control expenditure (versus
pollution control cost) data as an input into analyses of policy effects on macroeconomic aggregates. This
raises two potential issues: Expenditures include transfers between sectors or economic agents (e.g., sales
taxes on equipment purchased or payroll taxes firms pay to government), which are not included in a
measure of social cost; and they do not reflect foregone opportunities - alternative uses of resources
besides pollution control.11 This leads Portney (1981) to conclude that these models may only "determine
roughly how the costs of regulation may manifest themselves in the economy" but they are "no more
than suggestive, and at times they may even fall short of even this modest goal."

1-0 macro-econometric models gain several advantages over each model type used separately. Unlike
input-output models, they account for supply-demand conditions in the economy, including resource
constraints due to coupling with a macroeconomic framework. Feedbacks between supply and demand
occur in 1-0 macro-econometric models via econometric equations (in contrast, CGE models accomplish
this via a price mechanism and market clearing assumptions) (West, 1995). In addition, the hybrid model
provides a dynamic structure for the static 1-0 model. Finally, the hybrid approach can estimate changes
in demand for and production of intermediate goods due to their coupling with a detailed input-output
model while relaxing the linearity assumption in 1-0 models.

While CGEs assume full market clearing, 1-0 macro-econometric models assume imperfect knowledge of
product and factor markets, with an emphasis on tracking short run disequilibrium adjustments over time
(West, 1995). Thus, these models can account for non-equilibrium phenomena such as less than full
employment and business cycles (EPA, 2010). That said, 1-0 macro-econometric models still retain some
of the same limitations as 1-0 and macro-econometric models. Key among them is the fact that since the
models use econometrically estimated relationships to predict economy wide aggregates, they are still
often subject to the Lucas critique. The lack of a micro-theoretic foundation and inability to reflect
behavioral responses may also limit their potential for examining long run policy implications; the
additional sectoral detail from the 1-0 component "does not alleviate [the model's] shortcomings for the
purpose of long term forecasting" (White House Climate Change Task Force, 1997).

While West (1995) notes that the distinction between 1-0 macro-econometric and CGE models has blurred
over time as hybrid models incorporate price responses into product and factor demands, several other
differences are worth emphasizing. First, 1-0 macro-econometric models have been criticized for their lack
of adequate supply-side specification (West, 1995). Second, the hybrid model is more complicated than
either an 1-0 or macro-econometric model alone; it can be difficult to disentangle the mechanisms driving
model results (EPA, 2010). Third, as previously mentioned, they do not directly estimate consumer and
producer surplus.

11 When firms increase prices to offset new expenditures, consumers may postpone or forgo purchases they would
have made at lower prices; this loss in consumer surplus (because consumers are generally willing to spend more
for goods they buy than the prices they actually pay) are not included in expenditures (Portney, 1981).

18

-------