How Will Climate Change Policies Affect Domestic Manufacturing? Joseph E. Aldy and William A. Pizer* April 28, 2012 * Aldy is affiliated with Harvard University, Resources for the Future, and the National Bureau of Economic Research. [email protected]; 617-496-7213; Harvard Kennedy School, 79 JFK Street, Mailbox 58, Cambridge, MA 02138. Pizer is affiliated with Duke University, Resources for the Future, and the National Bureau of Economic Research. [email protected]; 919-613-9286; Box 90311, Duke University, Durham, NC 27708. This research is supported by a grant from the Electric Power Research Institute.
25
Embed
How Will Climate Change Policies Affect Domestic ...mcmanusb/AppliedMicroSeminar/Pizer_20120502.pdf · How Will Climate Change Policies Affect Domestic ... How Will Climate Change
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
How Will Climate Change Policies Affect Domestic Manufacturing?
Joseph E. Aldy and William A. Pizer*
April 28, 2012
* Aldy is affiliated with Harvard University, Resources for the Future, and the National Bureau of
That is, we take the differential change between beverages and aluminum in California and Iowa, and
compare it to the differential price change in California and Iowa. This estimates the differential price
response of beverages compared to aluminum. More generally, with data for multiple states, industries,
and years, we estimate the differential price response of each industry relative to a mean price response
across industries, averaged over states and years.
There are three important features to note about the triple difference approach. First, it
addresses the potential for confounding state-time effects (the δst’s) by looking at differences among
industries within a state. While helpful in removing a likely source of endogeniety – for example, an
increase in local economic activity that both increases employment and raises electricity prices – it also
removes a potential source of variation.
The related, second important feature to note is that the mean price response is not directly
estimated, only the response relative for each industry compared to the industry average. This follows
from the fact that we do not have industry-specific energy prices. As we remove the state-time effects
by taking the difference across industries in Equation (8), this would also remove the price response
unless the coefficients β(i) differ. We can work around this by specifying that β(i)=β ei where ei is the
energy intensity of industry i. That is, we expect the responsiveness of an industry with zero energy use
to be zero, and, for industries with non-zero energy use, the responsiveness to be proportional to the
energy share. With this assumption, differences among industries are used to fit a line through the
origin that defines absolute rather than relative elasticities for each industry.
The third important feature is that we remove any confounding industry-time effects (the γit‘s)
by looking at differences across states for each industry. While this again helps remove a potentially
confounding effect – for example, declines in heavy manufacturing producing energy price declines in
regions with those industries – it also again removes a potential source of variation.
May 1, 2012 Draft; Comments Welcome; Do Not Cite
9
The second, IV approach uses both changes in global oil prices and random fluctuation in annual
weather to instrument for energy prices. Global oil prices are generally exogenous to local industrial
demand but correlated with electricity prices due to the cost of fuel oil and distillate. Similarly, random
weather fluctuations should be exogenous to industrial production and employment decisions but, as
unusually cold winters and hot summers lead to increased competition for energy, influence electricity
prices. In order to construct our instrumental variables estimator, we follow a two-step approach where
we first estimate a model
ln��� = �� + � + ����� + ���� + ��� (14)
where πs are state-specific effects, θt are time-specific effects, zst are our annual weather variables for
each state, and wt is the oil price.
In general, zst includes 24 variables reflecting the number of heating- and cooling-degree days
(HDD and CDD) for each of the twelve months. Heating degree days for a given month are the sum over
each day of the month of 65 minus the average daily temperature (using zero if the temperature is
above 65). Cooling degree days are the same calculation, except the sum of the average daily
temperature minus 75. Thus, for example, “HDD_JANst” is the number of heating degree days in
January, measured for each state and year.
Using the estimated parameters from (14), we then construct predicted energy prices,
ln���� = ��� + �� + ������ + ����� (15)
which are then used to estimate (1). Because these predicted energy prices vary only due to oil prices
and in-state weather variation, we would expect them to be unrelated to any confounding state-level
economic variation.
Data
May 1, 2012 Draft; Comments Welcome; Do Not Cite
10
For outcome measures we focus on two series: average annual employment and annual total
wages collected by the Bureau of Labor Statistics over 1990 to 2009 (future work will look out value
added and revenue measures from the Bureau of Economic Analysis). In order to be able to relate these
industries to energy intensity estimates, we match them to benchmark input-output classifications used
by the BEA. The original BLS data is available at a level of detail that defines 473 manufacturing
industries; we collapse that to 53 industries. The data cover all 50 states and the District of Columbia.
We exclude petroleum refining from our 53 sectors in all of our analysis (leaving us with 52
industries). As an energy supply sector, we can expect it to behave in fundamentally different ways
from sectors that use energy to produce other products. Petroleum refining also uses energy as a
feedstock – more than 80 percent of its costs are energy, partly feedstock, partly energy use. Finally,
even ignoring feedstock use, it is more energy intensive by a factor of two than any other industry.
While there are other potentially problematic sectors that remain to be considered, and might be
accommodated in other ways, for the current analysis we simply remove petroleum refining and treat
other industries as comparable.1
This outcome data is merged with state-level price data from the Energy Information
Administration (EIA). The EIA collects data on state-level prices for a wide range of energy products:
coal, distillate fuel, gasoline, kerosene, natural gas, electricity, etc. It also tracks separate prices for four
or five sectors of users: residential, commercial, industrial, transportation, and (where relevant) electric
power. In this analysis, we focus on electricity prices for the industrial sector users.
We focus on electricity prices because it is the overwhelming source of energy for the
manufacturing sector (>80%). Moreover, it is generally related to the local price of other fuels. But
most importantly, the most frequent target of climate change regulation is the power sector: This was
1 Other potentially problematic sectors include chemicals, which also use energy products as a feedstock, and pulp and paper mills, which often cogenerate electricity from various byproducts (such as black liquor). However, the scale of these problems tends to be considerably smaller than that of petroleum refining.
May 1, 2012 Draft; Comments Welcome; Do Not Cite
11
the focus in the EU (along with a handful of other energy-related industries), it was the focus in several
of the legislative proposals in the 112th Congress, and it is the focus of the administration’s current
proposal for a clean electricity standard. While we may eventually want to expand our consideration to
other fuels, understanding the impact of electricity only regulation and price impact is an important
starting point.
To specify the function β(i) for the triple difference approach, we assume β(i)=ψei where ei is
energy intensity. Energy intensity is measured using the 2002 benchmark input-output tables from the
Bureau of Economic Analysis. Energy is defined as inputs of oil and gas extraction (211000), coal mining
(212100), electric power generation, transmission, and distribution (221100), natural gas distribution
(221200), and petroleum refining (324110). Energy intensity is calculated as the share of these inputs at
producer prices in total costs for a given sector i. For the IV (and OLS) approach, we assume β(i)=β is
fixed.
For the IV approach, we use oil prices and state-by-month heating degree day and cooling
degree day data as a set of instruments for electricity prices. Oil prices are measured as the average
daily closing price on the New York Mercantile Exchange for West Texas Intermediary (WTI) crude. Our
oil price instrument is allowed to vary in its effect by state, allowing us to distinguish aggregate
economic trends from state-specific oil-price dependence. Data on state-level, monthly heating and
cooling degree days are from the National Oceanic and Atmospheric Administration (series HCS 5-1 and
5-2; this data is only available for the lower 48 states).
Summary statistics for the data is presented in Table 1. The first line highlights that less than
one-tenth (51/732) of the state-level (versus national) price variation arises from within-state variability
based on sum-of-square calculations. Given an overall standard deviation (of logged prices) of 0.61, the
standard deviation within states is around 16%. This is an important benchmark as ultimately we want
to explore the impact of climate change policies that might raise energy prices by around 10%. The
May 1, 2012 Draft; Comments Welcome; Do Not Cite
12
second row and third rows provide information on the outcome variables. Almost 95% of the variation
in employment and wages arises from state-industry fixed effects, with about 5% arising from within
state-industry variation (with a small amount due to aggregate year effects). The fourth row shows that
the average energy intensity, excluding petroleum refining, is 2.3%. The last two lines provide
information about the instruments (to save space, we only report statistics for HDD in January).
Preliminary Results
Our initial results for estimating β in Equation (1) are reported in Table 2. For the OLS and IV
estimators, we assume β(i)=β. Results are presented separately for the weather instruments, the oil
price instrument, and both instruments together. For the triple difference estimator, we assume
β(i)=ψei where ei is the energy intensity of sector i. We then report ���̅ in the table. We show results for
both 1990-2009 and separately for 1990-1999 and 2000-2009, for both total annual wages and average
annual employment.
The only results that are consistent across both sub-periods periods are the triple difference
estimates that are positive but indistinguishable from zero (in the 1990s where the effects are
statistically significant, they are still extremely small). This suggests that the variation left after triple
differencing may not be particularly important. Future work will look at simpler difference-in-difference
models that remove either state or industry trends, but not both.
Among the remaining estimates, the 1990s consistently show a strong positive relationship
across all 4 estimates while the 2000s show a consistent negative relationship across all 4 estimates. A
significant difference between the 1990s and the 2000s is perhaps not surprising as the 1990s were a
period of significant deregulation of power markets in the United States. The 1992 Energy Policy Act
extended the 1978 Public Utility Regulatory Policies Act (PURPA) to allow open-access to the electricity
grid for all generators. Order 888 from the Federal Energy Regulatory Commission (FERC) in the summer
May 1, 2012 Draft; Comments Welcome; Do Not Cite
13
of 1996 led to wholesale power competition throughout the United States (Brennan et al 2002; EIA
2000). Progress towards retail competition continued through the late 1990s but stagnated in the wake
of the 2001 California energy crisis (EIA 2010b).
This change in regulation raises the question of whether contracts for industrial customers were
substantially altered, whether the process of deregulation itself may be influencing the estimation, or
whether something entirely unrelated is occurring. For example, if customers in the 1990s faced a block
structure for power and paid more per unit for higher use, that could explain the positive relationship.
Or, if state-level deregulation went forward at moments when their economies were doing well, and
correlated with (short-term) price increases, that could also explain the positive relationship.
In contrast, the negative elasticity across the 2000’s is consistent across the OLS and the various
IV estimates, at about 0.2. For our purposes, the most recent behavior is the most relevant in any case.
However, it will be important to understand what is driving the changing relationship over time in order
to ensure such changes are not expected to continue and/or what assumptions are relevant.
Variation Across Industries
To be completed (all models can be estimated with various functions for β as a function of industry; may
be possible to identify particularly vulnerable industries).
Carbon Pricing Simulation
We can use these statistically-estimated relationships to simulate the effects of a $15 per ton
CO2 price from a U.S. climate change policy. Based on the Energy Information Administration (2008)
modeling of an economy-wide cap-and-trade program, such an allowance price would increase
industrial sector electricity prices by about 8 percent, which is approximately equal to a one standard
deviation increase in energy prices in our sample. We pick this price based on similar allowance prices
expected at the start of cap-and-trade programs proposed in recent legislation, including EPA’s (2009)
May 1, 2012 Draft; Comments Welcome; Do Not Cite
14
estimate of a $13 per ton CO2 price under the Waxman-Markey Bill (H.R. 2454, 111th Congress), EPA’s
(2010) estimate of a $17 per ton CO2 price under the American Power Act (draft legislation from
Senators Kerry and Lieberman) as well as the first year carbon tax of $15 per ton CO2 in a 2009
Republican-sponsored carbon tax bill (H.R. 2380, 111th Congress).
Applying our estimated -0.2 elasticity to an estimated 8 percent price impact from recent
regulatory proposals, we would expect an employment decline of 1.6 percent. However, there are a
number of reasons to expect the actual “competitiveness” effect to be smaller. This elasticity is based
on shifting production across states; shifting production across countries is more costly. This elasticity
does not differentiate between employment declines owing to higher local energy prices when other
jurisdictions remain the same, versus higher energy prices in all jurisdictions. It is the difference
between these effects – the shifting of production to other unregulated, jurisdictions – that is the real
competitiveness effect.
Whether this effect of 1.6 percent should be viewed as large or small is unclear. One question is
the relative size of the “true competitiveness” effect – the consequence of inaction in other jurisdictions
compared to the consequences if all jurisdictions pursue similar policies. If the consequence of action in
all jurisdictions were, say, a 1.2 percent domestic effect, versus a 1.6 percent effect from U.S.-only
action, we might say this is quite large. One-quarter of the domestic effect would be associated with
leakage of employment, and presumably emissions, to other jurisdictions (or at least inaction in those
jurisdictions). However, compared to overall variability in employment over time, 1.6 percent is
relatively small. For example, employment in our manufacturing sample fell from 13.8 million to 9.9
million from 2000 to 2009. Further work should contemplate this question.
(Additional work will focus on variation of impacts across industry and states, as well as
alternate policies, such as a clean electricity standard and regulation through the Clean Air Act.)
May 1, 2012 Draft; Comments Welcome; Do Not Cite
15
Conclusion and Next Steps
Concerns about the impacts of regulation on domestic manufacturing activity continue to be an
important theme in political debates surrounding policies to address climate change, particularly when
other key trade partners are unlikely to pursue similar policies in the near term. There is also an
important environmental question of whether domestic emission reductions might “leak” into other
unregulated jurisdictions; for climate change, this would undermine any environmental benefits. The
scope for these effects depends on both the energy intensity of manufacturing and the ability of
production to shift jurisdictions, as well as the scale of the regulation. This is largely an empirical
question.
We estimate these effects using a 20-year panel of employment and wage data differentiated by
state and industry, coupled with state-level energy prices. Our preliminary results suggest that the
elasticity of employment and wages with respect to electricity prices is about -0.2. Coupled with an
expected 8 percent rise in electricity prices associated with recent cap-and-trade proposals, this
elasticity predicts a 1.6 percent decline in employment.
This result is sensitive to the period of analysis: Including the 1990s leads to equally positive
estimates. However, electricity deregulation may be adversely affecting our estimation in that period.
There are also reasons to believe our -0.2 elasticity is an over-estimate for national-level impacts.
Shifting production internationally is more costly than shifting domestically (the basis of our state-level
estimation). Further, we have not distinguished overall manufacturing impacts from domestic
regulation from those arising associated with domestic regulation absent foreign regulation.
As noted throughout, these are preliminary results. Our intention is to pursue a number of
further steps to complete the project. This includes:
1. Refining our estimation procedure. We need to consider difference-in-difference estimates to
complement the triple-difference approach. We also need to do a number of statistical checks
May 1, 2012 Draft; Comments Welcome; Do Not Cite
16
on our IV approach – for example, testing for weak instruments and/or the exogeneity of prices
(Stock and Yago, 2005; Kleibergen and Paap, 2006). Our weather data – monthly heating and
cooling degree days by state – could also be combined in different ways. For example, we could
combine the data into fewer variables and allow behavior to vary by state.
2. Including additional covariates. We have not fully utilized energy intensity differences among
industries in our model. We also have data on trade and measures of “footloose-ness” that we
can include to help explain differences among industries.
3. Focusing on subsets of states and industries. We know that many states do not contribute to
manufacturing and many industries do not have significant energy use. Additional work will
look at both the sensitivity of our results to various subsamples and weighting, as well as how
estimates vary across industries and states. This analysis could also explore in more detail
deregulation in different states, and how that might be influencing our results.
4. Considering other outcome variables. We have additional data on Gross State Product (which
includes capital as well as wages) that we have not yet explored as well as regional input-output
tables. We expect this to complement our initial focus on employment.
5. Additional simulations. We intend to consider energy price increases from other climate
policies, such as proposals for a clean electricity standard and use of existing authorities under
the Clean Air Act. We also expect to construct disaggregated estimates by state and industry.
Ultimately, we hope that these results will inform the debate over climate change policy design as it re-
emerges in coming years.
May 1, 2012 Draft; Comments Welcome; Do Not Cite
17
References
Aldy, Joseph E. 2012. “Promoting Clean Energy in the American Power Sector: A Proposal for a National
Clean Energy Standard.” Environmental Law Reporter 42: 10131-10149.
Aldy, Joseph E., Alan J. Krupnick, Richard G. Newell, Ian W.H. Parry, and William A. Pizer. 2010.
“Designing Climate Mitigation Policy.” Journal of Economic Literature 48(4): 903-934.
Aldy, Joseph E. and William A. Pizer. 2011. The Competitiveness Impacts of Climate Change Mitigation
Policies. NBER Working Paper 17705.
Antweiler, Werner, Brian R. Copeland, and M. Scott Taylor. 2001. “Is Free Trade Good for the
Environment?” American Economic Review 91(4): 877-908.
Barsky, Robert B. and Lutz Kilian. 2004. “Oil and the Macroeconomy Since the 1970s.” Journal of
Economic Perspectives 18(4): 115-134.
Blanchard, Olivier J. and Jordi Gali. 2009. “The Macroeconomic Effects of Oil Shocks: Why are the 2000s
So Different from the 1970s?” International Dimensions of Monetary Policy. Jordi Gali and Mark
Gertler, eds. Chicago: University of Chicago Press.
Brennan, Timothy, Karen Palmer, and Salvadore Martinez. 2002. Implementing Electricity
Restructuring: Policies, Potholes, and Prospects. Environmental and Resource Economics 22: 99–132.
May 1, 2012 Draft; Comments Welcome; Do Not Cite
18
Bureau of Economic Analysis. n.d. Gross State Product and Personal Income Data. Internet: