5757 S. University Ave.
Chicago, IL 60637 Main: 773.702.5599
bfi.uchicago.edu
An A�liated Center of
WORKING PAPER · NO. 2019-62
Do Renewable Portfolio Standards Deliver?Michael Greenstone and Ishan NathMAY 2019
Eric Hernandez
Do Renewable Portfolio Standards Deliver?∗
Michael Greenstone†
Ishan Nath‡
May 9, 2019
Abstract
Renewable Portfolio Standards (RPS) are the largest and perhaps most popular climate policy in
the US, having been enacted by 29 states and the District of Columbia. Using the most compre-
hensive panel data set ever compiled on program characteristics and key outcomes, we compare
states that did and did not adopt RPS policies, exploiting the substantial differences in timing
of adoption. The estimates indicate that 7 years after passage of an RPS program, the required
renewable share of generation is 1.8 percentage points higher and average retail electricity prices
are 1.3 cents per kWh, or 11% higher; the comparable figures for 12 years after adoption are a 4.2
percentage point increase in renewables’ share and a price increase of 2.0 cents per kWh or 17%.
These cost estimates significantly exceed the marginal operational costs of renewables and likely
reflect costs that renewables impose on the generation system, including those associated with their
intermittency, higher transmission costs, and any stranded asset costs assigned to ratepayers. The
estimated reduction in carbon emissions is imprecise, but, together with the price results, indicates
that the cost per metric ton of CO2 abated exceeds $115 in all specifications and ranges up to
$530, making it least several times larger than conventional estimates of the social cost of carbon.
These results do not rule out the possibility that RPS policies could dynamically reduce the cost
of abatement in the future by causing improvements in renewable technology.1
∗Working paper. Comments welcome.†University of Chicago and NBER‡University of Chicago1Thank: We thank Frank Wolak, Ken Gillingham, Nancy Rose, Matt Zaragoza-Watkins, Chris Knittel, Dick
Schmalensee, James Bushnell, Koichiro Ito, and participants at the MIT Public Finance and Industrial OrganizationLunches, UC Berkeley Energy Camp, Harvard Environmental Economics Lunch, and MIT CEEPR meetings for theircomments. We also thank Catherine Che and Henry Zhang for providing excellent research assistance. Contactinformation: [emailprotected], [emailprotected].
1
1 Introduction
Even as evidence mounts on the costs of climate change, the United States has had great difficulty
developing significant and enduring policy responses, particularly in the power sector which is a
primary source of greenhouse gas emissions. One major exception has been renewable portfolio
standards (RPS) that require that a certain percentage of electricity supply in a state is met by
generation from sources that are designated as renewable. The first RPS was passed in Iowa in 1991
and since then others have followed suit. As of 2015, RPS policies have been enacted in 29 states
and the District of Columbia, which together account for 62% of electricity generation.2 Further,
the ambition of these policies has grown dramatically. In the early years of implementation, RPS
policies typically required increases in the renewables share of electricity of a couple of percentage
points, but states have greatly ramped up their ambitions, with, for example, 2030 targets of 41%
(Massachusetts), 44% (Connecticut), 50% (New York), and 60% (California). Indeed, RPS have
been credited with greatly expanding the penetration of renewable technologies, most frequently
wind and solar, which rose from 0.1% of all generation in the United States in 1990 to 5.3% in 2015
. Further, their penetration rate has increased greatly in recent years and indeed they accounted
for approximately half of the new installed capacity since 2010.3
Despite the popularity of these policies, there is little if any systematic evidence on RPS’ impacts
on electricity prices or carbon emissions. A common approach to estimating their costs is to
calculate the difference in costs associated with a RPS– that is, compare the costs of a renewable
plant with the costs of a fossil fuel plant that it replaces. This type of calculation entails comparing
the levelized cost of energy (LCOE), calculated by dividing the total direct costs associated with
investment in new capacity by expected total lifetime energy production. The latest data from
the Energy Information Administration’s Annual Energy Outlook (EIA, 2019) suggests that solar
and wind plants can produce electricity at about 6 cents per kWh, while a natural gas combined
cycle plant produces at roughly 4 cents per kWh. Since to date RPS policies have only increased
renewable penetration by a few percentage points, it is this type of comparison of LCOEs that lead
observers to suggest that RPS policies have had only a minimal impact on electricity prices; one
recent study found that they increase retail electricity bills by about 2% (see, e.g., Barbose (2018)).
However, this comparison of LCOEs misses three key ways in which renewables impose costs on
the electricity generation system that need to be covered and are reflected in retail prices but can
be difficult to observe directly or measure systematically. First, and most obviously, renewables by
their very nature are intermittent sources of electricity. Solar plants cannot provide power when
the sun doesn’t shine and wind plants cannot provide it when the wind isn’t blowing. On average,
utility scale solar plants have a capacity factor (i.e., average power generated divided by its peak
potential supply over the course of a year) of about 25% and wind plants are not much higher at
2An additional seven states enacted non-binding targets under similar programs.3This fact comes from Bushnell et al. (2017).
2
34% according to the EIA. This means that a comparison of LCOEs between these intermittent
sources and “baseload” technologies that “always” operate (e.g., natural gas combined cycle plants
have capacity factors of 85%) is very misleading with respect to total system costs, because they
do not account for the additional costs necessary to supply electricity when they are not operating.
For example, given current cost structures, the installation of renewables are frequently paired with
the construction of natural gas “peaker” plants that can quickly and relatively inexpensively cycle
up and down, depending on the the availability of the intermittent resource.
Second, renewable power plants require ample physical space, are often geographically dispersed,
and are frequently located away from population centers, all of which raises transmission costs above
those of fossil fuel plants. A literature review of transmission cost estimates for wind power by the
Lawrence Berkeley National Laboratory (LBNL) finds a median estimate of about $300 per kW,
or about 15% of overall wind capital costs (Mills et al., 2009). This is approximately equivalent to
adding 1.5 cents per kWh to the levelized cost of generation for wind. More generally, a separate
analysis by the Edison Electric Institute in 2011 found that 65% of a representative sample of
all planned transmission investments in the US over a ten-year period, totaling almost $40 billion
for 11,400 miles of new transmission lines, were primarily directed toward integrating renewable
generation.4 The highly disproportionate share of transmission requirements for renewables relative
to their share of generation highlights the importance of accounting for the associated costs as part
of the total cost of renewable energy.
Third, RPS driven increases in renewable energy penetration can also raise total energy system
costs by prematurely displacing existing productive capacity, especially in a period of flat or de-
clining electricity consumption. Adding new renewable installations, along with associated flexibly
dispatchable capacity, to a mature grid infrastructure may create a glut of installed capacity that
renders some existing baseload generation unnecessary. The costs of these “stranded assets” do not
disappear and are borne by some combination of distribution companies, generators, and ratepay-
ers. Thus, the early retirement or decreased utilization of such plants can cause retail electricity
rates to rise even while near zero marginal cost renewables are pushing down prices in the wholesale
market. The incidence of excess capacity costs on ratepayers is likely greater in regulated markets
with vertical integration, although even in deregulated markets there are various mechanisms for
direct payments to producers unconnected to actual generation that can contribute to the rates
consumers face.5 Overall, there exists no comprehensive source of data on payments to displaced
electricity producers, and even the availability of such information would not provide an obvious
path to attributing these costs to the integration of renewables. Like many of the other ancillary
4The Edison Electric Institute collected a representative sample of transmission projects totaling over $61 billionfrom their members, who cover about 70% of the total US electricity market. See EEI (2011) and Mills et al. (2009).
5For instance, ISO New England made over $1 billion of capacity market payments unconnected to actual gener-ation in 2013, comprising 12% of their total wholesale market expenditures. Over 95% of these payments supportedexisting, rather than new, capacity. The Independent System Operator for New England covers production in Con-necticut, Maine, Massachusetts, New Hampshire, Rhode Island, and Vermont. They publish capacity market informa-tion in their annual market report: https://www.iso-ne.com/static-assets/documents/2015/05/2014-amr.pdf.
3
costs of renewable energy integration, directly observing the total costs associated with stranded
capacity is unlikely to be feasible.
As an alternative to what we believe is the nearly impossible task of directly measuring each of
the mechanisms by which RPS policies influence costs, this paper compares states that did and did
not adopt RPS policies, using the most comprehensive panel data set ever compiled on program
characteristics and key outcomes from 1990-2015. Importantly, there is variation in the timing of
the adoption of RPS programs across states, which lends itself to powerful event-study style figures
that reveal no meaningful evidence of pre-existing different trends in outcomes between adopting
and non-adopting states. Further, we are able to control for a series of potentially confounding
electricity policies.
There are three key findings. First, RPS policies’ statutory requirements for renewable gener-
ation frequently overstate their net impact on generation, because they often include generation
that existed at the time of the policy’s passage. For example, six years after Minnesota adopted its
RPS policy, its statutory or total requirement was that renewables account for 14.2% of generation.
Yet at the time of adoption, renewables already accounted for 5.3% of generation. So, its net re-
quirement in this year was 8.9%. Due to the substantial heterogeneity in the form and structure of
RPS policies, it is challenging to estimate the net requirements and there is no common source for
this information. For a handful of states in our sample, even the gross requirement differs across
data sources. Nevertheless, our best estimates are that 7 years after adoption the average adopting
states’ net requirement was 1.8% of generation and 12 years after it was 4.2%.
Second, electricity prices increase substantially after RPS adoption. The estimates indicate
that in the 7th year after passage average retail electricity prices are 1.3 cents per kWh or 11%
higher, totaling about $30 billion in the RPS states. And, 12 years later they are 2.0 cents, or 17%,
higher. The estimated increases are largest in the residential sector, but there are economically
significant price increases in the commercial and industrial sectors too. These estimates are robust
to controlling for local shocks to electricity costs in a variety of ways. Given the price increases, we
also test for impacts on economic activity and fail to find any impact on electricity consumption
or state level employment. There is some evidence of a decline in manufacturing employment, but
it would not be judged statistically significant by conventional criteria.
Third, the estimates indicate that passage of RPS programs leads to reductions in the generating
mix’s carbon intensity, although these estimates can be noisier and more sensitive to specification
than is ideal. The estimated decline in emissions intensity implies a reduction of 71-250 million
metric tons of CO2 across the 29 RPS states 7 years after passage. When the CO2 emissions
estimates are combined with the estimated impact on average retail electricity prices, the cost per
metric ton of CO2 abated exceeds $115 in all specifications and can range up to $530, making it
at least several times larger than conventional estimates of the social cost of carbon (Greenstone
et al., 2013; EPA, 2016).
4
Our paper builds on previous work in the economics and engineering literatures that considers
the costs and benefits of renewable electricity generation and the impact of RPS programs in
particular. One significant line of existing research investigates how baseload, dispatchable, and
intermittent resources interact on the grid and how this affects the value of generation from the
respective sources and renewables in particular (Denholm and Margolis, 2007; Borenstein, 2008;
Lamont, 2008; Joskow, 2011; Cullen, 2013). Recent work by Gowrisankaran et al. (2016) has made
particular progress in quantifying the costs of intermittency, and their model resembles the one we
present in Section 3. This line of research in economics runs parallel to an engineering literature
that uses an energy systems modeling approach to evaluate similar questions (Milligan et al., 2011;
Jacobson et al., 2015).
The literature on RPS program impact in particular has thus far largely consisted of ex-ante im-
pact estimation. Fischer (2010) and Schmalensee (2012) document the conceptual issues underlying
the costs of these programs and Chen et al. (2007) survey pre-program prospective assessments,
often commissioned by states considering adoption. The median estimate projected that RPS
standards would raise retail prices by 0.7%, though the range of projections included significant
heterogeneity. The authors also note the importance of underlying assumptions, which focus on
capital infrastructure and fuel input costs. A limited body of post-implementation evaluations of
certain RPS programs has found slightly larger costs of approximately 2-4% (Heeter et al., 2014;
Tuerck et al., 2013), although this literature has largely taken place outside peer-reviewed journals
and generally does not account for all the ways these programs can affect system costs. An im-
portant exception to this is Upton and Snyder (2017), who use a difference-in-difference synthetic
controls framework to show that RPS programs substantially raise electricity prices and modestly
reduce emissions at the state-level.6
The paper proceeds as follows. Section 2 provides background on RPS policies and their typical
implementation. Section 3 constructs a model to explicate the channels through which integrating
renewable generation could raise costs. Section 4 outlines our data sources and presents summary
statistics on the electricity sector prior to RPS passage. Section 5 describes our empirical strategy,
and Section 6 presents and discusses the results. The paper then finishes with Interpretation and
Conclusion sections.
6Tuerck et al. (2013) and associated work by those authors also constitute exceptions to this pattern. They accountfor intermittency and other associated costs using techniques such as engineering estimates, and produce somewhathigher cost estimates of close to 5% of retail prices, though these are still smaller than the effects implied by ourestimates.
5
2 Renewable Portfolio Standards
By 2009, 29 states and the District of Columbia had adopted mandatory portfolio standards, while
an additional seven states had passed optional standards.7 These programs currently cover 62%
of electricity generation in the US. Figure 1 is a map of the United States that indicates which
states have enacted RPS programs, with the colors indicating the years of enactment. Most RPS
programs require that retail electricity suppliers meet a percentage of demand with energy from
renewable sources.8 Once in place, the standard typically increases along a predefined schedule
until a specified fraction of generation is achieved. For example, California’s policy specifies a goal
of 33% retail sales from renewables by 2020, with interim targets of 20% by 2013 and 25% by
2016. While the standards sometimes exempt certain providers, most often smaller municipal or
cooperative suppliers, they cover 82% of electric load in a state on average.9
The key feature of RPS programs is that compliance requires production from a set of designated
technologies with the frequent motivation of aiming to help spur innovation that lowers those
technologies’ costs over time. In practice, the list always includes wind and solar, but whether
other technologies are included differs from state to state. Nuclear power is excluded from the
policy in all but two states (Massachusetts and Ohio), although it is also a zero carbon energy
source.
Electricity providers must demonstrate compliance with the program through Renewable Energy
Credits, or RECs, which certify that a given unit of electricity production qualifies to meet the
standard. Most RECs are awarded by various regional authorities encompassing several states,
which issue unique serial numbers for every megawatt hour of generation produced by registered
generators. The approximate coverage of these systems is shown in Appendix Figure A.1. This
independent tracking seeks to prevent double counting of generation used for RPS compliance.
While there is some scope for transferring RECs between regional systems, in practice most RPS
compliance occurs within tracking regions, a fact we will return to later on when considering the
impact of RPS on generation outcomes.
Once awarded, credits can be sold separately from the underlying electricity, enabling flexible
transfer of the rights to environmental benefits and providing additional revenue to renewable
suppliers.10 In most cases, individual generators must be further approved by the state office
7West Virginia also passed an Alternative and Renewable Energy Portfolio Standard in 2009 with characteristicssimilar to an RPS but which we do not consider. While renewables received some preference in this program, a muchbroader set of generation sources qualified, including “Advanced Coal Technology,” and there was no guaranteedcompliance from renewable sources. This program was also repealed before its first binding requirement came intoeffect.
8Our data classify qualifying generation as one of wind, solar, biomass, geothermal, landfill gas, or ocean power,with some states also allowing small hydroelectric.
9The statistic on load covered comes from the North Carolina Clean Energy Center’s Database of State Incentivesfor Renewables & Efficiency (DSIRE).
10A minority of RPS programs have the more stringent requirement that credits be “bundled” with electricitydelivered into the state, as demonstrated by transmission to a state balancing authority.
6
administering the RPS to assure that they comply with the specific requirements for generators set
forth by that state. In restructured markets, retail providers then purchase RECs generated by these
approved facilities, either via brokers or directly through individual contracts. In non-restructured
markets, retail providers may also use RECs generated by their own renewable facilities. The
serial numbers of the RECs obtained are filed for compliance and their retirement verified with
the relevant tracking system. Depending on program rules, excess RECs may also be “banked” for
use in later years, though there are typically vintage restrictions requiring relatively recent credits
be used. Therefore, REC prices reflect the marginal costs of producing electricity from one of the
designated technologies, relative to the least expensive alternative, but they do not capture the
systemwide costs of supplying that electricity, which additionally reflect the costs associated with
intermittency, transmission, and compensating owners of stranded assets.
Most RPS programs enforce compliance using a system of Alternative Compliance Payments
(ACPs), which effectively fine retail providers for failing to acquire sufficient RECs to cover their
sales. These payments are large, averaging about $50 per MWh.11 Such penalties are substantial,
representing about half of the average revenue per MWh observed in 2011. In addition to a penalty,
ACPs also provide an effective cost-ceiling for the REC market, as they provide an outside option for
compliance. While in practice few retail suppliers fulfill their requirements through ACP payments,
REC markets in some states have periodically traded at the ACP level, suggesting that marginal
sources of compliance can be relatively high cost.
While statutory requirements like Maine’s 40% target appear quite large, they often ramp
up gradually from lower levels and may not reflect the amount of marginal generation actually
mandated by RPS policies. Intuitively, if an RPS requirement were entirely covered by existing
sources at its inception, in a competitive market we would expect producers to bid down the price
of RECs to zero. Distinguishing the amount of new renewable generation required to comply with
RPS policy is quite difficult in practice, since covered sources of generation vary from state to state
even within narrowly defined categories. For instance, some states allow small-scale hydropower
but not large-scale hydropower to qualify for their RPS. Further, six states, including Maine,
explicitly mandate that part of their RPS be met using newly constructed renewable capacity. Our
best estimate of the “net” requirement imposed by RPS policies takes the gross amount of MWh
required for RPS compliance, as reported by LBNL, and subtracts existing generation from the
broad categories of covered sources in the year prior to RPS passage.
Figure 3 reports each states’s total and net requirements as of seven years after the state passed
RPS legislation, ordering states by the calendar year in which they first adopted an RPS. While
these numbers do not fully account for the complications described above, they do show a clear
pattern of statutory requirements overstating the amount actually necessary to achieve compliance.
For instance, seven event years after passage, the gross requirement in Michigan is 6.2%, but the
11In the case of mandates for generation specifically from solar energy, they can climb even higher, sometimesexceeding $400 per MWh.
7
net requirement after subtracting existing generation in the year of passage is only 2.6%. On
average, seven event years after RPS passage, RPS states have a total requirement of 5.1%, but a
substantially lower net requirement of 1.8%. In the remainder of the paper, we primarily focus on
estimates of net requirements, described in greater detail in Section 4.1.
Figure 2 plots the number of RPS programs passed into law in each year.12 The majority of
programs were not passed until after 2000. While a number of states adopted RPS policies during,
or subsequent to, broader electricity market restructuring, RPS programs have also been adopted
in a number of traditionally regulated markets. Figure 2 also plots real national average retail
electricity prices (right y-axis) which declined from about 12 cents per kWh to 10 cents per kWh
from 1990 through 2002 but by the end of the sample in 2015 returned to 12 cents per kWh.13 This
break in the decline in prices and subsequent upwards turn loosely corresponds with the number
of states that passed RPS programs in those years. Whether this relationship is causal will be
examined in much greater detail below.
3 Conceptual Framework
As discussed above, standard LCOE estimates measuring the direct capital and maintenance costs
of various generation sources provide an incomplete summary of the impact of transitioning elec-
tricity production to renewable sources on consumer prices. We set out a simplified model of the
decision-making process of a retail electricity provider to illustrate the mechanisms through which
renewable integration can raise costs, and consequently retail prices. The model demonstrates how
intermittency, transmission, and the displacement of existing capacity infrastructure interact to
raise the total costs incurred by a utility. Notably, the model highlights the wide range of parame-
ters and nontransparent data inputs that would be required to calculate these costs directly. The
paper’s empirical procedure sidesteps this difficulty by summarizing the aggregate effect of these
mechanisms through the reduced-form impact of RPS standards on retail electricity prices.
For simplicity, the model assumes a vertically integrated setting with a single utility respon-
sible for both power capacity and retail provision. The intuition from this framework translates
straightforwardly to a deregulated setting with a retail provider purchasing electricity from com-
peting generators, except for the assumption that ratepayers always pay the full cost of installed
capacity. As discussed below, the extent to which owners of capital bear the losses from excess
capacity stranded by integrating renewable sources will be one factor that contributes to the overall
effect on retail prices.
12Iowa was the first state to establish a binding standard in 1991, requiring the states’s two investor-owned utilitiesto build or contract for 105 MW of renewable capacity. Although Iowa originally enacted an Alternative Energy Lawin 1983, the program wasn’t given a concrete goal or made compulsory until a revision in 1991, so we consider thatthe first year of passage.
13All monetary figures are reported in January 2019 dollars.
8
3.1 Representative Utility Model
A representative utility chooses capacity investments and daily generation sources to fulfill two
requirements: ensuring that they meet the full electricity demand of their customers every day
and that their annual electricity production meets the RPS requirement.14 Utilities have three
types of production capacity available with which to meet daily electricity demand: renewables, R,
baseload power, B, and dispatchable “peaker” plants, D, the latter two of which we assume come
from non-renewable sources. Baseload generation produces a constant daily amount governed by
annual capacity, Bt, and cannot be adjusted in response to daily demand. Renewable generation is
stochastic and drawn from a distribution F (R), with mean, R, standard deviation, σR, and support
[R, R]. F (R) is a function of installed renewable capacity, Rt. The daily demand for electricity is
also drawn from a distribution, G(E), with mean E, standard deviation σE , and support [E, E].
So given the available capacity of Bt, Rt, and Dt in year t, the utility observes the daily draws of
Es and Rs and chooses the level of dispatchable power, Ds, to satisfy customer demand.
Es = Bt +Rs +Ds, (1)
Es ∼ G(Et), Rs ∼ F (Rt).
With knowledge of this daily optimization problem, the utility chooses investment in new ca-
pacity at the beginning of each year. Total capacity in period t consists of the depreciated capital
from last period plus new investments in each of the three categories of electricity sources:
Ct = Bt−1(1− δB) +Rt−1(1− δR) +Dt−1(1− δD) + IB + IR + IP . (2)
The utility chooses annual investments in new capacity to fulfill its two primary requirements.
First, the RPS requirement dictates the proportion of annual electricity production that must come
from renewables. For mandated renewable percentage, M, the utility must satisfy the following:∑365s=1Rs∑365s=1Es
≥M. (3)
Under RPS requirements, failure to meet this condition will cost the utility a per-unit fine,
f, for the amount by which renewable generation falls below the threshold. To avoid paying the
fine, utilities must have enough installed renewable capacity, Rt, to produce enough electricity to
meet this requirement. Determining what constitutes enough renewable capacity also may not be
straightforward. If draws from the F (R) distribution are correlated across days, simply ensuring
that E[Rs]E[Es]
= M might not be sufficient to ensure compliance with the RPS mandate in a year
with systematically low realizations for renewable generation. The utility will trade off the cost of
14The period in which instantaneous demand must be met can equivalently be thought of as an hour rather thana day.
9
increasing renewable capacity, Rt, with investments, IR, against the fine for noncompliance when
making their choice over optimal Rt.
Second, the utility must ensure it can supply enough energy every day of the year. We assume
there is an infinite penalty for failing to meet demand. Since both energy demand and renewable
production are stochastic, the utility must have enough dispatchable generation available to fill
the largest possible daily need. In particular, the utility chooses Dt such that it can meet total
electricity needs on a hypothetical day with the highest possible demand draw, E, and the lowest
possible renewable generation draw, R.
Dt = E −Bt −R. (4)
In addition to choosing investment, the utility also has the option to prematurely retire capacity
at the beginning of each period. The carrying costs of retired capacity are lower and for simplicity
we assume that capacity that has not been retired will be run. Under certain conditions, they may
choose to retire baseload capacity because too much baseload generation could prevent the utility
from meeting the RPS requirement. If BtE[Es]
> 1 −M , for instance, then renewable production
would be expected not to meet its mandate even without any dispatchable production. To ensure
compliance with the RPS mandate, the utility must estimate the amount of dispatchable production
necessary during the year and then scale back Bt such that the expected sum of baseload and
dispatchable generation does not exceed 1−M as a proportion of all production.
Total costs for the utility include the fixed costs of installed capacity, associated transmission
and distribution requirements, and the variable costs associated with each type of power. The
utility finances new investments such that they make a constant annual payment over a horizon
of T years. The annualized prices of installed capacity, pB, pR, and pD, incorporate differences
in the cost per MWh for baseload, dispatchable, and renewable sources, as well as any differences
in financing costs or investment tax incentives. New transmission investments in each period,
which are also financed over a T-year horizon with annualized payment pT , are a function of new
installations across the three categories and depreciation of the existing transmission capital stock,
with geographically dispersed renewable installations such as wind and solar likely having greater
associated requirements. Since renewables require no fuel inputs, they incur no variable costs
whereas baseload and dispatchable power have average costs acB and acP for each unit generated.
For the purposes of this model, these average costs capture not only the cost of fuel inputs, but also
any startup and shutdown costs associated with the operation of these generating sources. Thus,
the utility’s total costs in period t are as follows:
10
TCt =t∑
k=t−TpBkIBk +
t∑k=t−T
pDkIDk +t∑
k=t−TpRkIRk
+
t∑k=t−T
pTkTr(IRk, IBk, IDk) + 365BtacB +
365∑s=1
DsacD. (5)
The retail rate is given by total costs in period t divided by total kilowatt-hours of energy produced
plus a markup, µ, assigned by the regulator. Thus:
Retail Rate in Year t = (1 + µ)TCt∑365s=1Est
. (6)
3.2 Empirical Requirements for Estimating the Full Costs of a
RPS
This framework illustrates the major practical difficulties involved in developing the costs of RPS
programs piece-by-piece. This simplified model reveals that even if renewable and non-renewable
production have the same LCOE, defined by the prices of installed capacity and fuel inputs, tran-
sitioning a mature grid infrastructure could increase costs through a wide variety of channels. The
list of excess costs includes:
• investments in new dispatchable capacity to protect against shortfalls of intermittent renew-
able generation,
• investments in new transmission infrastructure to accommodate the geographic locations of
new renewable capacity,
• premature retirements of baseload capacity and/or transmission infrastructure that serves
nonrenewables to reduce nonrenewable production enough to meet RPS mandates.
Further, the incidence of this last category between ratepayers and owners of capital is unclear ex
ante, although ratepayers seem more likely to bear the costs in traditional regulated “cost-plus”
markets, compared to restructured ones. Regardless of the ultimate incidence, these costs are part
of the full costs of the introduction of a RPS program. However, it is worth noting that this
last category category is “transitional” in nature, while the first two are permanent features of
increasing renewables’ share of production.
It is instructive to consider the challenges with constructing a bottom-up or structural estimate
of the costs of an RPS policy. First, it would require data or estimates of several moments from
the distributions of daily energy demand, G(Et), and daily renewable generation, F (Rt), the pre-
existing level of installed capacity by generation type, Bt, Dt, Rt, the respective depreciation rates,
11
investment prices, and fuel input prices for each of these three energy categories, and the transmis-
sion investments necessary to incorporate renewable capacity. Second, the estimates would need to
make a series of assumptions for how utilities project electricity demand, renewable intermittency,
the need for dispatchable generation to protect against insufficient or excess supply, as well as the
decision criteria for retiring baseload generation. Third, estimating the model would require going
beyond the representative utility setup and incorporating interactions between heterogeneous gen-
erators and retail providers in restructured and non-restructured markets; these interactions have
proven to be quite complex to model as they involve questions of market power and doing so in
this context would undoubtedly be both a great research topic and a difficult problem to solve.
Fourth, the incidence of these costs between ratepayers and owners of capital is also a complicated
question and, as we noted above, is likely affected by the regulatory environment.
Our approach circumvents this complex interplay of underlying mechanisms with a reduced-form
approach that captures the costs imposed on ratepayers due to all potential mechanisms through
which RPS policies raise costs. If generators or distributors bear part of the costs, our approach
will not capture the full social costs of RPS policies.
Finally, we note that coincident to the increase in the number of RPS programs and the scope
of their ambitions, there have been important changes in the operation of electricity markets. As
one example, several Regional Transmission Organizations began holding centralized auctions for
capacity market payments in the mid-2000s as RPS programs began to proliferate.15 Since their
initiation, these payments comprise a substantial fraction of overall market costs - reaching 9-28%
of total costs in ISO New England, the New York ISO, and the PJM Interconnection between
2008 and 2016 (GAO, 2017). These three RTOs cover all or part of 15 of the 29 RPS-adopting
states in our main sample, and the Mid-continent Independent System Operator (MISO) added
a capacity market auction covering 4 more RPS states in 2013. Further, (Bushnell et al., 2017)
document that similar payments to maintain “Resource Adequacy” take place in other locations as
well and that they likely also preceded the centralized auctions in those four RTOs. We take the
significant share of these types of payments for generator availability after RPS implementation,
be it through auctions or resource adequacy payments, as suggestive evidence that RPS program’s
mandated increase in intermittent renewable generation imposes systemwide costs on electricity
markets, very likely due to these technologies’ intermittency.
Thus, it is at least plausible that an important share of RPS programs’ total costs comes from
the indirect costs that they impose on the electricity supply system. These costs are not evident
from a simple comparison of LCOEs or RECs prices. Of course, the qualitative evidence about
the growth of capacity markets or resource adequacy payments is not decisive and could be due
to other factors, so the remainder of the paper exploits a differences in differences research design
that is generated by the staggered adoption of RPS programs by some states and the non-adoption
15ISO-NY began their current system of auctions in 2003, PJM in 2004, NE-ISO in 2007, and MISO in 2013.
12
by other states.
4 Data Sources and Summary Statistics
In order to assess the retail price and other impacts of RPS programs, we construct a state-level
panel from 1990 to 2015 with data on RPS programs, electricity prices, generation capacity and
outcomes, and CO2 emissions. We believe this is the most comprehensive data set ever compiled
on RPS program characteristics and potential outcomes. This section describes each data source
and presents some summary statistics describing the context of the policy.
4.1 RPS Program Data
Since 1990, 29 states and the District of Columbia have adopted RPS programs. We construct
indicators for the year in which legislation for a mandatory RPS program first passed in each state,
compiled using state legislative documents, state government websites, and summaries from the
U.S. Department of Energy. While there is typically a few years’ lag between policy enactment
and the onset of binding mandates for renewable generation, costs to electricity providers, and
consequently customers, are likely to begin accruing when they start planning for and investing
in the required future capacity. Data from the Lawrence Berkeley National Laboratory (LBNL)
also include information about qualifying renewable sources under each program, including whether
there are specific requirements for solar generation.
To better characterize each state’s implementation, we also collect more detailed information on
year-by-year requirements. Most RPS programs require an increasing percentage of electricity sales
to come from renewable sources, leading to increased stringency over time.16 However, as mentioned
earlier, the statutory percentage requirement may overstate the additional generation required if a
large number of existing generators are eligible for compliance. To account for this, we construct a
measure of RPS net requirements as the difference between statutory requirements and pre-existing
renewable generation. We collect data from LBNL on total generation required from renewables
in each RPS state in each year of enforcement (Barbose, 2018), and define pre-existing compliance
as total generation from qualifying categories of renewables in the year before RPS legislation was
passed. The difference is the amount by which each state had to expand renewable generation
to comply with the policy - our measure of net requirements.17 Recall, Figure 3 highlighted the
substantial differences between the total and net requirements.
16Iowa and Texas have fixed capacity requirements for new renewable generation, which will tend to decreasestringency over time if demand is increasing.
17Some states include waste-to-energy and similar forms of power generation in their RPS, but we do not have datafor these sources, so we cannot account for these in our net generation estimates. In addition, there are 3 states -California, Montana, and Minnesota - for which the gross MWh requirements reported by LBNL differ by more than3 percentage points from the statutory percentages reported by DSIRE.
13
In addition to data on RPS programs, we also collect information from the North Carolina Clean
Energy Center’s Database of State Incentives for Renewables & Efficiency (DSIRE) on the presence
of other state programs that may influence the amount of renewable generation and the retail price
of electricity (Barnes, 2014). We have information on the implementation dates of three types of
programs: net metering, which pays consumers for electricity they add to the grid with distributed
generation such as solar PV, green power purchasing, which gives consumers the option of paying
to have renewable energy account for a certain percentage of their consumption, and public benefits
funds, which place a surcharge on retail electricity prices to fund programs such as research and
development, energy efficiency investments, and low-income energy assistance. This information is
used to account for the presence of potentially confounding programs.
4.2 Electricity Sector
Information on electricity sector variables is drawn from Energy Information Administration (EIA)
survey forms. Electricity prices are computed from EIA Form 861, a mandatory census of retail
sales by electric power industry participants.18 Respondents report sales and revenues separately
for commercial, industrial, and residential sectors. Average price is then computed based on average
revenue per megawatt-hour sold for each sector and for total retail sales.
Electricity generation by state and fuel source is compiled from EIA forms 906, 920, and 923,
which concern power plant operations. This data is broken down by fuel type, ensuring plants with
multiple fuel sources are accurately reflected in aggregate numbers. Generating capacity by state
and fuel source is compiled from EIA Forms 860 and 867, along with starting year and month and
location. These surveys cover all grid-connected generators larger than 1 MW in capacity currently
able to deliver power. For simplicity, we aggregate the EIA’s fuel type categories, measuring
generation by hydroelectric, solar, wind, coal, natural gas, nuclear, other renewables, and other
fuels.19
To measure CO2 emissions, we use estimates derived by the EIA from power plant operations
data taken from forms 767, 906, and 923. Their estimation process involves converting fuel use to
BTUs to provide a common comparison measure. Next, fuel uses that do not generate emissions are
subtracted out. Finally, source-specific carbon emission coefficients are used to convert to metric
tons of carbon.20 The result is a yearly panel of state emissions from electricity generation.
As part of our analysis, we also attempt to look at the difference between RPS impacts in
18The 3,300 respondents cover essentially the universe of retail suppliers, including electric utilities, energy serviceproviders, power marketers, and other electric power suppliers.
19“Other Renewables” includes biomass, geothermal, and wood-based fuels, while “Other” covers remaining sources,including pumped storage, blast furnace gas, and other marginal fuels. See the Electric Power Monthly published bythe EIA for a full accounting of possible disaggregated fuel sources.
20More details on this process, including the conversion factors used, can be found in “Methodology and Sources”section of the Monthly Electric Review published by the EIA.
14
regulated versus deregulated markets. Using data compiled for an earlier paper by Fabrizio et al.
(2007), we code an indicator for whether or not a state ever deregulates their electricity market,
defined by retail market access for non-utility-owned generation plants.21
4.3 Manufacturing Employment
If RPS programs do in fact raise electricity prices, there may be downstream impacts on industries
for which energy is a large input to production. To assess this, we construct a panel of employment
in each state by industry code using data from the County Business Patterns (CBP). One issue
with these data is that employment statistics are often suppressed when the industry code and
establishment size potentially disclose information about a specific business. Following previous
papers, we apply an imputation procedure to estimate employment for these cells, using the national
average for the industry in that cell size. To allow comparisons across years, we recode NAICS
industry codes used in later years to SIC industry codes, redistributing employment proportionally
based on concordances provided by the census.22 We then calculate total and manufacturing
employment for each state in each year.
4.4 Summary Statistics
Before describing our empirical approach in detail, we briefly present some summary statistics
from the data and report on some comparisons of treatment and control states in the year prior to
RPS passage. Table 1 presents summary statistics for treatment states, defined as those in which
legislation passes in the following year, and control states, which consist of states that did not pass
RPS by 2015. The summary statistics for control states are averaged across the set of control states
that correspond to each RPS state’s year of passage.
The statistics in Table 1 show some level differences between RPS states and control states
in the year prior to legislation. RPS states tend to have somewhat more expensive electricity
— 11.4 cents per kWh versus 9.4 in control states — larger populations, and better resources for
producing solar and wind energy. Such level differences do not threaten the identification of our
difference-in-differences design, but may be informative about the degree to which our results would
be representative of the impact of a national RPS policy. The RPS states in our analysis are also
more likely to have other simultaneous programs affecting renewable energy, including public benefit
funds, net metering, and green power purchasing programs. We control for the time-varying passage
of these programs at the state by year level in our analysis. Finally, we note that the pre-existing
trends of electricity prices in treatment and control states are similar, with an average six-year
21We thank Fabrizio, Rose, and Wolfram for generously sharing this data.22For further details, and code used, see Autor et al. (2013) and the accompanying data files. For 2012 and 2013,
where official concordances are unavailable, we allocate employment proportionally based on 2011 employment usingthe official code mapping 2012 to 2007 NAICS.
15
decrease in electricity prices of 0.6 cents per kWh in both RPS states and control states prior to
the year of passage. Our analysis in the next section will control for differences in pre-trends, but
the similarity of these trends lends validity to the key identification assumption of equal trends in
electricity prices in RPS and non-RPS states in the years before RPS passage.
5 Empirical Strategy
Our empirical approach begins with an event study-style equation:
yst = α+18∑
τ=−19
στDτ,st +Xst + γs + µt + εst, (7)
where yst is an outcome of interest in state s in year t. We include state fixed effects γs to control
for any permanent, unobserved differences across states. Year fixed effects, µt, non-parametrically
control for national trends in retail prices. The variables Dτ,st are separate indicators for each year
τ relative to the passage of a RPS law, where τ is normalized to equal zero in the year that the
program passed; they range from -19 through 18, which covers the full range of values of the τ ’s.23
For states that never adopt an RPS program, all Dτ,st are set equal to zero. As non-adopters, they
do not play a role in the estimation of the τ ’s but they aid in the estimation of the year effects, µt,
as well as the constant, α.
The στ ’s are the parameters of interest as they report the annual mean of the outcome variable
in event time, after adjustment for state and year fixed effects. An appealing feature of this design
is that because states passed RPS programs into law in different calendar years, it is possible to
separately identify the στ ’s and the year fixed effects µt. In the remainder of the analysis, we will
particularly focus on the στ ’s that range from -7 through 6. This is the maximum range for which
the στ ’s can all be estimated from a fixed set of states. Restricting the treatment period in this
way holds the advantage of eliminating questions about the role that differences in the composition
of states identifying the various στ ’s plays. This range is determined by Nevada, which passed its
law in 1997 on one side of the range, and Kansas, which passed its law in 2009 on the other side of
the range. We will present event-study figures that plot the estimated στ ’s against τ . These figures
provide an opportunity to visually assess whether there are differential trends in the outcome
variables prior to RPS passage, which help to assess the validity of the difference in differences
identification strategy. The event-study figures also demonstrate whether any impacts on outcomes
emerge immediately or over time, which will inform the choice of specification to summarize the
average effect of RPS policies.
To summarize the information contained in the event-study plots and formally assess the pro-
23Iowa adopted a RPS in 1991, which means that only one pre-RPS year is available. Consequently, we drop Iowafrom the primary sample although its inclusion does not alter the qualitative findings.
16
gram impact, we estimate two equations. In the first, we assume that the difference in differences’
identification assumption of parallel trends holds and allow for RPS programs to have only a mean-
shift effect on retail electricity price:
yst = δ0 + δ11(−19 ≤ τ ≤ −8)st ∗ 1(RPS = 1)s + δ21(7 ≤ τ ≤ 18)st ∗ 1(RPS = 1)s
+ δ31(0 ≤ τ ≤ 6)st ∗ 1(RPS = 1)s +Xst + γs + µt + εst. (8)
Here, the parameter of interest is δ3, which measures the mean of the outcome variable in the first
7 years after the passage of RPS policies, in RPS states, relative to the preceding 7 years, after
adjustment for state and year fixed effects. The coefficients δ1 and δ2 measure the mean of the
outcome in the unbalanced samples in the years before and after the 14 year period where the
sample is balanced, in RPS states. These are nuisance parameters.
Most RPS programs have requirements that increase gradually over time after legislation is
passed, so it is likely that the impact on electricity prices will increase correspondingly. Therefore,
a specification like a trend break model seems better equipped to summarize the effect of RPS pro-
grams on outcomes because it allows the programs’ effect to grow over time. Further, specifications
that allow for the possibility of differences in pre-adoption trends require weaker assumptions to
produce valid estimates of the impact of RPS programs. For these reasons, we also fit an equation
that allows for differential trends before and after RPS programs are passed into law:
yst = δ0 + δ11(−19 ≤ τ ≤ −8)st ∗ 1(RPS = 1)s + δ21(7 ≤ τ ≤ 18)st ∗ 1(RPS = 1)s
+ δ31(0 ≤ τ ≤ 6)st ∗ 1(RPS = 1)s + β0τst + β11(−19 ≤ τ ≤ −8)st ∗ 1(RPS = 1)s ∗ τst+ β21(7 ≤ τ ≤ 18)st ∗ 1(RPS = 1)s ∗ τst + β31(0 ≤ τ ≤ 6)st ∗ 1(RPS = 1)s ∗ τst+Xst + γs + µt + εst. (9)
To summarize the cumulative effects, we calculate and report the impact seven years after RPS
passage, which is given by δ3 + 6β3. Finally, we report standard errors that are clustered by state
from the estimation of Equations (8) and (9) to allow for correlation in the errors within state over
time.
6 Results
6.1 Net RPS Requirements and Retail Electricity Prices
We begin with an examination of the net RPS requirements. Figure 4a plots the event-year means
of net RPS requirements against τ . Recall that event time is normalized so that the program
passage year occurs at τ = 0 and the vertical line at τ = −1 indicates the last year before program
passage. It is apparent that the RPS programs’ passage into law leads to increases in the required
17
use of the RPS technologies that begin almost immediately and increase every year. Seven years
after passage, the average RPS state’s net requirement is 1.8 percentage points of generation. It is
noteworthy that this is substantially smaller than the increase in total or gross requirement which
is 5.1%. through the end of the balanced sample which is 7 years later.
Figure 4b reports on the estimation of equation (7) for the average retail price. where prices
are normalized so that they equal zero at τ = −1. Recall, the estimated στ ’s are adjusted for
state and year fixed effects. There are two primary points that emerge. First, there is no evidence
of a meaningful difference in the trends of prices, either upwards or downwards, among adopting
states in the six years preceding RPS programs becoming law, from τ = −7 to τ = −1. Thus,
for example, there doesn’t appear to be any evidence that prior to RPS passage, adopting states
were differentially passing other policies that influence electricity prices positively or negatively or
facing differential cost shocks. More broadly, this figure supports the validity of the difference in
differences research design. Second, it is apparent that retail prices increased after program passage,
but not all at once; the figure suggests that a model that allows for a trend break describes the
data well. It is striking that the trend in prices appears to very closely shadow the trend in net
RPS requirements.
Columns (1a) and (1b) in Panel A of Table 2 present results from the estimation of equations (8)
and (9) that confirm the visual impression that retail electricity prices increase after RPS programs
become law. The mean-shift specification suggests that RPS programs raised prices by 0.5 cents
on average in their first 7 years. In the mean shift and trend-break model, the estimates indicate
that retail prices in RPS states rise by roughly 0.16 cents each year post-passage, with statistically
insignificant pre-trends and post-passage mean-shift.
Given these results and the visual event-study evidence suggesting that RPS programs affect the
trend in prices, we treat Equation (9) as our primary specification. We focus on the effect 7 years
after RPS passage, which is calculated as δ3+6β3. Overall, the estimates from this regression suggest
that RPS policies have increased retail electricity prices by about 1.3 cents per kWh seven years
after passage. This increase is statistically significant and economically substantial, representing
an increase of about 11.1% over the mean retail price at τ = −1. Such a large increase in the retail
price of electricity is striking, given the modest net requirements 7 years after passage. Further,
these estimates are much larger than LCOE differences alone would suggest, indicating that the
indirect costs of RPS mandates are an important component of their total costs.
We next consider whether RPS policies exhibit heterogeneous effects by the category of customer.
The EIA divides retail sales among three sectors, residential, commercial, and industrial, that
together account for total retail sales.24 Residential is the largest sector for most years in our data,
24According the EIA, the sectors are composed of:
• Residential: “living quarters for private households,”
• Commercial: “service-providing facilities and equipment of: businesses; Federal, State, and local governments;and other private and public organizations,”
18
comprising about 37% of sales in 2015.25 On a per-customer basis, though, the commercial and
industrial consume significantly more. A typical commercial customer uses nearly seven times the
typical residential consumption, while the typical industrial customer uses more than 120 times the
typical residential consumption. As noted in Table 1, retail rates also vary among these groups,
with residential customers paying the highest rates while industrial customers pay the lowest. This
differentiated pricing may reflect demand elasticities that are correlated with usage, leading utilities
to price discriminate by charging lower prices to their most intensive, and therefore price sensitive,
customers (Bjørner et al., 2001).
The event-study figures derived from the fitting of Equation (7) for these outcomes are presented
in Appendix Figure A.4. There is little evidence of difference in trends between adopting and non-
adopting states prior to RPS passage. Industrial prices appear to shift upwards substantially in the
first year after passage, while the commercial and residential sectors adjust more gradually. Overall,
changes by sector track closely with net requirement changes, though perhaps with a slight lag.
The statistical sectoral price analyses for the balanced sample are reported in columns (2) - (4)
of Panel A in Table 2. As in our analysis of total prices, sectoral prices appear best captured by the
mean-shift and trend-break model, so we focus on estimates from Equation (9) in the (2b), (3b),
and (4b) columns. In all three sectors, the point estimates represent substantial price increases
in the first 7 years after RPS passage; they are 12.8% for residential, 7.7% for commercial, and
9.2% for industrial, although only the residential one would be judged statistically significant by
conventional criteria.
The appeal of the Panel A results is that there is a balanced sample for all event years, but
this sample restriction limits the number of post-years. In Panel B, we extend the post-period
through τ = 11 which allows us to estimate the effect of the RPS programs through 12 years after
passage. However, the number of RPS states that reach τ = 11 in the sample declines from 29 in
the balanced sample to 16, so the cost is that there is not a constant sample of states for all event
years.
The Panel B results tell much the same story of higher prices. As RPS programs are in force
longer here, their net requirements increase and their impact on electricity prices also increases.
The column (1b) estimates indicate that at twelve years after passage, the average retail price has
increased by 2.0 cents per kWh or 17% and at the same point net RPS requirements have risen to
4.2 percentage points of generation (although gross or total RPS requirements are higher at 11.1
percentage points).26 The remaining columns reveal that over this longer time horizon the higher
electricity costs remain evident in all three sectors, with the residential sector experiencing the
• Industrial: “all facilities and equipment used for producing, processing, or assembling goods.”
For complete definitions, see the EIA’s Electric Power Monthly.25Authors’ calculation, from the EIA Electricity Data Browser.26Appendix Figures A.2a and A.2b present the accompanying extended period figures for net requirements and
average retail prices. See Appendix Figure A.3 for a plot of gross, i.e. total, RPS requirements.
19
largest increase.
Table 3 explores the robustness of the Table 2 Panel A results to a variety of changes in Equation
(9). In column (1), we drop the two states with nuclear in their original RPS (i.e., Massachusetts
and Ohio) as these states’ policies are closer to a zero carbon energy standard and in column (2) we
drop Hawaii due to its unique geography. Neither of these sample restrictions meaningful change
the qualitative findings. The remaining columns aim to adjust for the possibility of local shocks to
electricity prices that might confound the adoption of RPS programs; specifically, columns (3) and
(4) include year by census region and year by census division fixed effects, respectively. There are
4 Census regions and 9 Census divisions. The estimated increases in electricity prices are modestly
smaller here than in Table 2, however the differences are small compared to the standard errors.
Our conclusion is that these models that handle local shocks more flexibly leave the qualitative
findings unchanged.
6.2 Heterogeneity in RPS Price Effects
To this point, we have assumed that the effect of RPS programs on average retail prices are constant
across states. However, there are several important characteristics that might differ across states
and could affect the magnitude of the impact of RPS programs or their incidence on ratepayers
versus owners of capital. This subsection explores this possibility by taking the trend and mean
shift model (i.e., Equation (9)) and fully interacting it with an indicator for membership in a
subsample of interest. Table 4 presents the results from this exercise for average retail prices and
residential retail prices by reporting the effect of RPS programs among states not in the subsample
and the marginal effect for the subsample. The latter estimate tests whether the seven year effect
differs in the subgroup of interest and the full effect for this group is the sum of the two reported
estimates.
Panel A examines whether RPS program effects differ for late adopters, defined as those with
laws that were passed after 2004, the median year of passage in the data. This specification tests the
hypothesis that the costs of RPS programs might be lower in the later years of the sample, perhaps
due to decreasing costs for renewable energy or learning about how to more efficiently integrate
renewables into the grid. Panel B explores differential impacts among states that have restructured
electricity markets. Panel C examines the effect of setting specific requirements that can only be
fulfilled by solar energy, which restricts flexibility to use the cheapest available renewable resource.
Panel D estimates effects for “heavy coal” states, defined as those above median percentage coal
generation in 1990, to test whether these states encounter higher costs to incorporating renewables.
This type of subgroup analysis is very demanding of the data, but some intriguing, albeit sugges-
tive, patterns emerge. There is little evidence to support the hypothesis that the costs for ratepayers
were lower in late (i.e., post-2004) adopting states. The point estimates in Panel B indicate that
the impact on prices is smaller in states where electricity markets have been restructured, which is
20
consistent with the possibility that it is easier to pass on the costs of stranded assets to ratepayers
in vertically integrated non-restructured settings. However, the magnitude of the standard errors
warrants caution in drawing strong conclusions. The point estimates suggest that solar set asides
substantially increase prices, which is consistent with the fact that solar REC prices can be several
times larger than general REC prices, but here too the imprecision of the estimates tempers the
strength of any conclusions. Finally, the costs appear to be higher in heavy coal states, but the
same problem of noisy estimates is evident.
6.3 Economic Activity
Since the estimates suggest that RPS programs lead to substantial increases in electricity prices, it
is natural to examine whether there are impacts on the real economy. We begin by testing whether
electricity consumption responds to these increases. Some previous studies suggest that consumers
typically appear to be responsive to the average prices, which is our variable of interest, rather than
marginal prices, potentially due to inadequate real-time information about current consumption
(Borenstein, 2009; Ito, 2014). In columns (1a) and (1b), there is little evidence of a change in
electricity consumption.
The remaining columns of Table 5 report on the estimation of the same equations for total
employment and manufacturing employment. Energy costs are a relatively high share of total
costs in manufacturing. There is little evidence of an impact on overall employment as would be
expected. The estimates suggest roughly 2% to 4% declines in manufacturing employment but
neither would be judged statistically significant by standard criteria.
6.4 Generation
A number of previous papers have examined the impact of RPS programs on state renewable gen-
eration (see Shrimali et al. (2012) for an excellent overview of the varied findings). In general, they
find that program heterogeneity appears to have some impact, while requirement stringency gener-
ally does not. Considering individual state responses, however, is likely confounded by spillovers, as
most RPS programs allow out-of-state resources within the REC region to comply.27 To allow for
these spillovers, we aggregate our state-level data to the REC region by taking state-level measures
of technology-specific generation shares, taking CO2 emissions intensity (in metric tons per MWh)
and whether an RPS program was law, and calculating a weighted average at the REC region level
where the weight is the MWh of generation in the relevant state by year observation. REC permits
can be traded within a REC region and the ten REC regions are shown in Appendix Figure A.1.
We then estimate versions of Equation (9), except now an observation is at the region by year level,
rather than state by year level.
27Johnson (2014) find that future RPS levels are associated with current regional capacity additions.
21
Table 6 presents estimates for generation sources observed in the EIA data. There is a case
for estimating unweighted (Panel A) and weighted (Panel B) versions of Equation (9) here. The
case for the unweighted regression is that the data generating process takes place at the REC
region level, with substantial cross-state spillovers due to the tradable REC permits. The case for
weighting by the number of states in a REC region depends on whether one wants to count more
heavily regions that are comprised of more states. Since the analysis of RPS on retail prices takes
place at the state level, weighting REC regions by the number of states to recover the effect on
the average state provides the most directly comparable results for the impact of RPS on prices,
generation, and carbon intensity.
The table reports separate estimates both 7 and 12 years after RPS passage for generation
shares of hydro, solar, wind, other renewables, coal, natural gas, petroleum, and nuclear, as well
as CO2 intensity. Although these technology share regressions are noisy, a few interesting findings
emerge. First, RPS passage is associated with substantial increases in the wind and hydro shares
of generation. The increase in wind generation is consistent with anecdotal evidence about wind
playing an important role in RPS compliance. However, we underscore that the wind estimates
are statistically significant in some specifications but certainly not all. Second, the estimates sug-
gest that RPS programs displace petroleum as its share declined meaningfully, although again the
standard errors preclude definitive conclusions. Third, although RPS programs likely have coun-
tervailing effects on natural gas generation - with renewables likely to displace baseload generation
but require an increase in the use of peaker plants as backup for their intermittent production -
our estimates are too noisy to provide a test with real empirical content.
Column (9) reports on specifications where CO2 intensity is the dependent variable. Just as
with the generation outcomes, the REC-level value of this variable is calculated as the weighted
average of state CO2 intensity, where the weight is the MWh of generation in the relevant state
by year. The mean of this variable in the year prior to program passage is 0.64. The Panel A
estimates indicate modest declines in CO2 intensity that have associated t-statistics below 1. In
Panel B, the estimated emissions intensity declines by about 16% (=.101/.641) seven years after
RPS passage and by 23% twelve years after passage. Both of these estimates are close to being
statistically significant at the 10% level.28
Overall, the table reveals that RPS programs are associated with changes in the generation mix
that are admittedly sensitive to specification and often imprecise. The most consistent evidence
appears to be that RPS programs led to reductions in the CO2 intensity of generation, although
the imprecision of these estimates also remains a source of concern. In the next subsection, we
combine the emissions intensity results with the price effect results to learn about the costs per
metric ton of CO2 abated.
28See Appendix Figure A.5 for event-study figures associated with these four estimates of the impact of RPSprograms on CO2 emissions intensity that illustrate the source of the column (9) estimates.
22
7 Interpretation
Our estimates suggest that RPS passage has imposed substantial costs on consumers of electricity
to date. To make this concrete, we calculate the higher charges that electricity customers paid
during the first 7 years after RPS passage in the 29 adopting states. This is calculated as the
product of the estimated increase in prices in each post-passage year (from the fitting of Equation
(9)) and total electricity consumption in the 29 RPS states in the analysis. The other side of the
ledger is the reduction in CO2 emissions in the 29 RPS states. This is calculated as the product
of the estimated effect of RPS passage on CO2 intensity and electricity generation separately for
each year post-passage. Recall, the estimated reduction in emissions intensity is about 3.5 times
larger in Panel B, compared to Panel A, of Table 6, so the results will be sensitive to the decision
of whether to weight observations on REC regions.
A natural summary statistic of RPS programs’ efficacy is the cost per metric ton of CO2 abated
and Table 7 uses this paper’s estimates to develop several of these measures. Specifically, the
first row of each panel reports the cumulative effect of RPS programs in their first 7 years after
passage, using the estimated impact on electricity prices in Table 2 and the unweighted (Panel A)
and weighted (Panel B) regressions for CO2 intensity from Table 6. Without discounting, the total
additional RPS costs over the first 7 years are about $125 billion in the 29 adopting states. The
cumulative reduction in CO2 emissions over the first 7 years after passage is 240 million metric tons
in Panel A and 1,010 million metric tons in Panel B.
Column (3) reports the cumulative estimated costs per ton of CO2 abated during the first 7
years after passage and they are $530 and $124 in the two panels, with the wide range underscoring
the sensitivity of the estimate to the estimated impact of RPS programs on CO2 intensity. The
second and third rows of each panel report on the cost per metric ton of CO2 abated in the 7th
and 12th years after passage. The cost per ton abated increases modestly between years 7 and 12
in both panels.
Overall, the estimates of the cost per metric ton of CO2 abated are high by almost any metric.
For example, the Obama Administration pegged the social cost of carbon (i.e., the monetized
damages from the release of an additional ton of CO2 in the year 2019) at roughly $51 in current
dollars (Greenstone et al., 2013; EPA, 2016). Thus, it appears that the costs of RPS programs
exceed their carbon reduction benefits (again, these benefits would be larger if these programs
reduce the future cost of renewable technologies that end up being deployed). Further, they exceed
the price of a permit to emit a ton of CO2 in all the major cap-and-trade markets globally by more
than an order of magnitude. For example, the current prices in the CA, Regional Greenhouse Gas
Initiative, European Union ETS, and Quebec markets are currently about $15, $6, $25, and $15,
respectively.29 Put another way, RPS programs appear to be achieve a small fraction of the CO2
29Because there are mandates inside these cap-and-trade programs, the permit price may not be reflective ofmarginal abatement costs across the entire covered sectors.
23
reductions per dollar of cost, relative to cap-and-trade markets.
There are several caveats and implications of these results that bear noting. First, the analysis
is “reduced form” so we cannot assign precise shares of the RPS programs’ full costs to differences
in generation costs, intermittency, transmission, and stranded assets. Further, it seems reasonable
to assume that these shares vary over time and in ways that further complicate trying to infer their
contributions. For example, it seems plausible that any stranded asset costs are declining at the
same time that intermittency costs are increasing, because the net requirements grow over time.
Second, there are two reasons that the cost per metric ton calculations may understate the full
social costs of RPS programs. This is because the price effects only reflect the portion borne by
ratepayers. However, it seems reasonable to presume that at least some of the costs will be borne
by owners of capital (e.g., generators or transmission), particularly in states with restructured
electricity markets. Further, it is possible that some of the costs are shared by all the participants
in wholesale electricity markets, which in several cases includes states with and without RPS
programs. If the costs are partially reflected in retail prices in non-adopting states, then the
difference in differences approach would understate the full costs borne by ratepayers because it
would miss the portion borne by ratepayers in non-adopting states and understate the effect in
adopting states.
Third, more broadly, a randomized control trial is unavailable here, so there will always be a
form of unobserved heterogeneity that could explain the results without RPS programs playing
a causal role. For example, our measures of other state programs that influence retail electricity
prices are limited in their detail, only measuring the years in which states adopted three of these
types of programs. So while our estimates are adjusted for the presence of three of these types of
programs, this may fail to capture their full impact on electricity prices and that could cause us
to understate or overstate the impacts of RPS programs on retail electricity prices, depending on
their correlation with RPS programs.
Fourth, it is often claimed that renewable policies provide an external benefit by reducing the
costs of future generation that is generic and cannot be fully appropriated by the firm that is
expanding its operations. If there are such spillovers or positive externalities, then our estimates of
the costs per metric ton of abatement will be systematically too high because they will not account
for the benefits received by customers outside of the RPS state’s jurisdiction. In principle, these
benefits could be global and thus quite substantial. The coincidence of the proliferation of policies
that support renewable energy and the decline in solar prices over the last decade are consistent
with the possibility of such spillovers. However, research that isolates the magnitude of any such
spillovers from other factors is probably best described as emerging, making this is a rich area for
future research (Gillingham and Stock, 2018).
24
8 Conclusion
This paper has provided the first comprehensive evaluation on the impacts of RPS programs,
which are perhaps the most popular and prevalent carbon policy in the United States. First, these
programs mandated increases in renewable generation that are often smaller than is advertised.
Seven years after passage, RPS programs require a 1.8 percentage point increase in renewable’s
share of generation, and 12 years after it is 4.2 percentage points. Second, RPS program passage
leads to substantial increases in electricity prices that mirror the program’s increasing stringency
over time. Seven years after passage, we estimate that average retail prices are 1.3 cents per kWh
or 11% higher than they otherwise would be. The corresponding effect twelve years later is 2.0
cents per kWh or 17% higher. Third, the estimates indicate that passage of RPS programs lead
to reductions in the generating mix’s carbon intensity, although these estimates can be noisier and
more sensitive to specification than is ideal. Putting the results together, the cost per metric ton
of CO2 abated exceeds $115 in all specifications and ranges up to $530, making it at least several
times bigger than conventional estimates of the social cost of carbon.
A particularly striking finding is that the indirect costs of RPS programs, which have not been
possible to comprehensively measure to date, appear to account for the majority of RPS program
costs. A recent study suggests that the direct costs of RPS increase retail electricity prices by
2% (Barbose, 2018), which is substantially smaller than our estimates that prices are about 11%
higher 7 years after passage. Although there are several differences between these two studies,
it seems likely that the indirect costs, including intermittency, transmission, and stranded asset
payments, account for a substantial fraction of RPS program costs. This finding suggests caution
in extrapolating declines in the direct generation cost of renewable energy to its overall impact
on electricity prices, and suggests that reducing indirect costs associated with grid integration
could represent the more important barrier to substantially increasing renewable energy’s share of
generation and meaningfully decreasing carbon dioxide emissions.
Overall, the paper’s results underscore the importance of research on policy and technology
mechanisms to reduce the costs of renewable energy, and imply that mechanisms to facilitate the
integration of intermittent sources onto the grid, such as advanced storage technologies or time-
of-use pricing, could be especially beneficial. While the potential damages from global climate
change have been widely documented, it is almost self-evident that failing to cost-effectively re-
duce emissions will ultimately limit the magnitude of these cost reductions. Further, policies that
substantially increase the price of electricity tend to have a regressive impact that hits low-income
consumers hardest, and therefore may be especially unattractive in developing countries that ac-
count for a large and growing share of global emissions. The most effective climate policy in
technologically advanced and innovative nations such as the United States will reduce emissions
domestically, but also involves developing low-carbon energy systems that are cost-effective enough
to promote adoption in the rest of the world.
25
References
Autor, D., D. Dorn, and G. Hanson (2013): “The China Syndrome: Local Labor Market
Effects of Import Competition in the United States,” American Economic Review, 103, 2121–
2168. 22
Barbose, G. (2018): “U.S. renewable portfolio standards: 2018 annual status report,” Lawrence
Berkeley National Lab (LBNL), Berkeley, CA. 1, 4.1, 8
Barnes, J. (2014): “DSIRE dataset on renewable portfolio standards,” Database of State in-
centives for Renewables & Efficiency (DSIRE), North Carolina State University, Raleigh, NC.
4.1
Bjørner, T., M. Togeby, and H. Jensen (2001): “Industrial Companies’ Demand for Elec-
tricity: Evidence From a Micropanel,” Energy Economics, 23, 595–617. 6.1
Borenstein, S. (2008): “The Market Value and Cost of Solar Photovoltaic Electricity Produc-
tion,” Working Paper. 1
——— (2009): “To what electricity price do consumers respond? Residential demand elasticity
under increasing-block pricing,” . 6.3
Bushnell, J., M. Flagg, and E. Mansur (2017): “Capacity markets at a crossroads,” Report
to the Department of Energy, Office of Energy Policy and Systems Analysis, Washington DC. 3,
3.2
Chen, C., R. Wiser, and M. Bolinger (2007): “Weighing the Costs and Benefits of State
Renewables Portfolio Standards: A Comparative Analysis of State-Level Policy Impact Projec-
tions,” Tech. Rep. 61580, LBNL. 1
Cullen, J. (2013): “Measuring the environmental benefits of wind-generated electricity,” Ameri-
can Economic Journal: Economic Policy, 5, 107–133. 1
Denholm, P. and R. M. Margolis (2007): “Evaluating the limits of solar photovoltaics (PV)
in traditional electric power systems,” Energy policy, 35, 2852–2861. 1
EEI (2011): “Transmission projects: At a glance,” Edison Electric Institute (EEI). 4
EIA (2019): “Annual Energy Outlook 2019,” U.S. Energy Information Administration (EIA). 1
EPA (2016): “Social cost of carbon,” Environmental Protection Agency (EPA): Washington, DC,
USA. 1, 7
26
Fabrizio, K., N. Rose, and C. Wolfram (2007): “Do Markets Reduce Costs? Assessing the
Impact of Regulatory Restructuring on US Electric Generation Efficiency,” American Economic
Review, 97, 1250–1277. 4.2
Fischer, C. (2010): “Renewable Portfolio Standards: When Do They Lower Energy Prices?”
Energy Journal, 31, 101–120. 1
GAO (2017): “Four Regions Use Capacity Markets to Help Ensure Adequate Resources, but FERC
Has Not Fully Assessed Their Performance,” U.S. Government Accountability Office (GAO)
Report to Congressional Committees. 3.2
Gillingham, K. and J. Stock (2018): “The cost of reducing greenhouse gas emissions,” Journal
of Economic Perspectives, 32, 53–72. 7
Gowrisankaran, G., S. S. Reynolds, and M. Samano (2016): “Intermittency and the value
of renewable energy,” Journal of Political Economy, 124, 1187–1234. 1
Greenstone, M., E. Kopits, and A. Wolverton (2013): “Developing a social cost of car-
bon for US regulatory analysis: A methodology and interpretation,” Review of Environmental
Economics and Policy, 7, 23–46. 1, 7
Heeter, J., G. Barbose, L. Bird, S. Weaver, F. Flores-Espino, K. Kuskova-Burns, and
R. Wiser (2014): “A Survey of State-Level Cost and Benefit Estimates of Renewable Portfolio
Standards,” Tech. Rep. 6A20-61042, NREL/TP. 1
Ito, K. (2014): “Do Consumers Respond to Marginal or Average Price? Evidence from Nonlinear
Electricity Pricing,” American Economic Review, 104, 537–563. 6.3
Jacobson, M. Z., M. A. Delucchi, M. A. Cameron, and B. A. Frew (2015): “Low-cost
solution to the grid reliability problem with 100% penetration of intermittent wind, water, and
solar for all purposes,” Proceedings of the National Academy of Sciences, 112, 15060–15065. 1
Johnson, E. P. (2014): “Measuring the Productive Inefficiency in Renewable Electricity Genera-
tion,” Working Paper. 27
Joskow, P. L. (2011): “Comparing the costs of intermittent and dispatchable electricity generat-
ing technologies,” The American Economic Review, 101, 238–241. 1
Lamont, A. D. (2008): “Assessing the long-term system value of intermittent electric generation
technologies,” Energy Economics, 30, 1208–1231. 1
Milligan, M., E. Ela, B.-M. Hodge, B. Kirby, D. Lew, C. Clark, J. DeCesaro, and
K. Lynn (2011): “Integration of variable generation, cost-causation, and integration costs,” The
Electricity Journal, 24, 51–63. 1
27
Mills, A., R. Wiser, and K. Porter (2009): “The cost of transmission for wind energy: A
review of transmission planning studies,” Lawrence Berkeley National Lab (LBNL), Berkeley,
CA. 1, 4
Schmalensee, R. (2012): “Evaluating Policies to Increase Electricity Generation from Renewable
Energy,” Review of Environmental Economics and Policy, 6, 45–64. 1
Shrimali, G., S. Jenner, F. Groba, G. Chan, and J. Indvik (2012): “Have State Renewable
Portfolio Standards Really Worked? Synthesizing Past Policy Assessments to Build an Integrated
Econometric Analysis of RPS Effectiveness in the U.S.” Working Paper. 6.4
Tuerck, D., P. Bachman, and M. Head (2013): “The Economic Impact of Wisconsin’s Re-
newable Portfolio Standard,” 26. 1, 6
Upton, G. and B. Snyder (2017): “Funding renewable energy: An analysis of renewable portfolio
standards,” Energy Economics, 66, 205–216. 1
28
29
9 Figures
Figure 1: RPS Passage by State
Sources: US Department of Energy and state government websites.Notes: States that have adopted any RPS policy are colored according to the year in which the RPS legislation was first passed.
30
Figure 2: Number of RPS Programs Newly Passed into Law, by Year
4
6
8
10
12
14
Ave
rage
Ret
ail E
lect
ricity
Pric
es
1
2
3
4
5
6
7R
PS P
rogr
ams
Pass
ed
1990 1994 1998 2002 2006 2010 2014year
RPS Programs (left) Prices (Cents per kWh) (right)
Sources: Department of Energy and state government websites (number of policies) and EIA (prices).
31
Figure 3: RPS Total and Net Requirements, by State
1991 1997 1998 1999 2001 20022004 2005 2006 2007 2008 2009
0.0
5.0
10.0
15.0
20.0
25.0
30.0
Iowa
Nevada
Connect
icut
Maine
New Je
rsey
Texas
Wiscons
inArizo
naCalif
ornia
Massach
usetts
Colorad
oHaw
aiiMary
land
New M
exico
New York
Pennsy
lvania
Rhode I
sland
Delawa
re
District
of Co
lumbia
Montana
Washing
tonIlli
nois
Minneso
ta
New Ham
pshire
North C
arolin
aOreg
onMich
igan
Missouri Ohio
Kansas
Requ
irem
ent (
%)
Sources: Department of Energy and state government websites; Lawrence Berkeley National Lab (LBNL).Notes: States are sorted by the year in which their RPS policies were first passed. The bars are colored according to RPS passage year. The total height of each bar denotes thegross RPS requirement at τ = 6; the non-patterned portion of each bar denotes net requirement at τ = 6. The data for gross RPS requirements are from the LBNL, in MWh,and are converted to percentages by dividing by contemporary generation at τ = 6. Note that these percentages do not exactly equal the prescribed statutory percentages inthe regulation.
32
Figure 4: Estimated Effects of RPS Programs on Net Renewable Requirements and RetailElectricity Prices
(a) Net RPS Requirements
-1
-.5
.5
1
1.5
2
2.5
Net
% R
equi
rem
ent
-6 -4 -2 0 2 4 6Year Relative to Program Passage
(b) Retail Prices
-1
-.5
.5
1
1.5
2
2.5
Cen
ts p
er k
Wh
-6 -4 -2 0 2 4 6Year Relative to Program Passage
Source: EIA; LBNL; Department of Energy and state government websites.Notes: Graph (a) shows the mean net RPS requirement percentage for event years τ = -7 to τ = 6. Graph (b) shows coefficientsfor στ for τ = -7 to τ = 6 from the event study specification in Equation (7) for retail electricity prices on indicator variablesfor years relative to program passage, controlling for state, year, and other programs fixed effects. Blue lines show the pointestimates and gray lines contain the 95% confidence interval. Gross RPS requirement data are from the LBNL. Electricity pricedata and electricity generation for calculating net requirement are from the EIA. RPS program passage dates and requirementsare from the Department of Energy and state government websites. Standard errors are clustered at the state-level.
33
10 Tables
Table 1: Summary Statistics
Mean RPS
Mean Control
P-value RPS vs Control
(1) (2) (3)
Price (2018 Cents/kWh)Total 11.4 9.4 0.01Residential 13.4 11.3 0.01Commercial 11.8 9.8 0.01Industrial 8.5 6.9 0.01
Price Change ! = -1 vs -7 (2018 Cents/kWh) -0.6 -0.6 0.92Total Sales (TWh) 76.2 64.3 0.38Population (Millions) 7.0 4.7 0.11CO2 Emissions (Million mt) 48.0 49.2 0.90Renewable Potential (PWh)
Solar 9.1 6.6 0.34Wind 1.1 0.9 0.40
GenerationTotal (TWh) 80.5 73.3 0.64RPS Eligible (TWh) 8.9 5.9 0.36RPS Eligible (% of Total) 13.5 13.0 0.89
Generating CapacityTotal (GW) 20.3 18.4 0.60RPS Eligible (GW) 2.5 1.6 0.36RPS Eligible (% of Total) 14.2 14.3 0.99
Other Programs (%)Public Benefit Funds 0.41 0.11 0.00Net Metering 0.66 0.45 0.04Green Power Purchasing 0.07 0.02 0.29
Notes: “Mean RPS” is for RPS states in the year prior to RPS passage. A control is defined for each RPS state as the meanacross non-RPS states and RPS states that have yet to pass RPS, in the year prior to the reference RPS state’s RPS passage.“Mean Control” is the average across these controls. Column (3) reports p-values from a two-sample t-test between Column (1)and (2) that allows for unequal variances across groups. Iowa is excluded from these summary statistics due to the particularlyearly passage of its RPS.
34
Table 2: Estimates of RPS Impact on Retail Electricity Prices
(1a) (1b) (2a) (2b) (3a) (3b) (4a) (4b)
Panel A: 7 Post-Passage Years, Balanced SampleMean Shift (δ3) 0.54 0.30 0.48 0.17 0.50 0.30 0.56 0.69
(0.35) (0.25) (0.38) (0.24) (0.36) (0.24) (0.37) (0.46)Trend Break (β3) 0.16* 0.26*** 0.10 0.01
(0.08) (0.09) (0.09) (0.10)Effect of RPS 7 years after passage 1.27** 1.71** 0.91 0.78(6β3 + δ3) (0.61) (0.66) (0.62) (0.48)
Panel B: 12 Post-Passage Years, Unbalanced SampleMean Shift (δ3) 0.66 0.35 0.65 0.26 0.59 0.33 0.62 0.57
(0.41) (0.31) (0.45) (0.31) (0.42) (0.31) (0.42) (0.40)Trend Break (β3) 0.15** 0.22*** 0.10 0.07
(0.07) (0.07) (0.08) (0.08)Effect of RPS 12 years after passage 1.98** 2.71*** 1.38 1.34(11β3 + δ3) (0.81) (0.89) (0.88) (0.84)
Mean at ! = -1 11.4 11.4 13.4 13.4 11.8 11.8 8.5 8.5
State FE Yes Yes Yes Yes Yes Yes Yes YesYear FE Yes Yes Yes Yes Yes Yes Yes YesOther Programs Yes Yes Yes Yes Yes Yes Yes YesN 1300 1300 1300 1300 1300 1300 1300 1300
Average Retail PriceAverage Retail Price
ResidentialAverage Retail Price
CommercialAverage Retail Price
Industrial
Notes: Columns (1a) through (4b) show estimates from Equations (8) and (9), where (a) columns correspond to Equation (8) and (b) columns correspond to Equation (9),with total retail electricity price and sector-specific retail electricity prices as the response variables. In Panel A, coefficient estimates are for states with data 7 years beforeand 7 years after RPS passage. In Panel B, coefficient estimates are for states with data 7 years before and 12 years after RPS passage. Using Equation (9) notation, theeffect of RPS 7 years after passage is 6β3 + δ3, and the effect of RPS 12 years after passage is 11β3 + δ3. Standard errors are clustered at the state-level.Asterisks denote p-values: < 0.10 (*), < 0.05 (**), < 0.01 (***).
35
Table 3: Robustness Checks for RPS Impact
(1) (2) (3) (4)
Panel A: Total
Effect of RPS 7 years after passage 1.13* 1.15* 1.08* 0.90(6β3 + δ3) (0.63) (0.60) (0.58) (0.56)
Panel B: Residential
Effect of RPS 7 years after passage 1.55** 1.59** 1.60*** 1.40**(6β3 + δ3) (0.68) (0.65) (0.59) (0.60)
Panel C: Commercial
Effect of RPS 7 years after passage 0.68 0.77 0.70 0.65(6β3 + δ3) (0.63) (0.61) (0.58) (0.58)
Panel D: Industrial
Effect of RPS 7 years after passage 0.69 0.64 0.32 0.29(6β3 + δ3) (0.51) (0.47) (0.55) (0.58)
Other Programs X X X XExcludes States with Nuclear in Original RPS XExcludes Hawaii XState FE X X X XYear FE X XYear-Region FE XYear-Division FE XN 1248 1274 1300 1300
Retail Electricity Price
Notes: The (a) columns report the aggregate effect 7 years after RPS passage from the mean-shift model given by Equation(8). The (b) columns report the same effect from the trend-break model given by Equation (9). Coefficient estimates are forstates with data 7 years before and 7 years after RPS passage. Year-Region fixed effects are for all combinations of years andCensus regions. Year-Division fixed effects are for all combinations of years and Census divisions. The two states with nuclearin their original RPS are Massachusetts and Ohio. One specification excludes Hawaii due to its geographic isolation and thusits inability to trade electricity across state borders. Standard errors are clustered at the state-level.Asterisks denote p-values: < 0.10 (*), < 0.05 (**), < 0.01 (***).
36
Table 4: Heterogeneous Effects of RPS Programs on Retail Electricity Prices
Total ResidentialPanel A: Late Adopters
Effect of RPS 7 years after passage 1.19 1.53(6β3 + δ3) (0.82) (0.92)
(Effect of RPS)*Late -0.12 0.27(1.51) (1.51)
Panel B: Ever RestructuredEffect of RPS 7 years after passage 1.99* 2.35**(6β3 + δ3) (1.17) (1.17)
(Effect of RPS)*Restructured -0.86 -0.69(1.33) (1.38)
Panel C: Has Solar Set-AsideEffect of RPS 7 years after passage 0.70 1.07(6β3 + δ3) (0.75) (0.90)
(Effect of RPS)*Solar Set-Aside 1.22 1.36(1.19) (1.20)
Panel D: Heavy Coal StatesEffect of RPS 7 years after passage 0.78 1.31(6β3 + δ3) (0.84) (0.96)
(Effect of RPS)*Heavy Coal 0.95 0.82(1.21) (1.27)
State FE Yes YesYear FE Yes YesOther Programs Yes YesN 1300 1300
Notes: The coefficients give the aggregate effect of RPS programs on total and residential retail prices 7 years after passageestimated from the trend-break model. The top row in each panel shows the coefficient for the subset of states not in thegiven category and the bottom row shows the difference in the coefficient for the given subset. All coefficient estimates are forstates with data 7 years before and 7 years after RPS passage. Using Equation (9) notation, the effect of RPS 7 years afterpassage is 6β3 + δ3. Standard errors are clustered at the state-level.Asterisks denote p-values: < 0.10 (*), < 0.05 (**), < 0.01 (***).
37
Table 5: RPS Effect on Sales and Employment
Total Total Manufacturing Manufacturing(1a) (1b) (2a) (2b) (3a) (3b)
Mean Shift (δ3) -0.01 0.001 -0.037(0.02) (0.016) (0.028)
Effect of RPS 7 years after passage 0.01 0.024 -0.022(6β3 + δ3) (0.03) (0.022) (0.035)
State FE Yes Yes Yes Yes Yes YesYear FE Yes Yes Yes Yes Yes YesOther Programs Yes Yes Yes Yes Yes YesN 1300 1300 1200 1200 1200 1200
EmploymentSalesTotal
Notes: The dependent variable in Columns (1a) and (1b) is the log of total sales in MWh. The dependent variable in Columns (2a) and (2b) is the log of total employment ineach state; in Column (3a) and (3b) is log manufacturing employment. The (a)-columns show the mean-shift estimates from Equation (8) for sales or employment. The(b)-columns report the aggregate effect 7 years after program passage from the trend-break model given by Equation (9). Coefficient estimates are for states with data 7 yearsbefore and 7 years after RPS passage. Using Equation (9) notation, the effect of RPS 7 years after passage is 6β3 + δ3. Standard errors are clustered at the state-level.Asterisks denote p-values: < 0.10 (*), < 0.05 (**), < 0.01 (***).
38
Table 6: Estimates of RPS Impact on Generation and CO2 Emissions (Trend Break)
Hydro Solar WindOther
Renewables Coal Natural Gas Petroleum Nuclear CO2 intensity(1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A: Unweighted
Effect of RPS 7 years after passage 2.87 -0.08 1.22** 0.53 -1.27 1.26 -1.68 -2.78 -0.029(6β3 + δ3) (2.46) (0.13) (0.56) (0.48) (3.70) (3.80) (1.39) (3.03) (0.034)
Effect of RPS 12 years after passage 4.87 0.04 2.84*** 0.26 3.37 -2.17 -3.83 -5.62 -0.043(11β3 + δ3) (3.80) (0.18) (1.05) (0.87) (7.70) (8.32) (3.54) (4.94) (0.057)
Panel B: Weighted
Effect of RPS 7 years after passage 9.97 -0.39 0.98 0.86 -6.28 -4.98 -1.42 0.95 -0.101(6β3 + δ3) (6.19) (0.29) (1.35) (0.77) (4.38) (7.14) (2.63) (4.12) (0.062)
Effect of RPS 12 years after passage 16.33 -0.23 0.93 1.24 -5.90 -9.89 -4.47 1.14 -0.149(11β3 + δ3) (9.06) (0.28) (1.77) (1.35) (6.44) (14.92) (5.75) (7.06) (0.089)
Mean at ! = -1 4.53 0.00 0.07 2.42 48.28 14.60 7.56 21.98 0.641Region FE Yes Yes Yes Yes Yes Yes Yes Yes YesYear FE Yes Yes Yes Yes Yes Yes Yes Yes YesOther Programs Yes Yes Yes Yes Yes Yes Yes Yes YesN 260 260 260 260 260 260 260 260 260
Notes: Columns (1) through (8) show estimates from Equation (9), each with a specific generation source as the dependent variable. Column (9) also shows estimates fromEquation (9), but uses the CO2 emissions intensity as the dependent variable. Coefficient estimates are either for states with data 7 years before and 7 years after RPSpassage, or for states with data 7 years before and 12 years after RPS passage. Using Equation (9) notation, the effect of RPS 7 years after passage is 6β3 + δ3, and the effectof RPS 12 years after passage is 11β3 + δ3. Panel A is a region-level generation-weighted average of the states in the region, unweighted by count of states in each REC-region.Panel B additionally includes state count as regression weights. Standard errors are clustered at the REC region-level.Asterisks denote p-value < 0.10 (*), < 0.05 (**), < 0.01 (***).
39
Table 7: Estimated Cost of Abating CO2 Emissions from RPS
CO2 Reduction(mm ton)
Cost to Consumers (bn $)
Cost per Ton Reduced ($)
(1) (2) (3)
Panel A: Unweighted
Cumulative Effect of RPS 236.1 125.2 530(for first 7 years after passage)
Effect of RPS 7 years after passage 71.5 29.5 412(6β3 + δ3)
Effect of RPS 12 years after passage 62.8 28.2 449(11β3 + δ3)
Panel B: WeightedCumulative Effect of RPS 1,011.8 125.2 124(for first 7 years after passage)
Effect of RPS 7 years after passage 249.8 29.5 118(6β3 + δ3)
Effect of RPS 12 years after passage 217.6 28.2 129(11β3 + δ3)
State Count 7 years after passage 29 29 29State Count 12 years after passage 16 16 16
Notes: Column (1) shows estimates from Equation (9) estimated at the REC level, where Panel A excludes and Panel B includes state-count weights. Column (2) showsestimates from Equation (9) estimated at the state level, so no state-count weights are used in either panel. Column (3) is the ratio of column (2) to (1). The cumulative effectof RPS is the sum of the year-by-year effects for τ = 0 through τ = 6 inclusive.
40
11 Appendix
Figure A.1: REC Tracking Markets
Source: EPA.
41
Figure A.2: Estimated Effects of RPS Programs on Net Renewable Requirements and RetailElectricity Prices (Extended Post Period)
(a) Net RPS Requirements
-1
1
2
3
4
5
6
Net
% R
equi
rem
ent
-6 -4 -2 0 2 4 6 8 10Year Relative to Program Passage
(b) Retail Prices
-1
1
2
3
4
Cen
ts p
er k
Wh
-6 -4 -2 0 2 4 6 8 10Year Relative to Program Passage
Source: EIA; LBNL; Department of Energy and state government websites.Notes: Graph (a) shows the mean net RPS requirement percentage for event years τ = -7 to τ = 11. Graph (b) shows coefficientsfor στ for τ = -7 to τ = 11 from the event study specification in Equation (7) for retail electricity prices on indicator variablesfor years relative to program passage, controlling for state, year, and other programs fixed effects. Blue lines show the pointestimates and gray lines contain the 95% confidence interval. Gross RPS requirement data are from the LBNL. Electricity pricedata and electricity generation for calculating net requirement are from the EIA. RPS program passage dates and requirementsare from the Department of Energy and state government websites. Standard errors are clustered at the state-level.
42
Figure A.3: Estimated Effects of RPS Programs on Gross Renewable Requirements (Ex-tended Post Period)
-2
2
4
6
8
10
12
14G
ross
% R
equi
rem
ent
-6 -4 -2 0 2 4 6 8 10Year Relative to Program Passage
Source: LBNL; Department of Energy and state government websites.Notes: The graph shows the mean gross RPS requirement percentage for event years τ = -7 to τ = 11.
43
Figure A.4: Electricity Prices Before and After RPS Passage, by Sector
-1
-.5
.5
1
1.5
2
2.5
Cen
ts p
er k
Wh
-6 -4 -2 0 2 4 6
Total
-1
-.5
.5
1
1.5
2
2.5
Cen
ts p
er k
Wh
-6 -4 -2 0 2 4 6
Residential
-1
-.5
.5
1
1.5
2
2.5
Cen
ts p
er k
Wh
-6 -4 -2 0 2 4 6
Commercial
-1
-.5
.5
1
1.5
2
2.5
Cen
ts p
er k
Wh
-6 -4 -2 0 2 4 6
Industrial
Source: EIA; LBNL; Department of Energy and state government websites.Notes: Graphs show coefficients for στ for τ = -7 to τ = 6 from the event study specification in Equation (7). This specification regresses the dependent variable - retailelectricity prices - on indicator variables for years relative to program passage, controlling for state, year, and other programs fixed effects. Blue lines show the point estimatesand gray lines contain the 95% confidence interval. Electricity price data are from the EIA. Standard errors are clustered at the state-level.
44
Figure A.5: CO2 Emissions Intensity Before and After RPS Passage
-.15
-.125
-.1
-.075
-.05
-.025
.025
.05
.075
.1
Emis
sion
s In
tens
ity
-6 -4 -2 0 2 4 6
Unweighted
-.15
-.125
-.1
-.075
-.05
-.025
.025
.05
.075
.1
Emis
sion
s In
tens
ity
-6 -4 -2 0 2 4 6
Weighted
-.2
-.15
-.1
-.05
.05
.1
.15
.2
Emis
sion
s In
tens
ity
-6 -4 -2 0 2 4 6 8 10
Unweighted
-.2
-.15
-.1
-.05
.05
.1
.15
.2
Emis
sion
s In
tens
ity
-6 -4 -2 0 2 4 6 8 10
Weighted
Source: EIA; LBNL; Department of Energy and state government websites.Notes: Graphs show coefficients for στ from the event study specification in Equation (7). This specification regresses the dependent variable - CO2 emissions intensity - onindicator variables for years relative to program passage, controlling for REC regions and year fixed effects, as well as other programs fixed effects whose values are a generation-weighted average of the states’ indicator values within a given REC region. The plots labelled ”Weighted” use state-count weights, and the ones labelled ”Unweighted” do not.The top two plots show a narrower time frame, from τ = −7 to τ = 6, where we have a balanced panel of 29 states. The bottom two plots show a larger time frame in which wehave an unbalanced panel that varies from 29 to 16 states. Blue lines show the point estimates and gray lines contain the 95% confidence interval. Standard errors are clusteredat the REC region level.
45