The cost of carbon dioxide abatement from state renewable portfolio standards

Resource and Energy Economics 36 (2014) 332–350

Contents lists available at ScienceDirect

Resource and Energy Economics

jo urnal homepage: www.elsev ier .com/ locate / ree

The cost of carbon dioxide abatement from staterenewable portfolio standards�

Erik Paul Johnson ∗

School of Economics, 221 Bobby Dodd Way, Georgia Institute of Technology, Atlanta, GA 30332,United States

a r t i c l e i n f o

Article history:Received 11 December 2012Received in revised form 30 December 2013Accepted 3 January 2014Available online 18 January 2014

JEL classification:Q28Q42Q52Q58

Keywords:EnergyElectricityCost of carbon abatementRenewable portfolio standards

a b s t r a c t

Renewable portfolio standards (RPSs) have become a popular toolfor state governments to promote renewable electricity generationand to decrease carbon dioxide emissions within a state or region.Renewable portfolio standards are a policy tool likely to persist formany decades due to the long term goals of many state RPSs andthe likely creation of a federal RPS alongside any comprehensive cli-mate change bill. Even though RPSs have become a popular policytool, there is little empirical evidence about their costs. Using thetemporal and regional variation in RPS requirements, I estimate thelong-run price elasticity of supply of renewable electricity genera-tion to be 2.67 (95% CI of 1.74, 3.60). Using my preferred elasticityestimate, I calculate the marginal cost of abatement from RPSs is atleast $11 per ton of CO2 compared to a marginal cost of abatementof $3 per ton in the Regional Greenhouse Gas Initiative.

© 2014 Elsevier B.V. All rights reserved.

1. Introduction

Renewable energy has become a salient policy issue at both the state and federal levels. Many stateshave adopted policies aimed at promoting the growth of renewable electricity within their state todecrease carbon dioxide (CO2) emissions, most prominently through a renewable portfolio standard

� This paper has benefited from valuable input from Ryan Kellogg, Meredith Fowlie, Michael Moore, and many seminarparticipants at the University of Michigan. All errors and omissions are my own.

∗ Tel.: +1 404 385 3891; fax: +1 404 894 1890.E-mail address: [email protected]

0928-7655/$ – see front matter © 2014 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.reseneeco.2014.01.001

dx.doi.org/10.1016/j.reseneeco.2014.01.001

http://www.sciencedirect.com/science/journal/09287655

http://www.elsevier.com/locate/ree

http://crossmark.crossref.org/dialog/?doi=10.1016/j.reseneeco.2014.01.001&domain=pdf

mailto:[email protected]

dx.doi.org/10.1016/j.reseneeco.2014.01.001

E.P. Johnson / Resource and Energy Economics 36 (2014) 332–350 333

Fig. 1. Timing of renewable portfolio standard adoption by state.

(RPS). An RPS is a mandate that retail electricity providers purchase a specified fraction of their elec-tricity sales from renewable sources. RPSs typically begin with a small requirement for renewableelectricity and incrementally increases over a 15–25 year period. A typical RPS is passed by a state leg-islature a few years before retail providers are required to meet the standard. This allows time for newcapacity to be built to meet the RPS requirements. For example, Massachusetts’s RPS, that was passedin 1997, requires retail providers to demonstrate that 1% of their electricity sales come from renewablegeneration in 2003 with the amount of required renewable electricity increasing by between one-halfand one percentage points in every subsequent year. The end goal for Massachusetts’s RPS occurs in2020, when 15% of electricity sales must come from renewable sources. If a retail provider fails tomeet its requirement in a given year, it must pay a penalty proportional to the difference between thetarget and the amount of renewable electricity it purchased.

In 1997, three states had renewable portfolio standards (Iowa, Massachusetts, and Nevada) whereasby the end of 2011, 37 states and the District of Columbia had passed an RPS into law.1 (Fig. 1 displayswhen states have passed RPSs.) Since the electricity sector accounts for 42% of CO2 emissions nation-ally, RPSs may have the ability to substantially decrease CO2 emissions at a low cost. However, therehas been little quantitative examination of their effectiveness or the cost of CO2 abatement from RPSs,particularly accounting for heterogeneity in state policies.

The goal of this paper is therefore to estimate the long-run price elasticity of supply of renewablegeneration. The price elasticity is a key parameter in estimating the cost of CO2 abatement from RPSsand allows me to compare the cost of abatement between state level RPSs and a regional cap-and-tradeprogram.2 I use an instrumental variable approach to estimate this elasticity to avoid simultaneity biasin my elasticity estimate. I find that the price elasticity of renewable electricity capacity is 2.67. Usingmy estimates, I calculate the cost of exclusively using an RPS to decrease the carbon dioxide emissionsin the northeastern US. This elasticity suggests that the cost of abating an equivalent amount of CO2from an RPS in the northeastern US is between nearly four and eleven times larger than the costs ofCO2 abatement under a regional cap-and-trade program (the Regional Greenhouse Gas Initiative).

To identify the long-run supply price elasticity, I use variation from the pre-specified RPSimplementation schedules. The incremental changes in demand for renewable generation from the

1 Some of these RPSs are voluntary goals and do not have enforcement mechanisms.2 There are some cost estimates of a federal RPS in the literature, for instance see Palmer and Burtraw (2005), but these

estimates come from simulation models of the electricity sector rather than empirically estimating the response to policies.

334 E.P. Johnson / Resource and Energy Economics 36 (2014) 332–350

implementation schedules create an exogenous change in the demand for renewable electricity. Thesechanges provide me with an instrument for the price renewable generators receive for electricity.

In order to correctly measure the changes in demand for renewable capacity due to RPSs, I use ameasure of the strength of the incentives created by a particular state’s RPS. This measure is differentthan what has been used in most previous work on RPSs. Menz and Vachon (2006) and Adelaja andHailu (2008) use cross-sectional data to examine the effectiveness of RPSs in promoting the devel-opment of wind generators.3 However, both of these papers treat all RPSs the same by estimatingthe effect of RPSs on new capacity using a simple indicator for a state having an RPS. Both papersfind that RPSs are correlated with a greater presence of wind generators in that state,4 but due totheir cross-sectional approaches these papers are likely to suffer from omitted variables bias due totime-invariant factors in each state. In fact, Lyon and Yin (2007) suggest that a large wind potential ina state increases the probability of that state passing an RPS, which suggests that these studies mayinstead be incorrectly attributing the effects of an RPS to renewable generating potential.

Shrimali and Kniefel (2011) use panel data on state renewable capacity and attempt to discernwhich of the variety of renewable electricity policies including RPSs are most effective at increasingin-state renewable electricity capacity, examining four common renewable generating technologiesseparately. They find conflicting evidence that RPSs are associated with more renewable generatingcapacity, but also fail to account for much of the heterogeneity in RPSs. Yin and Powers (2010) doaccount for much of the heterogeneity in policies and adopt a measure of the RPS requirement similarto this paper’s measure. By using their preferred method of incorporating this heterogeneity, they finda significant impact of RPSs on the share of renewable generation.

The papers mentioned above, with the exception of Powers and Yin, assume that all RPSs createidentical incentives for wind generators regardless of how difficult the policies are to meet. However,there is substantial heterogeneity in the investment incentives that RPSs create. For instance, the firstyear that Pennsylvania’s RPS was implemented, the state had more than enough renewable capacityto meet the requirement; whereas the first year that Delaware’s RPS was implemented enough newrenewable generation had to be built to power approximately 2% of the state’s electricity demand.

An important shortcoming of all of these papers is that they treat RPSs as policies that only cre-ate incentives for in-state renewable generators. However, nearly all RPSs create incentives for allrenewable generators in a region, not just within a state. This is because most RPSs specify that eli-gible renewable electricity can be purchased from within a regional group of states to comply withthe RPS. Thus, treating RPSs as an incentive for only renewable generators within the state biasesthe results toward zero since some generation will be built in “untreated” neighboring states to meetan RPS. Therefore, in contrast to previous work, I aggregate RPSs into regional level policies to accu-rately measure wholesale generator’s incentives and produce unbiased estimates. Fig. 2 shows theRPS requirements for each state in a region and displays how just considering only one state’s RPSrequirement in a region can drastically underestimate the incentives renewable generators face in theregion.5

The price elasticity estimates help to inform estimates of the excess burden of CO2 reductions fromRPSs since they are not a first-best policy. Holland et al. (2009) show under general conditions thatpolicies that govern the rate of pollution rather than the level cannot be efficient. The intuition in thispaper is that a regulation governing the rate of pollution is a tax on the pollution intense good and asubsidy on the lower pollution intensity good at the same time. Though RPSs do not govern the rate ofCO2 emissions similar to instruments described in Holland, Hughes, and Knittel, since they implicitly

3 There are also a few qualitative assessments of RPS policies. Wiser et al. (2005) examine many of the policy design issuesassociated with RPSs and identify broad principles that could be considered best practices. Langniss and Wiser (2003) alsodo a qualitative assessment of the Texas RPS and suggest that it has likely been an effective driver of renewable generationdevelopment in Texas.

4 Menz and Vachon (2006) find somewhat mixed results depending on the specification but all point estimates are positiveif not statistically significant.

5 RPS requirements, usually stated in percentage terms, have been translated into capacity requirements in the graph. Adiscussion of how this is calculated can be found in Section 4.2.


Fig. 2. RPS requirements by region.

tax carbon-intensive electricity production and subsidize less carbon-intensive electricity production,the intuition can be extended to RPSs.6

Despite their inefficiency, one reason state politicians may prefer an RPS to a cap-and-trade pro-gram, even though it is not a first-best policy, is that it may be harder for retail electricity providersto avoid the requirements of an RPS than a cap-and-trade program.7 While a cap-and-trade programprovides an incentive to locate electricity generators outside of the jurisdiction of the program,8 RPSssidestep some of this leakage problem by applying the regulation to consumption instead of produc-tion. Locating a fossil electricity generator outside the geographical boundaries of an RPS does notchange the requirement about the renewable percentage of electricity sold. To the extent that thereis leakage in an RPS, it will come from electricity consumers locating outside of the state with anRPS because of higher electricity prices due to the RPS. However, consumption leakage is likely to berelatively small, particularly in the first years of an RPS since the renewable requirement only appliesto a fraction of total consumption.

The remainder of the paper is organized as follows. In the next section I discuss the details of RPSs,electricity markets, and the dimensions on which there is heterogeneity in RPS policies. In Section 3,I develop a model to ground our thinking about renewable generating capacity investment. In Sec-tion 4, I discuss the empirical methodology I use and the key variables. Section 5 describes the dataI use to examine RPSs. Section 6 discusses my results which is followed by a discussion of the policyimplications of my estimates in Section 7. Section 8 concludes.

6 For a formal model examining the implicit tax and subsidies in an RPS, see Fischer (2009).7 Moreover, it has proven politically difficult to regulate greenhouse gas emissions through either a carbon tax or a cap-and-

trade program, either at the national or regional level. For instance, the US House of Representatives passed a cap-and-tradebill in 2009 (usually referred to as the Waxman-Markey Climate Change Bill), but the bill was never brought to the Senate floor.There has been slightly more success with regional cap-and-trade programs such as the Western Climate Initiative and theRegional Greenhouse Gas Initiative in the Northeastern US, but the coalition of states has proven somewhat volatile as NewJersey has since dropped out of the program.

8 See Fowlie (2009) for a discussion of leakage in a cap-and-trade context.


2. RPS policy background

The first modern RPS was passed by Massachusetts in 1997 and was part of the electric utilityrestructuring legislation whereby electric generating capacity was separated from retail operationsof electric utilities.9 After Massachusetts, many other New England and Mid-Atlantic states followedsuit.10 By December 2011, 37 states had passed an RPS. Fig. 1 shows the timing and spatial distributionof RPS passage. In this paper I focus on the subset of states that have passed RPSs, have restruc-tured electricity markets, and transparent markets for renewable energy credits.11 This is because theincentives for potential electricity generators are clear due to the competitive wholesale electricitymarket.

In addition to selling electricity into the wholesale market, generators that use renewable sourcesalso create a renewable energy credit with every megawatt hour (MWh) of electricity produced.Renewable energy credits (RECs) are a pure financial product in most markets that retail electricityproviders are required to purchase to show compliance with a state’s renewable portfolio standard.12

Typically a REC describes the attributes of the electricity that was produced such as the location of thegenerator, the fuel that was used, and the date that the electricity was produced. Using this informa-tion, retail providers can purchase RECs that qualify to meet a particular state’s RPS. At the end of theyear, retail providers surrender the RECs that they have purchased to the state regulator to meet theRPS requirement.

Nearly all states have a system of fines, called alternative compliance payments (ACPs), for retailproviders that are short of their required number of RECs. These ACPs effectively set a price ceiling inthe market for RECs. If a retail provider has not purchased the required number of RECs, the providermust purchase ACPs to make up the difference from the state. The level of the alternative compliancepayment is usually determined by the state’s public utility commission and is generally above themarket price for RECs, giving retail providers an incentive to purchase RECs instead.13 In many states,retail providers end up paying relatively few fines. For instance, in 2003, Massachusetts collected lessthan 1% of the RPS requirements through alternative compliance payments.

These RECs provide a second stream of revenue for renewable generators. Since the average cost ofrenewable generation tends to be higher than that of fossil generation, the revenue from selling RECsencourages new renewable generating capacity to be built. Thus, the total price renewable generatorsreceive for each MWh of electricity is the price of the electricity plus the price of the REC.

3. Model

In this section I develop a simple model of investment in electricity generating capacity to illustratethe effect an RPS has on the incentives of electricity producers that motivates my empirical specifica-tions. Consider a representative firm deciding whether to invest in new generating capacity. For thepower plant to be profitable, the revenue the generator produces over its lifetime must exceed itscapital and operating costs:

T∑t=0

ˇtE[(pet + pREC

t )qt]≥K0 +T∑

t=0

ˇt(mt + E[ftqt]) (1)

9 Iowa passed a law in 1983 that required two retail electricity providers install a combined 105 MW of renewable capacity.However since this requirement never increases it is somewhat different from modern RPSs.

10 This was usually done as part of restructuring legislation or shortly afterwards.11 The states I will examine in this paper are in the New England ISO (Connecticut, Maine, Massachusetts, New Hampshire,

Rhode Island, and Vermont), the PJM control region (Delaware, Maryland, New Jersey, Pennsylvania, Virginia, West Virginia,and parts of Illinois, Indiana, Michigan, and Ohio), and the Electric Reliability Council of Texas (ERCOT).

12 California requires retail providers to enter into long term contracts with renewable electricity producers to purchaseboth the electricity and RECs to fulfill the state’s RPS obligation. This requirement was lifted by the California Public UtilityCommission in March 2010. California retail providers may purchase unbundled RECs to fulfill the RPS requirement in a limitedamount in 2010 and 2011, with the market becoming completely unrestricted in 2012.

13 Some states explicitly link the alternative compliance payment to a multiple of the market REC price, while others such asMassachusetts re-evaluate the penalty every few years to make sure the price is still above the market price for RECs.


where pet is the price of electricity at time t, pREC

t is the price of the REC at time t,14 qt is the quantityof electricity the generator provides at time t, K0 is the initial capital cost of the generator, mt is thevariable operating and maintenance costs associated with the generator at time t, ft, is the fuel cost attime t, T is the number of years of the useful life of the new capacity, and ̌ is a discount factor.15 Inequilibrium this condition holds with equality. In absence of an RPS, retail providers have no preferenceover the fuel used for electricity. Therefore pREC

t is zero and the equilibrium fraction must be at a pointwhere the price is equalized across types of generators.

A new generator will enter the market when there is sufficient excess quantity demanded over thelife of the generator such that inequality (1) holds. Two ways to induce generators to enter the marketare to decrease the cost of the capital investment or to increase the price the generator will receivefor its electricity over the new capacity’s life span. These two levers have been used by the federalgovernment to induce more renewable generators to enter the market in the form of the InvestmentTax Credit and the Production Tax Credit, respectively.

Renewable portfolio standards also induce renewable generators to enter the market by increasingthe price that a generator will receive through artificially shifting preferences for types of generation.16

These new preferences over types of generation cause a price wedge between fossil generation andrenewable generation, pREC

t .Thus Eq. (1) implies that renewable generators consider the price path of both the price of electricity

and the price of RECs when considering entry decisions. So long as the price of RECs is expected to begreater than zero, renewable generators have an additional incentive to enter the market. Notice alsothat entry decisions depend on the flow of revenue to the generator over the life of the generator, notjust the contemporaneous revenue.

The profitability condition implies that each generator has a critical total price (price of electricityplus the price of a REC) at which it will enter the market. Therefore, as contemporaneous prices andexpectations about future prices change the model predicts that we should see generators enteringthe market consistent with the profitability condition.

We can derive a supply curve for renewable generators by aggregating each firm’s decision aboutwhether to enter the market. Each generator enters the market if their profitability condition holds.Thus, the total new generating capacity in the market at time t is:

Qt =∑

i

I

[T∑

t=0

ˇtE[(pet + pREC

t )qit]≥Ki +T∑

t=0

ˇt(E[mit + ftqit])

]× Ci (2)

where i indexes generators and C is the capacity of generator i.17 Notice that there is a generator-specific capital cost, and each generator can expect a different amount of output. These two termsrationalize why we observe some renewable generators in existence in areas without a binding RPS.Consider a wind developer looking at potential locations to install a wind turbine. Not all locations areof equal value to the developer due to the fact that the wind blows at different speeds and differenttimes at each location. Sites where the wind blows more frequently, all else equal, will be worth more tothe developer since the turbine will create more electricity and has a marginal cost near zero. Thus, thebest locations will be developed first with each subsequent wind turbine being placed in a marginallyinferior location, necessitating a marginally higher price for the electricity generated by that turbine tomake it profitable. This suggests that existing renewable capacity satisfies the profitability condition

14 For generators not eligible for renewable portfolio standards, pRECt is zero.

15 This profitability condition abstracts away from any payments that generators receive for participating in ancillary servicemarkets where generators may be paid to be on standby, ready to produce electricity if called upon by the market operator.Typically renewable generators are not eligible to participate in these markets due to the unpredictability of wind and solargeneration. However, an increase in the amount of wind and solar generators participating in the electricity market will causean increase in demand for standby capacity services from fossil generators.

16 In a market with constant demand, this increase in demand for renewable generation decreases the demand for fossilgeneration by the same amount.

17 Note that the same condition holds for fossil generators, except their expectations over the price of RECs do not enter theirprofitability condition.


in Eq. (1) even when pRECt = 0 but as demand for renewable capacity increases, renewable generators

will need to receive a higher price for their electricity. Thus, the upward slope of the supply curve isdriven by heterogeneity in the value of locations and capital costs.

In order to aggregate across generators I need to assume that all generators have the same expec-tations over the trajectory of prices (electricity and RECs) over the life of each generator. With thisassumption, I can rewrite Eq. (2) as

Qt = f (ptotalt , ptotal

t+1 , . . ., ptotalt+T ) + �Xt (3)

where ptotal = pelectricity + pREC and Xt is a set of variables capturing the other factors that effect a gener-ator’s entry decisions such as fuel costs.

This equation suggests that I can estimate the price elasticity of supply of renewable generationusing a log–log specification by regressing the log of quantity of new renewable capacity on the log ofprice and other factors that affect entry decisions. However, this presents the traditional problem ofsimultaneous equations bias since price and quantity are determined by the intersection of supply anddemand. Since, price is an endogenous regressor, I instrument for the price that renewable generatorsreceive to consistently estimate this equation. The instrument needs to be correlated with the stream ofpayments renewable generators will receive into the future since the supply equation depends on thestream of payments over the lifetime of the generator. However, since RECs can usually be banked forat least two of years, the price of a REC today provides information about the future price of RECs. Thismeans that for renewable generators (where pREC /= 0) the total price of electricity that a generatorreceives today contains information about the total price of electricity the generator receives in thefuture. Based on this REC bankability and therefore the assumption that the contemporaneous totalprice of electricity for renewable generators contains information about the total price of electricityin the future, Eq. (3) can be simplified to only contain the contemporaneous total price of electricityfor renewable generators.

4. Empirical strategy

In order to estimate the long-run price elasticity, I use the implementation schedules of state RPSs asan instrument for the price that renewable generators receive for their electricity. RPS implementationschedules are typically written into the original RPS legislation and the RPS requirement increases eachyear that the RPS is in effect until the end goal is met. These schedules are unlikely to be correlatedwith unobserved supply shocks and therefore be a valid instrument since the schedules are writteninto the legislation, which rarely changes.

My model suggests two natural estimating equations to estimate the long-run price elasticity ofsupply. The first-stage equation estimates the price response to an exogenous change in the demandfor RECs. Due to the implementation schedule of each state’s RPS, the demand for RECs change in apredictable way. As derived in Section 3, new renewable generating capacity should respond to theentire flow of payments over the life of the generator. To capture this variation I construct a variable,RPS Requirementit,t+5, that measures of changes in RPSs’ stringency averaged over the current andsubsequent five years.18 This leads to a first stage equation of the following form:

log(ptotalit ) = ̌ log(RPS Requirementit,t+5 years) + �Xit + ˛i + �t + �it (4)

where ptotalit

= pelectricityt + pREC

t , the ˛i’s are region fixed effects, the � t’s are year and month fixed effects,and Xit’s are a set of controls and other policy variables that may effect the incentives of renewablegenerators and the price of substitute fuels such as natural gas and fuel oil. The region level fixed effectsabsorb time invariant differences across regions such as differential renewable generating potential,as Lyon and Yin (2007) suggest may be important in the decision to adopt an RPS. Year and monthfixed effects absorb differences across time that are constant across region. These are important since

18 I will discuss how this variable is constructed in Section 4.2. The results are not sensitive to the choice of a five year average.The results are similar for average RPS requirements over a variety of years, though are noisier the longer the time period thatis averaged.


over our period of examination various federal tax incentives have taken effect (and occasionally notbeen immediately renewed) such as the Investment Tax Credit and the Production Tax Credit thataffect the financial desirability of building renewable generation.19 The other variables, Xit, control forother policies that affect the incentives for renewable electricity providers unrelated to RPSs. Thesevariables allow for a more isolated estimate of the effect of only the RPS.20

As discussed in Section 3, generators should be making entry decisions based on the time pathof prices, not just contemporaneous prices. However, most RPSs allow RECs that were created and/orpurchased in one year to be banked and used for compliance in the following two (or more) years.21,22

Therefore the contemporaneous REC price contains information about the price of RECs in the futureso the first stage equation effectively captures the total price of electricity of the life of the generators.Using the predicted price from the first stage, I estimate a second stage that produces the price elasticityof renewable capacity from the equation:

log(RenewableCapit) = ̌ log(̂ptotalit

) + �Xit + ˛i + �t + �it . (5)

4.1. RPS implementation schedules and adoption

The two main sources of variation I use to identify the price elasticity are the sequence in whichstates enact RPSs and the implementation schedules. This subsection explores, to the extent possible,the assumption that this variation is exogenous to the supply of renewable electricity generatingcapacity. I first explore the implementation schedules and then the timing and possible motivatingfactors of RPS adoption and other state specific details.

Since the instrument is the entire RPS implementation schedule, this needs to be uncorrelatedwith in-region unobservable characteristics. Two main concerns come to mind when considering theexogeneity of the RPS implementation schedules. First, states that are early adopters of RPSs may haveparticularly aggressive implementation schedules, either due to a strong desire to promote renewableelectricity generation or because they have a lot of renewable resources that can be exploited.

A second concern about the exogeneity of RPS implementation schedules is that states that have alot of renewable generating capacity at the time the RPS is passed will have more aggressive imple-mentation schedules. Since more aggressive implementation schedules likely lead to higher REC pricessooner, existing renewable generators have much to gain by lobbying state legislatures for morestringent requirements.

Both of these concerns can be addressed empirically by examining the state implementation sched-ules. Since most RPSs follow a nearly linear implementation schedule, I estimate the slope of theimplementation schedule by regressing each schedule on a time trend. I then examine the correlationbetween these slopes and variables that address the concerns raised above about the endogeneity ofimplementation schedules.

To address the first concern that early adopter states have more aggressive implementation sched-ules, I regress the slope of the implementation schedule requirements (in MW of required newcapacity) on the year that each state’s RPS went into effect. The coefficient on the year the RPS wentinto effect is not statistically different from zero with a coefficient of −7.7 and a standard error of 17.7.The point estimate suggests that early adopters require an extra 8 MW of renewable capacity eachyear of an RPS but is clearly not statistically different from zero (p = 0.67). Moreover, an additionalrequirement of 8 MW per year is a relatively small difference given that the average increase in RPSrequirements is 166 MW per year.

19 For a discussion of the history of these policies and their consequences see Metcalf (2009), Wiser et al. (2007), and the JointCommittee on Taxation (2009).

20 The other policy variables are discussed in detail below.21 All states in my sample allow banking of RECs for two years except Texas, which allows three years of banking and Con-

necticut which currently does not allow banking. However, since Connecticut’s RPS has regional requirement and other statesin the region allow banking there is still likely some transmission of prices across time.

22 While not all states publicly report the how many RECs each electricity provider banks each year, states that do reportaggregate banking show that up to 20% of compliance RECs are banked for compliance in future years.


To address the second concern that states with a larger renewable sector before RPS passage willhave a more aggressive implementation schedule, I regress the slope of the implementation scheduleon the renewable capacity in that state at the time of RPS passage. The coefficient is not statisticallydifferent from zero with a coefficient of 0.18 and a standard error of 0.11. This suggests that the imple-mentation schedules are not a function of the renewable interests already established in a particularstate.

This may be because many states choose “round” numbers for both their end goal, such as “20%renewable electricity by 2020,” and a linear implementation schedule. Likely, these end goals areless amenable to manipulation by pressure groups and since the intervening years’ requirements areessentially a linear interpolation back through time, the implementation schedule is not changed muchby pressure groups.

Another concern about my approach is that the RPS policies may spill over into other regions thatdo not have RPSs or less stringent RPSs. However, there are likely only very small spillover effects inmy setup since the unit of observation is a regional electricity market. Many states publish a list of allof the approved generation facilities that are eligible to produce RECs that meet the state RPS. Whileoccasionally there are power generators located in states not included in the wholesale power market,a vast majority of the approved generation facilities are indeed located in states in the wholesale powermarket.23

Examining the order that states adopted their RPS policies, there does not appear to be much of apattern. Some of the states with the largest renewable potential from both wind and solar are in theGreat Plains and the Southwest. While many of the states in the Southwest do indeed have RPSs theyare not uniformly early policy adopters. Conversely, most of New England and the North Atlantic stateshave adopted RPS policies, some being among the first adopters but do not have a large renewablegenerating potential.24

Moreover, many of the first adopters of RPS policies adopted their RPS as part of the electricityrestructuring legislation. The electricity restructuring legislation in many of these states was a majorpiece of legislation that separated the retail and wholesale electricity markets. Most of the deregulationof the electricity markets were motivated by high retail electricity prices in the state and generally agroup of states in a region deregulated the wholesale electricity market at similar times. There is verylittle reason to believe that the deregulation legislation is correlated with unobserved covariates thataffect renewable electricity capacity.

Lyon and Yin (2007) empirically examine which states get RPSs. Their findings suggest that windpotential in the state increases the probability of RPS adoption (though not potential in other fuelsthat are typically included in RPSs such as solar or biomass). This will not be a problem for me since Iwill be controlling for this variation through my region level fixed effects.

Lyon and Yin also find that high local pollution levels, as measured by the fraction of the popula-tion living in counties that are designated as “nonattainment” under the Clean Air Act, increase theprobability of adoption of an RPS as well as some evidence that organized renewable energy lobbyinggroups increase the probability of adoption. In contrast to Rabe’s (2006, 2008) qualitative examina-tion, they also find that a state’s unemployment rate decreases the likelihood that a state will adoptan RPS. Rabe (2008) finds that states often emphasize the potential economic benefits of RPSs such ascreating “green” jobs or gaining a competitive advantage as a first mover in renewable energy tech-nology, but this does not seem to be a driving factor empirically as measured by the unemploymentrate. These papers give me confidence that RPS adoption is likely to be uncorrelated with many of theunobservables that would invalidate the instrument. Moreover, since RPS policies affect neighboringstates as well as the states in which they are passed, they are even less likely to be strongly correlatedwith in-region unobservables.

In addition to variation in adoption and implementation schedules, states vary as to the fuelsthat qualify for use in the RPS and if pre-existing renewable generation is eligible to fulfill the RPS

23 To the extent that there are spillovers across regions, my estimates will be biased toward zero.24 For instance, the six states in the New England ISO have less than one tenth of the renewable generating potential of Kansas

alone and half as much as 30 other states individually (Lopez et al., 2012).


requirements. Though there is considerable variation in the fuels that qualify to meet RPSs as a whole,often the variation is in the second or third tiers of the requirement.25 For the purposes of this paper, Iconcentrate only on the first tier of RPSs in which fuel requirements are relatively stable across statesand usually only include fuels that an average person would consider renewable such as wind (100%of states), solar (89% of states), and biomass (89% of states). States also vary considerably as to whetherthey allow renewable capacity that existed at the time the RPS was passed to produce RECs to meetthe RPS. However, this can be seen as a level shift in the requirement, reducing the incentive to investin new renewable generation but not changing the dynamic incentives.

4.2. Key variables

The primary variable of interest in the first stage regression is a variable constructed to measurethe stringency of a particular state’s RPS, RPS Requirementit,t+5 years. Most states, with the exception ofIowa and Texas, set their RPS goals out as a percentage of electricity sales, measured in megawatthours (MWh).

However, since I will be using data on renewable capacity (measured in megawatts) instead ofgeneration, I convert each RPS from a percentage (or megawatt hour) requirement to a capacityrequirement. To translate megawatt hour requirements into a megawatt capacity requirement, I use ameasure of historical capacity factor for renewable generators. To construct this measure, first capac-ity factors for each type of renewable capacity are calculated from historical data. Then, since each ofthe four most common renewable technologies (wind, solar, biomass, geothermal) have very differentpotentials in each region and are therefore are likely to be used differently in each region, weights arecreated for each technology based on theoretical potential (Lopez et al., 2012) for each technology inthat region and the observed capacity factors are weighted and summed to get an average renewableenergy capacity factor.26

I then adjust the requirement by the amount of eligible renewable generation available at thetime the RPS was passed. This means it is possible for an RPS to create zero incentive in some orall years. Consider a hypothetical RPS that requires 1% of generation to be from renewable sourcesin its first year. If the state has a total electricity demand of 10,00,000 MWh, there will need to be100,000 MWh of renewable electricity produced to meet the RPS. This means that a capacity of 100,000 MWh/8760 hours per year = 11.4 MW must be installed. However, if the state’s only renewableresource is wind with a capacity factor of 35%, then a total of 11.4 MW/0.35 = 32.6 MW must actuallybe installed to meet the requirement. If a state already has 20 MW of installed capacity, the RPS Require-ment for state i in year t will then be 32.6 − 20 = 12.6 MW. If the second year of our fictional RPS has a2% requirement, then the RPS Requirement variable in the second year will be 32.6 * 2 −20 = 49.2 MWsince the existing 20MW would still need to be subtracted off from the total.

Given this definition, it is possible for an RPS to create no new incentives for renewable generatorsif the prior capacity is sufficient to meet the RPS. For instance, Maine passed an RPS in September 1999that had a final goal that was less than the existing eligible capacity within the state. Subsequently, in2006 Maine passed another RPS that only new renewable facilities were eligible to meet the require-ment. This is the RPS we consider for Maine in this paper. It is somewhat common for RPSs not tocreate new incentives in the first year but this usually disappears quickly.

25 Many states have multiple tiers of RPS requirements that act equivalently as additional RPS requirements. These secondand third tier RPSs expand the definition of what a renewable fuel is, often including things such as cogeneration, municipalsolid waste, and small hydroelectric facilities in addition to generation eligible for the first tier requirement. Since there is moregeneration available to meet these requirements, the price of compliance RECs is lower, often less than a quarter of the first tierprice. Thus, generators eligible to sell their RECs for compliance with the first tier will do so.

26 Since this calculation is based on historical capacity factors for all renewable generators, it is slow to reflect changes inefficiency gains. Alternatively, I use the capacity factors for each renewable resource taken from the Department of Energy andused in the Annual Energy Outlook (DeMeo and Galdo, 1997). This should reflect the capacity factor for new construction ofthat resource. This has a negligible effect on the results.


After computing this variable on the state level, it are aggregated up to the regional level by weight-ing by each state’s electricity consumption share in the region.27 Fig. 2 displays the this variable foreach state in a region and the aggregated regional requirement, rescaled from megawatts to gigawattsof capacity. This RPS Requirement variable is then averaged over the contemporaneous and futurevalues to construct the instrument. Therefore RPS Requirementit,t+5, averages the contemporaneousstringency with the stringency over the next five years to account for the fact that generators respondto the contemporaneous requirement and increasing future requirements.

Finally, I need to construct the complete price that renewable generators receive for the electricitythey produce. As mentioned above, there are two revenue streams for renewable generators underan RPS: the revenue from each megawatt hour of electricity they sell into the electricity market andthe revenue they receive from each REC associated with each megawatt hour of electricity that retailproviders retire at the end of each year to comply with the RPS. This means that the complete pricefor renewable generators is ptotal

t = pelectricityt + pREC

t . As stated earlier, I limit my sample to regions thathave a robust wholesale electricity market and REC market. Thus, I will be focusing on three regionsof the country: New England, the Mid-Atlantic states in PJM, and Texas.28

5. Data

To estimate the empirical models, I use data on all existing electricity generators and productionfrom the Energy Information Administration (EIA). I also collect data on REC and electricity prices, RPSrequirements, and other relevant policy variables.

Capacity data are taken from the Energy Information Administration. All generators with a name-plate capacity of at least 1 MW, are connected to the electric power grid, and are able to deliver powerare required to fill out form EIA-860. The data files include information about each generator includingits capacity, all fuels used during that year, the year and month the generator began operation, the yearand month of retirement, the city and state that the plant is located in, and basic information aboutthe owner. Though these data are reported annually to the EIA, they can be translated into monthlydata on total generating capacity since the data report the first month of operation for each plant. Iuse the data reported in these surveys from 1999 to 2007.

In these data, each generator provides detailed data on the type of fuel used during that year forelectricity generation, including up to six or more fuels that were used. I consider the first fuel listed asthe generator’s primary fuel.29 For the purposes of this paper, I aggregate these fuels into 17 categories.

In order to translate a RPS requirement that is usually in terms of percent of sales of electricity andto create each state’s weight in the region I use data from the Energy Information Administration’sstate historical tables on electricity sales. These data report the total megawatt hours sold in the entireelectric industry for a given state in a given year. I also use data on electricity generation aggregatedto the state × year level by the EIA. These aggregate data are based on EIA Form 906.

Data on the wholesale price of electricity were collected from each Independent System Operator’s(ISO) web site. ISOs publish data on the market-clearing price of electricity for many locations in eachregion for every hour of the day. Where available, I use the published regional weighted average pricefor each hour and then average the price over each calendar month. Some ISOs do not publish a regionalelectricity price, instead only publishing data for each location in the ISO. Where this is the case, I take

27 The weights are computed using a state’s consumption share in 2003 to keep them across time. Changing the weights tothe contemporaneous consumption share in each region does not change the results.

28 I exclude California from the analysis because until mid-2010, there was not a market for RECs since the California PublicUtility Commission required retail providers to purchase both renewable electricity and its attributes (essentially RECs) togethervia bilateral (private) contracts. Therefore, there is not a market price for RECs to use in the second stage. I exclude Midweststates since there is not a developed market for RECs. I also exclude New York since the New York State Energy Research andDevelopment Authority (NYSERDA) centrally procures the RECs for the entire state’s commitment through an annual biddingprocess. It is not clear that this processes elicits the same price due to possible market power on behalf of the NYSERDA. Resultsincluding the New York ISO will be used as a robustness check.

29 One-third of plants report using two fuels and less than 5% report using more than two fuels. Of the plants that list usingtwo fuels, only 6% of generators that are categorized as using a renewable fuel list a non-renewable fuel as their second fuel,concentrated among generators that are categorized as biomass, landfill gas, and municipal solid waste.


Fig. 3. State REC prices.

a simple arithmetic mean of the prices across all locations to form an hourly regional price and thenaverage this mean over the entire calendar month.

In order to compute the complete price that renewable electricity generators receive for theirelectricity, I need to add the price of renewable energy credits (RECs) to the price of the electricity. Ihave collected average annual prices for RECs in states that allows RECs to be purchased separatelyfrom electricity. These prices are gathered from public utility commission documents, other agenciesadministering a state’s RPS, or other governmental reports (PUC Bureau of Conservation Economicsand Energy Planning, 2007; New York State Energy Research and Development Authority, 2012; NewJersey Office of Clean Energy, 2010; Stern et al., 2009; Wiser and Barbose, 2007). The raw price data forREC prices can be seen in Fig. 3. The state REC prices exhibit distinctly regional variation confirmingthat the market for RECs is indeed regional.

The policy variables are constructed from information compiled at North Carolina State’s Databaseof State Incentives for Renewables & Efficiency (DSIRE). DSIRE has cataloged all state incentives forrenewable energy including the date they were enacted, when and if they were modified, as well asmany details about each policy. Where necessary, this information was supplemented by consultingstate statutes.

The variables that were constructed include the date that a particular renewable energy policywas passed by the legislature, when the policy began to bind (if different), and the implementationschedules for RPSs. In addition, information for each RPS regarding what fuels are eligible to meetthe requirements, and in some cases maximum capacities for eligible facilities, were taken from thisdatabase.

In addition to collecting data on state RPSs from DSIRE, I collect data about other policies that havebeen implemented in some states that could change the incentives for renewable electricity genera-tors. These policies include net metering, public benefits funds, mandated government purchases ofrenewable electricity, and rules requiring retail providers to offer renewable electricity to consumerfor a price premium. It is important to note that all RPSs in the sample require that any renewableelectricity purchased for these other policies be purchased in addition to the RPS requirement.

Table 1 displays summary statistics for the policy variables listed above. The top panel displayssummary statistics after aggregating state policies to the region level, and the bottom panel displaysthe RPS requirements during my sample period for individual states. A time series of the effective RPScapacity requirements through 2030 for both the state and regional level can be seen in Fig. 2.

The sample consists of the three previously discussed wholesale markets: PJM, New England, andERCOT. The data for each region begins in the first month operation of the wholesale market where theprice is reported. Therefore, the PJM data begins in January 1999, the New England data begins in May1999, and the ERCOT data begins in April 2001. This produces a complete sample of 293 region × monthobservations through the end of 2007.


Table 1Summary statistics.

Mean Standard deviation Min Max N

Active RPS in region 0.556 0.498 0 1 293Statuatory RPS requirement in region (%) 0.600 0.747 0 2.476 293Statuatory RPS requirement in region (MW) 606 769 0 2280 293Effective RPS requirement in region (MW) 514 678 0 2280 293Statuatory RPS requirement when active in region (%) 1.079 0.697 0 2.476 163Statuatory RPS requirement when active in region (MW) 1089 732 0 2280 163Effective RPS requirement when active in region (MW) 923 670 0 2280 163Renewable capacity (MW) 1392 814 59 3156 293Mandatory green power option in region 0.041 0.199 0 1 293Government purchases of green power in region 0.491 0.501 0 1 293Public benefits fund in region 0.724 0.448 0 1 293Net metering laws in region 0.724 0.448 0 1 293

Active RPS in state (%) 0.200 0.400 0 1 1848Statuatory RPS requirement in state (%) 0.339 0.882 0 4.92 1848Statuatory RPS requirement when active in state (%) 1.698 1.261 0 4.92 369

Just over half of the region–months in my sample have an active RPS in the region, with a meanregional renewable requirement of 0.6% renewable generation and a maximum of 2.5%. The meanregional RPS requirement, conditional on an RPS being enforced in the region, is just over 1% ofelectricity consumption coming from renewable generation. Taking a look at the state-level data, Iobserve just 20% of state-months in our sample with an active RPS, with an average requirement of1.7% renewable generation, conditional on an operational RPS.

I aggregate state policies to a regional policy by weighting each state’s RPS requirement by thatstate’s fraction of electricity consumption in the region. Other state level policy variables (public ben-efits funds, green power options, etc.) are aggregated in a similar fashion to this, except each variableis simply an indicator for each state, so the variables take on the cumulative fraction of electricityconsumption in the region covered by those policies.30

6. Results

The results from the first stage regression are displayed in Table 2. Column 1 begins by simplyregressing the logarithm of the total price for renewable electricity (electricity price + REC price) onthe logarithm of the average effective RPS requirement for renewable capacity in that region over thefollowing five years.31 As mentioned above, the average requirement is used since it is correlated withfuture stream of payments over the lifetime of the generator.32

We see that the measure of the stringency of an RPS is statistically and economically significant.Column 2 adds control variables that can take values between zero and one depending on the fractionof electricity consumption in the region that is covered by one of the policies. Column 3 allows forboth a one-off effect in the region and an increasing effect over time as more states in the regionadopt these policies. This flexible specification makes sense intuitively, since I expect that the moreexpansive these policies are, the larger in magnitude the effect should be.

Examining the other coefficients in column 3, the coefficients match my intuition about the direc-tion of the effect. I expect a positive effect on REC prices from government purchases of green power

30 For simplicity, the weights used are calculated as the state’s fraction of consumption in the region during 2003. This keepsthe policy and RPS variables weakly monotonic across time. It is unlikely that generators can accurately predict the smallvariations in electricity consumption across regions for them to take these fluctuations into account. Changing the year usedfor the weights or using contemporaneous weights do not change the results.

31 All specifications are robust to the number of years over which the effective requirement variable is averaged.32 One possibility, given the frequency of the data, is that generators are only responding to the price of electricity instead

of the total price. To address this concern, all regressions are also run using the yearly average electricity price in addition toincluding both year fixed effects to capture level changes in the price common across regions and month fixed effects to captureseasonal variation in electricity prices. These regressions are nearly identical to the results presented below.


Table 2First stage regression estimates.

Dependent variable: log(total renewable electricity price)

(1) (2) (3) (4)

Log(RPS Requirementt,t+5 years) 0.192** 0.304** 0.306** 0.227**

(0.039) (0.051) (0.049) (0.045)Gov’t Power Purchase (Frac. of Consumption) 0.113* 0.118* −0.027

(0.042) (0.063) (0.077)Public Benefits Fund (Frac. of Consumption) −0.028 −0.036 −0.071

(0.070) (0.074) (0.062)Net Metering (Frac. of Consumption) −0.083* −0.106* −0.154*

(0.035) (0.037) (0.049)I(Green Power Option) 0.141* 0.065

(0.057) (0.066)I(Gov’t Power Purchase) 0.046 0.078

(0.074) (0.103)I(Public Benefits Fund) 1.041* 1.151**

(0.317) (0.336)

Observations 293 293 293 390R2 0.58 0.59 0.60 0.55F-Test that excluded instrument equal to zero 24.94 35.05 39.47 25.62

OLS estimates. Estimates include region, year, and month fixed effects as well as region specific trends. Standard errors robustto heteroskedasticity and autocorrelation using the Newey–West method.

* p < 0.05.** p < 0.01.

since this increases the demand for green power without decreasing the demand for RECs. Most statesdo not allow green power purchased through government purchases to count toward retail providers’REC fulfillment obligations; instead these purchases simply add buyers into the green electricity/RECmarket.

The last row of each column shows the F-statistic of a test that the excluded instrument in theregression is zero. All three columns reject the null that both coefficients are zero at all usual lev-els of significance. All standard errors in this table and the second stage regressions are estimatedusing Newey–West heteroskedasticity and auto-correlation robust standard errors. The number oflags included in the auto-correlation estimation was chosen using the procedure suggested by Neweyand West (1994).

Table 3 displays the results from the second stage regression that estimates the price elasticity ofsupply for renewable electricity generators.33 The variable of interest in this set of regressions is thefirst row, log( ̂Total Price). This is the predicted price of RECs plus the predicted price of electricity inthe region given the shift in demand induced by the stringency of the state RPS estimated in the firststage.

Column 1 shows a baseline specification without any additional controls, with columns 2 and 3progressing to a full set of flexible controls for other policies aimed at renewable generators. Thepreferred estimate is in column 3 with a price elasticity estimate of 2.67. In the next section I will usethis estimate to bound the cost of focusing on reducing greenhouse gas emissions through only anRPS-style policy.

Column 4 of Tables 2 and 3 shows the results from an regression that is identical to the column3 regression except that they also include observations from New York that had been previouslyexcluded due to the central procurement of RECs in New York. As can be seen, the inclusion of NewYork has only a small effect on the elasticity estimate.

33 Each specification in Table 3 uses the first stage regressions from the same numbered column in Table 2.


Table 3Second stage regression estimates.

Dependent variable: log(renewable generating capacity)

(1) (2) (3) (4)

Log ̂(Total Price) 2.427* 1.998** 2.670** 2.593**

(0.753) (0.407) (0.473) (0.576)Gov’t Power Purchase (Frac. of Consumption) 0.439* −0.161 −0.109

(0.143) (0.205) (0.152)Public Benefits Fund (Frac. of Consumption) 0.746** 0.691** 0.684**

(0.157) (0.185) (0.198)Net Metering (Frac. of Consumption) 0.495** 0.413** 0.475**

(0.063) (0.072) (0.080)I(Green Power Option) −0.346 −0.379*

(0.196) (0.162)I(Gov’t Power Purchase) 0.849** 0.906**

(0.211) (0.204)I(Public Benefits Fund) −3.945** −4.051**

(0.635) (0.753)

Observations 293 293 293 390First stage F-statistic 24.94 35.05 39.47 25.62

OLS estimates. Estimates include region, year, and month fixed effects as well as region specific trends. Standard errors robustto heteroskedasticity and autocorrelation using the Newey–West method.

* p < 0.05.** p < 0.01.

7. Policy implications for RPSs as a CO2 abatement tool

In this section, I use my estimates of the long-run supply elasticity to estimate the cost of decreasingcarbon dioxide emissions in states covered by the Regional Greenhouse Gas Initiative (RGGI) bypursuing carbon dioxide reductions exclusively through an RPS.

RGGI is a cap-and-trade program established in the northeastern United States to reduce green-house gas emissions from electric power plants to 10% below (approximately) 2005 levels by 2018.There were originally ten states participating in RGGI, including all of the states in the New Englandwholesale electricity market, New York, and parts of the PJM wholesale electricity market.34 In thesestates, RGGI regulates all fossil fuel fired electricity generators in the 10 states that have a capac-ity of 25 MW or more. Each quarter, new emissions permits are auctioned with approximately 50%of the auction proceeds being invested in energy efficiency and approximately 7% being invested inrenewable generation projects (Hibbard and Tierney, 2011).35

The states in RGGI had a total of 184 million tons of carbon dioxide emissions from the electricitysector in 200536 (RGGI, 2010). Beginning in 2009 and continuing through 2014, carbon dioxide emis-sions are capped at the baseline level of 188 million tons. Beginning in 2015, the carbon dioxide capis reduced by 2.5% annually until the final goal is met after 2018 when carbon dioxide emissions arereduced by 10% from the original cap. Under the RGGI cap-and-trade program, emission reductionsare most likely to come from using a different mix of fuel to produce electricity (more natural gas andrenewable sources, less coal) and energy efficiency investments. This suggests that the carbon pricein the RGGI market provides a good cost estimate of reducing greenhouse gas emissions within theelectricity sector through ways that differ from an RPS’s exclusive reliance on switching productionto renewable sources.

34 The original ten states participating in RGGI were: Connecticut, Delaware, Massachusetts, Maryland, Maine, New Hampshire,New Jersey, New York, Rhode Island, and Vermont. The major states in the PJM wholesale electricity market that are notparticipating in RGGI are Ohio, Pennsylvania, Virginia, and West Virginia. New Jersey dropped out of RGGI in 2011.

35 The balance of the funds goes toward the state general funds, direct bill assistance, education and outreach, and programadministration.

36 The baseline level of carbon dioxide emissions that the RGGI reductions are based on is 188 million tons of CO2.


I will examine two different levels of carbon reduction produced by a northeastern RPS, a 2.5%reduction of 2005 CO2 levels and a 10% reduction of 2005 CO2 levels, to compare to the cost of carbondioxide abatement through RGGI. In order to estimate the cost of carbon dioxide abatement under anRPS, in addition to knowing the price elasticity of supply or renewable generation that I estimated inthe previous section, I need to make a few assumptions. Whenever possible, I will make assumptionsthat will make an RPS look as favorable as possible (lowest cost of carbon dioxide abatement) so myestimates will be a lower bound on the cost of CO2 abatement under an RPS.

First, I need to make an assumption about what fossil fuel the new renewable capacity will bedisplacing. Coal fired generation is the most CO2 intensive fuel, therefore, to make an RPS look asattractive as possible, I will assume that each megawatt hour of renewable generation produces nocarbon dioxide and replaces a megawatt hour of coal production.37 To the extent that renewablegeneration produces carbon dioxide or displaces generation other than coal, an RPS would have ahigher cost of carbon abatement than I estimate.

Secondly, I need to assume a capacity factor (the fraction of the year that a generator produceselectricity) for the new renewable generation built to meet the RPS. In the previous analysis, theaverage, region specific capacity factors for renewable generation were all below 38% in every year.Since these are not capacity factors for new renewable generators, which may be higher, I use anweighted average of projected capacity factors for new generation over the period 2010–2020, in themiddle of many of the RPSs (Transparent Cost Database, 2012).38

Finally, I need to assume how the demand for electricity changes in the future. I will assume thatelectricity consumption does not change from the amount consumed in 2005. Given the decrease inelectricity consumption in the northeast due to the 2008 recession, this may slightly overestimate2015 consumption and will cause my estimates to be slightly high.39

In 2005, the total renewable generating capacity in the northeast was 2932 MW. If every megawatthour of renewable generating capacity displaces a megawatt hour of coal generation, a 36% increasein renewable generating capacity would achieve a 2.5% decrease in CO2 emissions in RGGI. Using mypreferred elasticity estimate of 2.67, this means renewable generators would need a price increase ofapproximately 14% in order to be profitable. Since the total price of electricity for renewable generatorsaveraged $82 per MWh over my sample, a 14% increase implies that renewable generators would needto receive approximately $93 per MWh to be profitable. This implies a marginal cost of CO2 abatementof over $11.

In addition to examining this extreme case that each megawatt hour of renewable electricity pro-duced replaces a megawatt hour of coal production, I examine two other scenarios. First, I use theestimates produced by Graff Zivin et al. (2012) of the marginal CO2 emissions from electricity pro-duction in the NPCC NERC region, which covers New York and New England and parts of Canada.Using these marginal emissions estimates, a 2.5% decrease in CO2 emissions requires a 63% increase inrenewable generating capacity. Using the preferred elasticity estimate, renewable generators wouldneed a 24% in price to enter the market and a marginal cost of CO2 abatement of $34 per ton of CO2.

A second alternative assumption, is that each megawatt hour of renewable electricity producedreplaces the carbon emissions of an “average” megawatt hour.40 This assumption provides anotheruseful point of comparison since final RPS goals represent large changes in generation. Therefore, usingmarginal emissions may be the incorrect emissions estimate. Under this assumption, to achieve a 2.5%decrease in CO2 emissions requires a 61% increase in renewable generating capacity with an impliedmarginal cost of CO2 abatement of $31 per ton of CO2.

37 For renewable resources, such as wind, that are not completely predictable, there is some carbon emissions from usingthese sources since more generators need to be on standby in case the electrical output is less than expected. Compoundingthis, usually these standby generators need to ramp up their output quickly which creates higher than average emissions perMMBTU consumed. I abstract from both of these issues.

38 Mean projected capacity for wind is 44%, 22% for solar, 85% for biomass, and 90% for geothermal.39 Electricity consumption in the northeast has grown by approximately 2.5,million MWh annually between 1990 and 2008.

Electricity consumption in 2005 was 282 million MWh.40 Average is defined as the total tons of CO2 produced by the electricity industry divided by the number of megawatt hours

consumed in a year.


Table 4Cost of CO2 abatement from an RPS.

Replace coal Replace marginal emissions Replace average fuel

2.5% reduction in CO2 levelsPercent increase in renewable capacity 36% 63% 61%Cost of CO2 abatement $11.20 $34.05 $31.30

[8.67, 15.83] [26.35, 48.10] [24.22, 44.22]

10% reduction in CO2 levelsPercent increase in renewable capacity 145% 253% 243%Cost of CO2 abatement $44.82 $159.83 $156.62

[34.68, 63.61] [105.39, 192.38] [96.89, 176.87]

All CO2 reductions are measured from the 2005 baseline levels, similar to the Regional Greenhouse Gas Initiative. Ranges ofcost of abatement prices are shown in brackets using a 90% confidence intervals of the elasticity estimate. Marginal emissionsestimates taken from Graff Zivin et al. (2012).

The final goal of RGGI is to reduce CO2 emissions by 10% from their 2005 levels. Based on anextrapolation that is well out of my sample, my elasticity estimate suggest that if all of the additionalrenewable generation displaced coal generation to meet an equivalent RPS, there would need to bea 145% increase in renewable capacity in RGGI states. This implies that the cost of CO2 abatementwould be $45. Using the two alternative, and more realistic, assumptions about the emissions that therenewable capacity would displace, the cost of CO2 abatement would be nearly $160 with nearly a250% increase in renewable generating capacity. A summary of these results can be found in Table 4.

These estimates of the marginal cost of CO2 abatement, ranging from $11 to $160 are substantiallyhigher than the cost of carbon abatement under the RGGI cap-and-trade system. Currently the RGGICO2 emissions permits being traded and auctioned are for the years when CO2 emissions are cappedat a level just over 2005 CO2 emissions. However, since these permits can be banked indefinitely intothe future, they give us a window into the expected marginal cost of CO2 abatement in the future. Atthe end of 2012, the price of emissions permits were approximately $2 per ton of CO2, and permitswere trading at approximately $3 per ton of CO2 in early 2009 with an average price of $2.37 per tonof CO2 over four years of trading. The maximum of RGGI permit price over this period was $3.72 perton of CO2 in 2009 shortly after the market began operation and a minimum of $1.90 in 2010 with2011 and 2012 consistently near $2.00. Since permits purchased today can be used to comply withRGGI indefinitely into the future, the current price is indicative of future CO2 abatement costs.41

However, one of the clear problems with comparing these estimates to the RGGI price is that RGGIpermits began trading in 2008, after the sample period. Moreover, the RGGI prices incorporated twoimportant pieces of information: the discovery of large economically recoverable natural gas reservesthrough the use of hydraulic fracturing and a decrease in electricity demand due to the 2008 recession.

The price of natural gas is likely important to the RGGI permit price for two reasons. First, naturalgas is usually the fuel on the margin and thus natural gas generators tend to set the market-clearingprice for electricity. Secondly, as mentioned earlier, one expected common RGGI compliance strategywas to switch from using coal-fired generation to natural gas-fired generation. To demonstrate thelarge effect that the new natural gas reserves have had on natural gas prices, consider that in June2008 the price of natural gas was over $12 per thousand BTU. Natural gas prices dropped drasticallyin a one year span after June 2008 to $4 per thousand BTU and has stayed in that range since.42 Incontrast, over my sample period the price of natural gas ranges from $2.40 to nearly $11 with a meanof $5.88 per thousand BTU. RGGI permits began trading in late 2008 after most of the fall in naturalgas prices. Similarly, the 2008 recession likely affects RGGI prices through electricity demand.

41 RGGI has two mechanisms built into its structure to curb potential price volatility. If the average price of CO2 permits isabove $7 for a 12-month period, more permits are released and generators are allowed to meet more of their obligations throughoffsets. If the average price of CO2 permits is above $10 for a 12-month period, a second mechanism is triggered and even moreoffsets can be used to meet CO2 obligations. It is widely expected that neither of these trigger events will happen, suggestingthat it is unlikely that the marginal cost of CO2 abatement is below $10 in the electricity sector in the states in RGGI.

42 The mean natural gas price over the time period that RGGI permits have been trading is $4.94.


To make the cost of CO2 abatement more comparable to RGGI prices, I regress RGGI prices on theprice of natural gas and electricity demand as well as a set of calendar month fixed effects and a dummyto indicate the months at which the RGGI price was equal to the auction reserve price.43 Using theseregression results, I extrapolate what the RGGI price would have been over my sample period. Thepredicted RGGI price at the end of 2007 is $3.00 (95% CI of $1.87–4.13). Moreover, the mean predictedRGGI price between 2002 and 2007 is $2.73. These estimates are similar in magnitude to the RGGIprices that were expected during the first few years RGGI was in force.44

Comparing the cost of CO2 abatement from my elasticity estimates to the extrapolated prices inRGGI suggests that within the electricity sector, the state level RPSs that have been passed are anexpensive way to decrease carbon dioxide emissions, costing between nearly four and eleven timesmore to reduce CO2 emissions by 2.5% than from a cap-and-trade program on the electricity sectorin the northeastern United States. However, it should be noted that CO2 abatement is not the onlypolicy goal of RPSs, though it is usually the primary goal. Other policy goals that are often cited whenan RPS is introduced are developing new renewable technologies in a state, spur employment thoughthe creation of local “green” jobs, and reduce a state’s dependence on foreign energy inputs. Thus,the excess cost of CO2 abatement from RPSs can be viewed as the financial cost of these other policyobjectives.

Since these estimates are constructed using RPSs that were being enforced in 2007 and before, itis possible that the cost of carbon dioxide abatement from other RPSs is lower than estimated here.However, taken as a whole, the states studied in this paper include approximately 45% of the totalRPS requirements in the US as of 2012 and account for 30% of electricity consumption in the US. Incontrasting, they account for just 18% of the renewable potential in the United States45 (Lopez et al.,2012). Thus, unless there is a drastic rearranging of RPS requirements in the United States, RPSs arelikely to be an expensive method of reducing CO2 emissions in the United States. Moreover, sinceboth RGGI and an RPS focus just on the electricity sector, the marginal cost of CO2 abatement in theeconomy as a whole has the potential to be lower than either of these estimates since there may becheaper ways to decrease CO2 emissions in other sectors of the economy.

8. Conclusion

This paper estimates the long-run supply elasticity of renewable electricity generating capacity.The price elasticity is an important parameter for policy makers to know since many states haveintroduced aggressive RPSs to increase the share of renewable electricity sold in their states. Also, theUS Congress has considered legislation on multiple occasions that would introduce a federal RPS. SinceRPSs’ main goal are to reduce carbon dioxide emissions, it is important to know the cost of the carbonabatement from these policies relative to other ways that could reduce carbon dioxide emissions.

In order to estimate this parameter, I use the policy variation in the implementation schedule ofrenewable portfolio standards across states that have restructured electricity markets. Since moststate RPSs can be met by renewable generation located anywhere in the wholesale electricity market,I aggregate individual state policies into region-level renewable portfolio standards. Each year, eachstate’s RPS increases in its stringency, creating the variation that I use to estimate the long-run supplyelasticity. In my preferred specification, I estimate that a 1% increase in the total price received forrenewable electricity (price of electricity plus the price of the renewable energy credit) results in a2.67% increase in the supply of renewable generation.

Politicians appear to prefer using RPS policies to those of broader policies such as cap and trade or acarbon tax. Part of the attraction is likely that the costs of this method of carbon dioxide abatement areless transparent to voters. However, these policies still come with a cost. My estimates suggest that the

43 The reserve price is indexed to inflation and was originally set at $1.863.44 See Center for Integrative Environmental Research (2007), Gittell and Magnusson (2008), and IFC Consulting (2006) for

studies of price expectations.45 Fifteen percentage points of this renewable potential comes from Texas, whereas Texas’s RPS makes up just 15 percentage

points of RPS requirements. This means all of the other states in this paper account for 30% of the total RPS requirements in theUS but account for just 3% of the renewable potential in the US.


cost of abating the last ton of carbon dioxide from an RPS in the northeastern US to reduce emissionsby 10% from their 2005 levels (approximately equal to a 10% RPS) would cost between $45 and $160per ton of carbon dioxide, depending on the type of fossil generation that the renewable generationwas replacing. My estimate for the cost of CO2 abatement is nearly four times more expensive thanthe maximum price of CO2 under the regional cap-and-trade program for the electricity sector usingthe most optimistic assumptions. Therefore, residents of states with RPSs paying an extremely highpremium for carbon dioxide, even though they appear to be more politically palatable policies.

References

Adelaja, S., Hailu, Y., January 2008. Projected Impacts of Renewable Portfolio Standards on Wind Industry Development inMichigan. Land Policy Institute, Michigan State.

Center for Integrative Environmental Research, January 2007. Economic and energy impacts from Marylands potential partici-pation in the Regional Greenhouse Gas Initiative. Technical Report. Maryland Department of the Environment.

DeMeo, E.A., Galdo, J.F., December 1997. Renewable energy technology characterizations. Technical Report. US Department ofEnergy.

Fischer, C., 2010. Renewable portfolio standards: when do they lower energy prices? The Energy Journal 31 (January (1)),101–120.

Fowlie, M.L., 2009. Incomplete environmental regulation, imperfect competition, and emissions leakage. American EconomicJournal: Economic Policy 1 (July (2)), 72–112.

Gittell, R., Magnusson, M., 2008. Economic impact in New Hampshire of the Regional Greenhouse Gas Initiative (RGGI): anindependent assessment. Technical Report. University of New Hampshire, Whittemore School of Business and Economics.

Graff Zivin, J.S., Kotchen, M., Mansur, E.T., 2012. Spatial and temporal heterogeneity of marginal emissions: implications forelectric cars and other electricity-shifting policies. NBER Working Paper 18462.

Hibbard, P.J., Tierney, S.F., 2011. Carbon control and the economy: economic impacts of RGGI’s first three years. The ElectricityJournal 24 (December (10)), 30–40.

Holland, S.P., Hughes, J.E., Knittel, C.R., 2009. Greenhouse gas reductions under low carbon fuel standards? American EconomicJournal: Economic Policy 1 (April (1)), 106–146.

ICF Consulting, 2006. RGGI package scenario. Technical Report.Joint Committee on Taxation, May 2009. Tax expenditures for energy production and conservation. Technical Report. Senate

Committee on Finance.Langniss, O., Wiser, R., 2003. The renewables portfolio standard in Texas: an early assessment. Energy Policy 31 (January),

527–535.Lopez, A., Roberts, B., Heimiller, D., Blair, N., Porro, G., 2012. U.S. renewable energy technical potentials: a GIS-based analysis.

Technical Report. National Renewable Energy Laboratory.Lyon, T.P., Yin, H., 2007. Why do states adopt renewable portfolio standards? An empirical investigation. The Energy Journal 31

(May (3)), 133–158.Menz, F.C., Vachon, S., 2006. The effectiveness of different policy regimes for promoting wind power: experiences from the

states. Energy Policy 34 (April), 1786–1796.Metcalf, G.E., 2009. Tax policies for low-carbon technologies. National Tax Journal 62 (3), 519–533.New Jersey Office of Clean Energy, 2010. New Jersey’s renewable portfolio standard rules: 2009 annual report. Technical Report.

New Jersey Board of Public Utilities.New York State Energy Research and Development Authority, 2012. The New York state renewable portfolio standard perfor-

mance report. Technical Report. New York State Energy Research and Development Authority.Newey, W.K., West, K.D., 1994. Automatic lag selection in covariance matrix estimation. Review of Economic Studies 61 (January),

631–653.Palmer, K., Burtraw, D., 2005. Cost-effectiveness of renewable electricity policies. Energy Economics 27 (November), 873–894.PUC Bureau of Conservation Economics and Energy Planning, 2008. 2007 annual report: Alternative Energy Portfolio Standards

Act of 2004. Technical Report. Pennsylvania Public Utility Commission.Rabe, B., June 2006. Race to the Top. Pew Center on Global Climate Change, Arlington, VA.Rabe, B., 2008. States on steroids: the intergovernmental odyssey of American climate policy. The Review of Policy Research 25

(September (2)), 105–128.Regional Greenhouse Gas Initiative, September 2010. Fact sheet: the Regional Greenhouse Gas Initiative (RGGI). Technical

Report.Shrimali, G., Kniefel, J., 2011. Are government policies effective in promoting deployment of renewable electricity resources?

Energy Policy 39 (September (9)), 4726–4741.Stern, F., Wobus, N., Pater, J., 2009. New York renewable portfolio standard market conditions assessment. Technical Report

10430. New York State Energy Research and Development Authority.Transparent Cost Database, 2012. Open Energy Information (en) (accessed September 17, 2013) http://en.openei.org/

wiki/Transparent Cost DatabaseWiser, R., Barbose, G., April 2008. Renewables portfolio standards in the United States – a status report with data through 2007.

Lawrence Berkeley National Laboratory Working Paper 154-E.Wiser, R., Bolinger, M., Barbose, G., 2007. Using the federal production tax credit to build a durable market for wind power in

the United States. Electricity Journal 20 (November (9)), 77–88.Wiser, R., Porter, K., Grace, R., 2005. Evaluating experience with renewables portfolio standards in the United States. Mitigation

and Adaptation Strategies for Global Change 10 (April), 237–263.Yin, H., Powers, N., 2010. Do state renewable portfolio standards promote in-state renewable generation. Energy Policy 38

(February (2)), 1140–1149.

http://refhub.elsevier.com/S0928-7655(14)00018-9/sbref0005

















































































































































































































































































































































































































http://en.openei.org/wiki/Transparent_Cost_Database

http://en.openei.org/wiki/Transparent_Cost_Database









































































The cost of carbon dioxide abatement from state renewable portfolio standards

Documents

Transcript of The cost of carbon dioxide abatement from state renewable portfolio standards