Page 1
The Disparate Influence of State Renewable Portfolio Standards (RPS) on U.S. Renewable Electricity Generation Capacity
Karen Maguire Assistant Professor
Department of Economics and Legal Studies Oklahoma State University
Stillwater, OK
Abdul Munasib Research Scientist
Department of Agricultural & Applied Economics University of Georgia
213 Stuckey Building, 1109 Experiment Street, Griffin, GA 30223
2015 OKSWP1502
Economics Working Paper Series Department of Economics
OKLAHOMA STATE UNIVERSITY http://spears.okstate.edu/ecls/
Department of Economics Oklahoma State University Stillwater, Oklahoma
339 BUS, Stillwater, OK 74078, Ph 405-744-5110, Fax 405-744-5180
Page 2
1
TheDisparateInfluenceofStateRenewablePortfolioStandards(RPS)onU.S.RenewableElectricityGenerationCapacity
KarenMaguireAssistantProfessor
DepartmentofEconomicsandLegalStudiesOklahomaStateUniversity
327BusinessBuilding,Stillwater,OK74075Phone:405‐744‐5112
E‐mail:[email protected]
AbdulMunasibResearchScientist
DepartmentofAgricultural&AppliedEconomicsUniversityofGeorgia
213StuckeyBuilding,1109ExperimentStreet,Griffin,GA30223Phone(770)229‐3419
E‐mail:[email protected]
Abstract
Several papers have used panel data analyses to examine the effectiveness of U.S. state‐levelRenewablePortfolioStandards(RPS)inpromotingrenewablecapacitydevelopment,but the findings are inconclusive. Estimation of average treatment effects, however, canmask the fact that RPS policies across states are disparate and the treatment states areheterogeneous. We use the Synthetic Control Method (SCM) to conduct individual casestudies of the early adopter states. Our findings indicate that the impact of RPS variedacrossstates.WefindTexastobeuniqueamongtheseearlyadoptersinthatRPSinTexashasledtoincreasedrenewablecapacity.Keywords: Renewable portfolio standard (RPS), renewable energy, electricity, synthetic
controlmethod(SCM)JELclassification:Q4,Q42,Q48,H7
Page 3
2
I.Introduction
As of January 2012, 29 U.S. states and theDistrict of Columbia had enacted a Renewable
PortfolioStandards(RPS)orothermandatedrenewableenergypolicies.RPSrequirethatelectricity
producerssupplyaportionoftheirelectricityfromdesignatedrenewableresourcesbyaspecified
futuredate.TheadoptionofRPS ismotivatedbyacomplexsetofpoliticalandeconomic factors,
includingincreasingconcernsoverclimatechangeandenergysecurity(YiandFeiock2012).While
severalpolicieshavebeenproposedtoaddresstheseconcerns,RPSisoneofthemost frequently
advanced policies to promote renewable energy development for electricity generation (Fischer
2010).WeexaminewhetherRPSishavingtheintendedeffectof increasingrenewablegeneration
capacity.
Severalpapershaveimplementedpaneldataanalysestostudytheroleofrenewableenergy
policiesinpromotingrenewablesdevelopment.1They,however,donotprovideanyconsensus.Yin
andPowers(2010)findthatRPShasapositiveinfluenceonthepercentageofnon‐hydrorenewable
generatingcapacity,but the finding ispredicatedon theconstructionofanRPSstringency index.
ShrimaliandKniefel(2011),ontheotherhand,foundanegativeimpactofRPSontheratioofnon‐
hydrorenewablecapacityovertotalnetgeneration.Carley(2009)focusesongenerationandfinds
that in theyearsafterRPSadoptionanadditionalyearofRPShasapositiveeffect,althoughRPS
implementation has no predictive power. Delmas and Montes‐Sancho (2011) analyzed capacity
rather than generation and found that RPS led to declining renewable electricity capacity.
Additionally, a number of studies on wind capacity found no impact of RPS. Hitaj (2013), for
instance,providesacounty‐levelanalysisandfindsthatRPSdidnothaveasignificantinfluenceon
1Carley (2009): 48‐state1998‐2006panel,Delmas andMontes‐Sancho (2011):panel of 650utilities from48‐statesover1998‐2007,Hitaj (2013): county‐level1998‐2007panel,Maguire (2014): state‐level1994‐2012panel, ShrimaliandKniefel(2011):50‐state1991‐2007panel,YinandPowers(2010):50‐state1993‐2006panel.
Page 4
3
wind capacity, andMaguire’s (2014) state‐level analysis also concludes that RPS did not have a
significanteffectonwindcapacity.
Theempiricalliteraturediscussedabovehasgenerallyfailedtofindconclusiveevidenceof
anaveragetreatmenteffectofRPSonrenewablesadoptionacrossRPSstates.Thishighlights the
needforanalysesthataccommodatethepossibilityoftreatmentheterogeneity(Keeleetal.2013),
particularlybecauseRPSareuniquestate‐levelpolicies.Estimationofaverageeffectscanmaskthe
factthatadopterstatesareheterogeneousandstateRPSpoliciesaredisparate.Theassumptionofa
uniform effect of RPS across states can be quite restrictivewhen the states differ in their policy
environment, electricity market characteristics, renewable resource potential, likelihood of
successfulimplementationoftheirRPS,andahostofobservedandunobservedcharacteristics.
TreatingdisparatestatelevelRPSasauniforminterventionisalsoinappropriate.RPSvary
in the amount of electricity generation that must be supplied from renewables, the types of
allowablerenewables,theyearofrequiredimplementationofthefinalmandate,andthemagnitude
and the timing of intermediatemandates.RPS also differ in thenature of theRenewableEnergy
Credit(REC)tradingmarkets,andthedegreeandscopeofrestructuringrequirements(seesection
II.3formoredetails).We,therefore,adoptacasestudyapproachtoexaminetheeffectofastate’s
RPSonitsrenewablecapacity.Weexaminetheperiod1991‐2008andfocusontheearlyadopter
states(seeAppendixAforalistofRPSstatesandfinalmandates).2Oursetoftreatmentstatesare
Nevada(1997),Connecticut(1998),NewJersey(1999),Maine(1999),Texas(1999)andWisconsin
2Theearliestavailablestate‐leveldataforgenerationcapacityis1990.Startingattheendof2008,fiveadditionalstatesadoptedRPS.Extendingouranalysisbeyond2008,therefore,wouldsignificantlyshrinkthesizeofthedonorpool.
Page 5
4
(1999),states thatenactedRPSbetween1997and2000.3Ouroutcomevariableof interest is the
generationcapacityofthemodernrenewables:wind,solar,geothermal,andbiomass.4
Weemploy theSyntheticControlMethod (SCM) for comparativecasestudies (Abadieand
Gardeazabal2003,Abadieetal.2010) toestimate the impactofRPS ineachof thesestates.SCM
constructsauniquecounterfactual(or‘synthetic’)foreachRPS(treatment)stateusingaweighted
averageofthenon‐RPS(control)statesbasedonasetofpre‐intervention(pre‐RPS)characteristics.
Byexaminingeachstateasastand‐alonecasestudyweareabletoallowforheterogeneouseffects
ofRPS.
Wefocusonlyonearlyadopterstates(i.e.,statesthatenactedRPSbetween1997and2000)
in order to allow for sufficient post‐intervention years to capture the effect of RPS.Unlike other
policies such as changes in gun laws or driving restrictions, RPS does not become immediately
binding on its effective date. The renewable mandates are implemented years after the RPS
effectivedatethroughaseriesofintermediategoalsandmandatesleadinguptothefinalmandate.
For instance, Nevada enacted RPS in 1997, and updated the policy in 2001 to establish the
minimum requirement that 2 percent of electricity be supplied from eligible renewable sources,
increasing every two years and culminating in a 15 percent mandate by 2013.5 In Texas, RPS,
passed in 1999, had intermediatemandates in 2002 and 2007with their finalmandate initially
bindingin2010andthensubsequentlyamendedto2015.Asimilarpatternisobservedintheother
RPSstateswherethefinalmandateiseffectiveonafuturedateprecededbyaseriesofintervening
targets. 3IowaistheonlystatethatpassedRPSbefore1997.ButitpasseditsRPSin1983,whichfallsoutsideourdatarange.4 Hydroelectric generation capacity is not considered a modern renewable resource and is excluded. Although itconstitutes52%ofrenewableelectricitygenerationintheU.S.in2013,becausemosthydroelectriccapacitywasaddedprior to the mid‐1970s it is not a newly developed resource.(http://www.eia.gov/energy_in_brief/article/renewable_electricity.cfm)5NevadaRPSwassignificantlyrevisedagainin2009,whichfallsbeyondourstudyperiod.
Page 6
5
OurSCMestimatesshowthattheimpactofRPSindeedvariesacrossstates.Texasisunique
amongtheearlyadopterstates in thatwefindapositive impactofRPSonrenewablecapacity in
Texas.WithinadecadeafterenactingRPS,Texasinstalledmorewindgenerationcapacitythanany
otherstate.6AswediscussindetailinsectionsII.2‐II.4theenergymarketcharacteristicsofTexas
are also quite unique: Texas is the only early adopter state with substantialmodern renewable
potential.Texasisalsotheonlymainlandstatewithitsowngrid,anditsRPS,specifiedintermsof
capacityandnotgeneration,isatypical.
Inwhatfollows,weprovidesomebackgroundinformationontheU.S.electricitymarketand
describetheRPScharacteristicsoftheearlyadopterstatesinsectionII,presentabriefdescription
oftheempiricalmethodologyinsectionIII,describethedatainsectionIV,anddiscusstheresultsin
sectionV.SectionVIconcludes.
II.RenewableGeneration,ElectricityMarkets,andRenewablePortfolioStandards
II.1.Renewablegeneration
Renewableenergysourcesprovided13percentof totalU.S.electricitygeneration in2013,
49percentofwhich is frommodern renewables;wind,biomass, geothermal, and solar, i.e., non‐
hydroelectric sources.Today, theUnitedStatesproducesmoreelectricity fromnon‐hydroelectric
renewablesourcesthananyothercountry,ChinaandGermanyranksecondandthird.7TheEnergy
InformationAssociation(EIA)predictsthatbetween2013and2040,non‐hydroelectricrenewables
willaccountfor24percentoftheoverallgrowthintheUnitedStateselectricitygeneration.Solaris
expectedtoincreasefrom8GWin2012to48GWby2040,whilewindispredictedtoincreasefrom 6In2013,Texasaccountedfor22percentofthe167millionMWhoftotalpowergeneratedfromwindnationwide.IfTexaswereacountryitwouldbesixthintheworldinwindcapacityfollowingChina,theUnitedStates,Germany,Spain,andIndia.SeeHurlbut(2008),EIA‐PTC:http://www.eia.gov/todayinenergy/detail.cfm?id=8870,EIA‐Texas:http://www.eia.gov/todayinenergy/detail.cfm?id=15851,EIA:http://www.eia.gov/state/?sid=TX,ERCOTTime‐line:http://www.ercot.com/about/profile/history,andOfficeoftheGovernor:www.TexasWideOpenForBusiness.com.7http://www.eia.gov/todayinenergy/detail.cfm?id=16051
Page 7
6
60GWto87GWoverthesameperiod.Inaddition,geothermalcapacityispredictedtotripleand
biomass capacity is predicted to double. Finally, modern renewable generation is predicted to
exceedhydroelectricgenerationandcomprisetwo‐thirdsofallrenewablegenerationby2040.8
II.2.ElectricityMarket
The electricity system in the United States consists of three regions: the Eastern
Interconnection, the Western Interconnection, and the Texas Interconnection. Grid connectivity
withinan interconnectionenablesutilities to importandexportgenerationacrossstates.9Within
the Interconnections, there are nine Independent System Operators (ISOs) and Regional
Transmission Organizations (RTOs) that coordinate the trading of electricity generation across
states (SeeFigure1).Theyprovide therates, termsandconditions for thewholesalemarketand
transmissionwithintheregion.
RenewableEnergyCredits(REC)marketsallowforthetradingofrenewableenergybetween
utilitieswithin a particular region.10REC aredesigned toprovide an accurate account of eligible
renewableenergyproduction,andtobetradablebetweenproducersandretailers.Forexample,in
NewEngland,theISONewEnglandRTOcoordinatesthetradingofrenewablegeneratedelectricity
across states using REC.11 Because in these states utilities are allowed to import and export
renewablegenerationfromotherstates,utilitiesmayimportratherthanaddadditionalrenewable 8http://www.eia.gov/forecasts/aeo/MT_electric.cfm#cap_natgas9http://www.un.org/esa/sustdev/publications/energy/chapter2.pdf.10Arizona,Nevada,TexasandWisconsinweretheearlieststatestoallowfororrequiretheuseoftradableRECtomeetRPS.11PowergeneratedfromrenewableresourcesisusedtocreateREC,whicharemeasuredinenergyunits.Forinstance,oneRECmayrepresent1MWhofqualifiedrenewableenergy.TheexistingRECmarketsandtrackingsystemsserveadistinct region: theNEPOOLGeneration InformationSystem(NEPOOLGIS)supportsasix‐statearea inNewEnglandcomprisingtheISONewEnglandcontrolarea,thePJMGenerationAttributeTrackingSystem(GATS)supportsthePJMcontrolarea,whichcovers13statesandtheDistrictofColumbia,whiletheERCOTRECprogramonlyoperatesinTexas.See(Doot,Belval,andFountain2007) formoredetails.TheNewEnglandISOwasestablishedbytheFederalEnergyRegulatory Commission (FERC) in 1997 andwas designated as an RTO in 2005, giving the organization additionalauthority over the regional grid (http://www.iso‐ne.com/about/what‐we‐do/in‐depth/industry‐standards‐structure‐and‐relationships).
Page 8
7
capacityifimportingisalow‐costalternativetomeettheirRPSmandates.Conversely,itmayalso
bemorecosteffectiveforautilitytobecomeanexporterofrenewablegeneration.Accordingtothe
National Renewable Energy Laboratory (NREL), “The primary regionalmarkets for REC exist in
New England and the Mid‐Atlantic states” (Heeter and Bird, 2010, p.6). The NEPOOL_GIS REC
tradingmarketfortheNewEnglandregionbeganin2002,whilethePJM‐GATSRECtradingmarket
servingtheMid‐Atlanticstatesbeganin2005(HeeterandBird,2010,p.9).12
One unique state in terms of interconnectivity is Texas. The Texas Interconnection is
separatedfromtherestofthenation,makingTexastheonlymainlandstatewithitsowngrid.Also,
theTexasRECtradingprogramwasunusualinthatitrequirestheRECgeneratedelectricitytobe
produced in Texas (Hurlbut 2008).13 Nevada is the only other early adopter state that limits
renewablegenerationtowithinstateproducers,buttheydoallowlimitedoutofstateproduction.
II.3.RenewablePotential
The renewable energy potential for each state varies significantly. Texas is the only early
adopterstatewithsubstantialmodernrenewablepotential.14According to theNREL’s renewable
potentialdata,Texasranksfirstinonshorewindandsolarphotovoltaicpotential,fifthinbiopower
(solid) potential, and eighteenth in geothermal‐hydrothermal potential.15 Other states with
significantrenewablespotentialthathaveenactedRPSincludeWashington,California,Oregonand 12TheNEPOOL_GISRECtradingactivityincludedimportsof20,163GWhandexportsofapproximately10,861GWhin2008.Thisrepresentsapproximately6percentand3percentoftotalU.S.renewablegeneration(modernrenewablesandhydroelectricgeneration)in2008.13TheElectricReliabilityCouncilofTexas(ERCOT)whichmanagestheTexasInterconnectionmanageselectricpowerforapproximately85%ofthestate’stotalelectricload.Formoredetails,seeOfficeoftheGovernor(www.TexasWideOpenForBusiness.com),ERCOT(http://www.ercot.com/about,http://www.ercot.com/content/news/mediakit/maps/NERC_Interconnections_color.jpg),andDSIRE(http://www.dsireusa.org/incentives/incentive.cfm?Incentive_Code=TX03R).14http://www.nrel.gov/gis/re_potential.html15Noneoftheotherearlyadopterstateshasatop10ranking inanycategory,exceptNevadawhichrankssecondingeothermal‐hydrothermalpotential.NRELbiopowerestimates includecrop, forest,primary/secondarymill residues,and urban wood waste from Milbrandt (2005). See NREL 2012 for more information on the calculation of eachrenewableenergypotentialmeasure.
Page 9
8
New York, but they passed their RPS on or after 2003. In addition, in these four states,
hydroelectricityconstitutesthelargestshareofrenewablegenerationandmostofthehydroelectric
capacityexistedinthesestatesbeforetheirrespectiveRPSwereenacted.
II.4.HeterogeneityofRPSacrossStates
RPSarestate‐adoptedpoliciesandthereissignificantvariationinthecharacteristicsofRPS
acrossstates,whichisoneoftherationalesforourcasestudyapproach.OurSCMestimatesallow
ustodeterminetheeffectofastate’suniqueRPSpolicy in thecontextof itsdistinctpoliticaland
marketcharacteristics.
RPSvarynotonlyinthemagnitudeandtimingofthefinalrenewablesmandatebutalsothe
magnitudeandtimingofintermediatemandates(AppendixAdetailsthecurrenttargetsforallthe
RPS states). For instance, Wisconsin’s RPS (passed in 1999) requires 10 percent renewable
generationby2015whileMaine’sRPS(alsopassedin1999)requires40percentby2017,oneof
themoststringentinthenation.TheTexasRPSmandateissetintermsofcapacityandnotinterms
ofthepercentageofgenerationrequiring10,000MWby2025.Kneifel(2007)identifiesthisasan
importantfeaturevis‐à‐vistheeffectivenessofRPS.16
In addition to the finalmandate, states vary in their definitions of ‘renewable resources’.
Thisvariationisafunctionoftheiruniqueresources,politicalconditions,andeconomicstandingin
the regional economy. The mandated renewable sources can include wind, solar, geothermal,
biomass, some typesofhydroelectricity, andother resources suchas landfill gas,municipal solid
waste,andtidalenergy.ForsomestatesmodernrenewablesarelargelycategorizedasClass1and
makeupanincreasingportionoftherenewablerequirementsovertime.Forinstance,Connecticut
16TheonlyotherstatethatsetitsRPSbasedoncapacitywasIowa,buttheirmandatewassmall.Iowa’sRPSmandated105MWofrenewablecapacity.(http://twww.dsireusa.org/incentives/incentive.cfm?Incentive_Code=IA01R&re=1&ee=1)
Page 10
9
and New Jersey mandated three categories of renewables each with their own generation
requirements. The early adopter states included modern renewables in their set of allowable
renewables.
There isalsovariationinthecoverageofthepolicy indifferentstates. Insomestatesonly
specific typesofutilities, investorownedutilities (IOUs),municipal,orruralelectriccooperatives
(Coops)arerequiredtomeetRPS.Forexample,InWisconsintheinitialRPSmandateappliedonly
toIOUsandCoops.TheTexasRPSappliedtobothIOUsandretailsupplierswhilemunicipalutilities
and Coops could opt in. The legislative path of the passing of RPS also varied across states.
WisconsinwasthefirststatetoimplementRPSwithoutrestructuringitselectricitymarket,whilein
therestoftheearlyadopterstates,RPSpassedaspartoflegislationthatincludedrestructuringof
theelectricitymarket.
III.SyntheticControlMethod(SCM)forComparativeCaseStudy
ThereareanumberofadvantagestousingSCMinthisstudy.First, inprogramevaluation,
researchersoftenselectcomparisonsonthebasisofsubjectivemeasuresofsimilaritybetweenthe
affectedandtheunaffectedregionsorstates.But,neitherthesetofallnon‐RPSstatesnorasingle
non‐RPSstatelikelyapproximatesthemostrelevantcharacteristicsofatreatment(orRPS)state.
SCMprovidesacomparisonstate(orsynthetic)thatisacombinationofthecontrolstates,adata‐
drivenprocedurethatcalculates‘optimal’weightstobeassignedtoeachstateinthecontrolgroup
based on pre‐intervention characteristics, thus making explicit the relative contribution of each
controlunit to thecounterfactualof interest (AbadieandGardeazabal2003;Abadieetal.,2010).
Withreduceddiscretioninthechoiceofthecomparisoncontrolunits, theresearcher is forcedto
demonstratetheaffinitiesbetweentheaffectedandunaffectedunits.
Secondly, even when aggregate data are employed, as the case is in this paper, there is
Page 11
10
uncertaintyabouttheabilityofthecontrolgrouptoreproducethecounterfactualoutcomethatthe
affected statewouldhaveexhibited in theabsenceof the intervention.AsBuchmueller,DiNardo,
and Valleta (2011) explain, in a ‘clustering’ framework, inference is based on the asymptotic
assumption,i.e.,thenumberofstatesgrowslarge.Thecomparisonofasinglestateagainstallother
states in the control group collapses the degrees of freedom and results inmuch larger sample
variancecomparedtotheonetypicallyobtainedundertheconventionalasymptoticframeworkand
can seriously overstate the significance of the policy intervention (Donald and Lang 2007;
Buchmueller, DiNardo, and Valletta 2011; Bertrand et al. 2004). We, therefore, apply the
permutationsorrandomizationtestthatSCMreadilyprovides(Bertrand,Duflo,andMullainathan
2004; Buchmueller, DiNardo, and Valletta 2011; Abadie, Diamond, andHainmueller 2010; Bohn,
Lofstrom,andRaphael2014).
Thirdly, because the construction of the optimalweights does not require access to post‐
interventioninformation,SCMallowsustodecideonastudydesignwithoutknowingitsbearingon
thefindings(Abadie,Diamond,andHainmueller2010).Theabilitytomakedecisionsonresearch
designwhileremainingblindtohowaparticulardecisionaffectstheconclusionsofthestudyisa
safeguardagainstactionsmotivatedbya‘desired’finding(Rubin2001).
Finally, Abadie, Diamond, and Hainmueller (2010) argue that unlike the traditional
regression‐based difference‐in‐difference model that restricts the effects of the unobservable
confounders to be time‐invariant so that they can be eliminated by taking time differences, SCM
allows such unobservables to vary with time. In particular, Abadie, Diamond, and Hainmueller
Page 12
11
(2010) show that with a long pre‐intervention matching on outcomes and characteristics a
syntheticcontrolalsomatchesontime‐varyingunobservables.17
III.1.TheSyntheticControl
AtypicalSCManalysisisfeasiblewhenoneormorestatesexposedtoaninterventioncanbe
compared to other states that were not exposed to the same intervention. In this paper, the
interventionisRPS,theoutcomeisrenewablecapacity,andthesetofexposedstatesaretheearly
RPSadopterstates.Thedonorpool(unexposed/controlstates)consistsofstatesthatdidnothave
thepolicyfortheobservedperiod.
To obtain the synthetic control we follow Abadie and Gardeazabal (2003) and Abadie,
Diamond, and Hainmueller (2010). For states 1,...,1 Ji and periods Tt ,...,1 , suppose state
1i isexposedtotheinterventionat ),1(0 TT .Theobservedoutcomeforanystate i attimetis,
(1) ititN
itit SYY ,
where NitY is the outcome for state i at time t in the absence of the intervention, the binary
indicatorvariable itS denotestheinterventiontakingthevalue1if 1i and 0Tt ,and it isthe
effectoftheinterventionforstate i attimet.
Wewant to estimate ),...,( 111 0 TT . Abadie, Diamond, andHainmueller (2010) show that,
under standard conditions, there exist ),...,( 12
JwwW such that pre‐intervention matching is
achievedwithrespecttotheoutcomevariableaswellascharacteristics(orpredictors),andwecan
use,
17 As Abadie et al. (2014) explains the intuition as, “… only units that are alike in both observed and unobserveddeterminantsof theoutcomevariable aswell as in the effect of thosedeterminantson theoutcomevariable shouldproducesimilartrajectoriesoftheoutcomevariableoverextendedperiodsoftime.”
Page 13
12
(2) },...,1{,ˆ 01
211 TTtYwY jt
J
j jtt
,
asanestimator for t1 .Theterm jt
J
j jYw
1
2 ontheright‐hand‐sideof(2) issimplytheweighted
average of the observed outcome of the control states for },...,1{ 0 TTt withweights W . The
proceduretoobtain W isinAppendixB.
III.2.Inference
Once an optimal weighting vector W is obtained, the “synthetic” is constructed by
calculating theweighted averageoutcomeof thedonorpool. Thepost‐intervention valuesof the
synthetic control serve as our counterfactual outcome for the treatment state. The post‐
intervention gapbetween the actual outcome and the synthetic outcome, therefore, captures the
impactoftheintervention.
To begin, we calculate a difference‐in‐difference estimate for the treatment state (Bohn,
Lofstrom,andRaphael2014,MunasibandRickman2015),
(4) presyntheticTR
preactualTR
postsyntheticTR
postactualTRTR YYYY ,,,, ,
where postactualTRY , is the average of the post‐intervention actual outcome of the treatment state,
postsyntheticTRY , istheaverageofthepost‐interventionoutcomeofthecounterfactual.Similarly, pre
actualTRY , is
theaverageofthepre‐interventionactualoutcomeoftreatmentstate,and presyntheticTRY , istheaverage
of thepre‐interventionoutcomeof thecounterfactual. If theoutcomechanged in response to the
interventionintime 0T wewouldexpect 0TR.
To formally test the significance of this estimate, we apply the permutations or
randomizationtest,assuggestedbyBertrandetal.(2004),Buchmuelleretal.(2011),Abadieetal.
(2010)andBohnetal.(2014),onthisdifference‐in‐differenceestimator.Specifically,foreachstate
Page 14
13
inthedonorpool,weestimatethedifference‐in‐differenceasspecifiedinequation(4)asifitwas
exposedtoRPSattime 0T (i.e.,applyafictitiousintervention).Thedistributionofthese“placebo”
difference‐in‐differenceestimatesthenprovidestheequivalentofasamplingdistributionforTR .
Tobe specific, if the cumulativedensity functionof the complete setof estimates is givenby
)(F ,thep‐valuefromaone‐tailedtestofthehypothesisthat 0TRisgivenby )( TRF (Bohnetal.
2014). Note that this answers the question, how often would we obtain an effect of RPS of a
magnitudeaslargeasthatofthetreatmentstateifwehadchosenastateatrandom,whichisthe
fundamental question of inference (Bertrand et al., 2004, Buchmueller et al. 2011, Abadie et al.
2010).
WecarryoutasecondtestwherewecalculatewhatwecalltheDIDrank.Itistherankingof
theabsolutevalueofthemagnitudeofthedifference‐in‐differenceofthetreatmentstateagainstall
the placebo difference‐in‐differencemagnitudes (Bohn et al. 2014,Munasib and Rickman 2015).
Forexample,ifDIDrankis1thentheestimatedimpactoftheinterventioninthetreatmentstateis
greaterthananyoftheestimatedplaceboimpacts.
IV.Data
We collected the data for the outcome variable, renewable capacity, from the EIA. The
information on state RPS is collected from the Database of State Incentives for Renewables &
Efficiency(DSIRE)database(seeAppendixA).Figure2demonstratesthatstatesthathaveadopted
RPS are largely the states that have renewable generation capacity additions. This, of course, is
confoundedbyvariousaggregatefactorssuchas theFederalProductionTaxCredit(PTC).Oneof
the rationales behind our case study approach is thatwe can purge out these aggregate effects,
factorssuchasthePTCapplytobothcontrolandtreatmentstates.
Page 15
14
Much of the remaining energy data, including electricity generation and price, generating
capacity, number of customers, etc., were also collected from the EIA. We used information on
geographicalfeaturessuchassunlightandnaturalamenitiesfromtheEconomicResearchService
(ERS)of theU.S.DepartmentofAgriculture(USDA)andtemperatures fromNationalOceanicand
AtmosphericAdministration(NOAA).Populationaswellaseconomicindicatorssuchaspercapita
personal income andmanufacturing earnings sharewere obtained from theBureauofEconomic
Analysis(BEA).PovertyratesarefromtheCensus.
In addition, we collected data on technical renewable potentials from Pacific Northwest
NationalLaboratory(PNNL)andNREL.Therearetwowindpotentialmeasures.Thefirstmeasure
is derived from wind potential estimates produced by the PNNL in 1991 (Elliott, Wendell, and
Glower1991,p.B‐1).Windpotentialcalculationsindicatetheamountofwindthatastateorregion
istheoreticallycapableofproducingunderaspecificsetoftechnologicalandlanduseassumptions,
excluding transmission limitations.18 The second measure is an updated 2010 wind potential
measurements constructed by NREL.19 Similarly, the photovoltaic potential, biopower (solids)
potential, and geothermal‐hydrothermal potential measures are also 2010 estimates from NREL
(NREL2012).Table1presentsasummarydescriptionforallthevariablesusedintheanalysis.
18Forinstance,theinstalledcapacitycalculationsarebasedonanassumptionof5MW/km2ofinstalledcapacity.19 The twomeasures differ based on technological and land use assumptions. For instance, the 1991measurewasconstructedwithanassumedturbineheightof 50mduetotheavailabilityofwindtechnologyatthetime,whilethe2010measurewasconstructedusingan80mturbineheight(NREL2010).
Page 16
15
V.Results
V.1.SCMEstimatesoftheImpactofRPSonRenewableCapacity
We construct the counterfactual (or synthetic) renewable capacity for each of our early
adopterstates(asdiscussedinSectionIII).Ourdonorpoolconsistsof26statesthatdidnotpassa
lawsimilartomandatoryRPSasof2008.
Figure 3 is a graphical representation of the SCM estimates of the impact of RPS on
renewablecapacityforthesixexposedstates.Ineachpanel,thepictureontheleftshowstheactual
andthesyntheticrenewablecapacitiesfortheperiod1990‐2008.Thepictureontherightpresents
thepermutations/randomizationortheplacebotests:thedarklineisthegapbetweenactualand
syntheticforthetreatmentstate,whereaseachgreylineisthegapbetweenactualandsyntheticof
aplacebo.ThedetailsoftheestimationarereportedinTable2.
TheleftpictureinpanelA(Nevada)showsthatthesyntheticrenewablecapacitycoincides
wellwith theactualrenewablecapacityover1990‐1996.OntherightpictureofpanelA,we find
thatNevada(thedarkline)doesnotstandoutfromtheplacebos(thegreylines).Asexplainedin
section III.2,weexamine thecomparisonof thepost‐predifferenceratios fromtheplacebo tests.
Along the first column of Table 2, we find that the DID rank is 25 and the p‐value of the DID
measuredoesnothaveasignificantp‐value.We,thus,concludethatRPSdidnothaveasignificant
impactonrenewablecapacityinNevada.
WeobservethesamepatternforConnecticut(panelBofFigure3andcolumn2ofTable2),
Maine(panelCofFigure3andcolumn3ofTable2),NewJersey(panelDofFigure3andcolumn4
ofTable2)andWisconsin(panelFofFigure3andcolumn6ofTable2).Ineachofthesecaseswe
findthattheDIDrankishighandnotstatisticallysignificant.
Page 17
16
For Texas, however, we find that RPS had a significant impact on renewable capacity
addition.OntheleftpictureofpanelEofFigure3,weseethattheactualcapacitystartstodeviate
fromthesyntheticinthepost‐interventionperiod(i.e.,1999,theyearofRPS)andkeepsdiverging.
OntherightpictureofPanelE,weseethatthegapbetweenactualandsyntheticforTexasstands
outinthemidstofalltheplacebogaps.Incolumn5ofTable2,wefindthatTexas’sDIDrankis1,
anditissignificantat1percent.ThemainconstituentsofTexas’ssyntheticasindicatedbythew‐
weights are (in order of importance): Indiana, Illinois and Virginia. The strongest predictors of
renewablecapacityforTexas’ssyntheticare(notshown):coalandnaturalgasgenerationshares,
percapitaincome,growthofcustomersandpercapitaincome,andshareofmanufacturingincome.
V.2.AlternativeSetofPredictors
Totest ifourestimatesarerobusttochangesinthesetofpredictors(forpre‐intervention
matching)we carry out robustness checkswith an alternative set of predictors.We include the
1991 wind potential measure and geographic and weather variables: January sunlight, summer
coolingdegreedays,summerheatingdegreedays.Alaskaisdroppedfromthedonorpoolbecause
the1991windpotentialmeasureandthegeographicvariablesarenotavailableforthisstate.Table
3presentstheseresults.BasedontheDIDranksaswellas thep‐valuesof theDIDmeasures,we
concludethatonlyincaseofTexas,RPShadasignificantimpactonrenewablecapacity.
V.3.AdditionalRobustnessChecksforTexas
IneachSCMreportedinTables2and3,foreachtreatmentstate,thestate’spre‐intervention
outcome(renewablecapacity)isincludedwithacommonsetofpredictors.Then,thematchingis
done to calculate the optimalw‐weight. In the case of Texas, therefore,matching is done on the
commonsetofpredictorsaswellastheoutcomevariable(renewablecapacity)fortheperiod1990‐
Page 18
17
1998.20 However, Texas’s renewable electricity market did not exist prior to 1998. So, we have
conductedarobustnesscheck,reportedincolumn1ofTable4,wherethematchingisdoneonthe
setofpredictorsthatincludesrenewablecapacityfor1998only.Wefindthatourinferenceremains
unchanged.ThemainconstituentsofTexas’ssyntheticasindicatedbythew‐weightsare(inorder
ofimportance):Oklahoma,Illinois,andSouthDakota.
AnotherissueisthattheTexasRPSincludessomedegreeofrestructuringintheelectricity
market.Todetermineiftheeffectofrestructuringisconfoundingthefindings,incolumn2ofTable
4wepresenttheSCMresultswherewehaveexcludedstatesthathadanykindofderegulation(i.e.,
thedonorpool has onlynon‐RPS andnon‐deregulated states). The set of predictors remains the
sameasthatinTable2.Again,wearriveatthesameconclusionthatRPShadasignificantimpacton
renewablecapacity.
V.4.Discussion:HeterogeneityofRPSImpacts
WefindthatofthesixearlyRPSadopters,TexasistheonlystatewhereRPShadanimpact
on modern renewable capacity. It is important to point out that Texas stands out among these
statesinanumberdifferentways.First,TexasisanexceptioninspecifyingRPSintermsofcapacity.
Allotherstates,withtheexceptionofIowa,specifyRPSasapercentageoftotalgeneration.Kneifel
(2007)arguesthatthetypeofmandateinfluencesitseffectiveness.
Second, the five early adopter states where we do not find an effect are also among the
smallestenergyproducingstates;NewJersey,whichisthelargestproducerofthesefivestates,had
only a 0.5 percent share of the total U.S. generation in 2012. Texas, on the other hand,was the
20ThisisthestandardprocedurefollowedinSCMduetoAbadieetal.(2010)andBohnetal.(2014).
Page 19
18
largest energy producing state for every year between 1990 and 2012.21 The size of the Texas
electricitymarketmayhavegivenTexasanedgeinaddingrenewablecapacity.
Third,inadditiontosize,gridinterconnectivityhasimportantimplicationsfortheexpansion
of renewable capacity. In New England, the ISO New England RTO coordinates the trading of
renewable generated electricity across states using REC. This may have influenced the pace of
withinstaterenewablecapacityadditionsinConnecticutandMaine,bothintheISONewEngland
region.22NewJersey,whichisinthePJMRTO,promoteswithinstatedevelopment,particularlyfor
solar generation. However, if approved by the New Jersey Board of Public Utilities, renewable
generation can also be generated from regional capacity (Daniel et al, 2014, p. 7). InWisconsin
tradablecreditsarecreatedonlywhenanelectricutilityorcooperativeprovides totalrenewable
energy to its retail electric customers in excess of the RPS requirements (See Berry 2002 for
details).Texas,ontheotherhand,istheonlymainlandstatewithitsowngridandunlikeotherREC
programs,theERCOTRECprogramonlyoperatesinTexas;togenerateaunitofRECtheelectricity
hastobegenerated(fromrenewables)andmeteredinTexas.
InNevada,utilitiesarerequiredtomeetaminimumof5percentoftherequiredrenewables
mandate through solar generated electricity. Nevada did not meet 100 percent of their RPS
obligationuntil2008.23 InNew Jersey, in2005, themandatewasrevisedwhereby theshare that
must come fromClass1 renewableswas set tobe17percentby2021.Until2005,however, the
21http://www.eia.gov/electricity/data/state/22Asarobustnesscheck,weconductedanSCManalysiswheretheNewEnglandregionisconsideredthetreatedunit.The year of intervention was the first year in which a state in New England passed RPS, 1999. The finding wasconsistentwiththestatelevelresults.TherewasnotasignificantinfluenceofRPSonrenewablecapacity.Theseresultsareavailableuponrequest.23In2009,beyondouranalysisperiod,thestringencyoftheinitialpolicywasincreasedandthefinalmandatewasincreasedto25percentby2025.Seehttp://www.dsireusa.org/incentives/incentive.cfm?Incentive_Code=NV01R.
Page 20
19
mandatewasthatthesharethatmustcomefromClass1renewableswas0.74percent.24Thismay
explainwhytherewasnocapacityexpansionthrough2008.
InMaine,atthetimeofthepassageofRPS,thegenerationconstraintwasnotbinding.Maine
hassignificanthydroelectricgenerationcapacity,andgenerationfromtheseresourcesexceededthe
initial mandate. The Maine RPS was subsequently updated to require that a portion of the
renewablecapacitybeinstalledafter2005.Themandatewassmallhowever,requiring1percentof
electricitybeproducedfromnewrenewablecapacityin2008.
InWisconsin, the initialRPSmandateappliedonly to InvestorOwnedUtilities (IOUs) and
RuralElectricCooperatives(Coops),requiringthemtoobtain2.2percentof theirelectricity from
renewablesourcesby2012.Thepolicywasstrengthenedin2006,withautility‐widerequirement
of10percentby2015.25
V.5.Discussion:EfficacyofTexasRPS
WeobservethatTexasproducersreached10,000MWofwindgenerationcapacityby2010
reachingtheRPStargetyearsaheadofthemandatedtimeline.This,however,doesnotindicatethat
RPS was not binding. In the presence of non‐convex adjustment costs, indivisibilities, and
irreversibilities ofwind generation capital, optimal investment is unlikely to be incremental and
more likely toexhibitburstsof large‐scalecapitalaccumulations(AddaandCooper2003,Cooper
and Haltiwanger 2006). As a result, the level and timing of optimal investment may very well
exceedandprecedethemandate,aswasthecaseinTexas.
Additionally, firms may have predated wind generation capacity in order to secure the
federalProductionTaxCredit(PTC)benefits.ThePTCappliestowindfarmsforthefirst10yearsof
24http://www.dsireusa.org/summarytables/rrpre.cfm25http://www.dsireusa.org/incentives/incentive.cfm?Incentive_Code=WI05R&re=0&ee=0.
Page 21
20
productionandlowersthecostofwindgeneratedelectricityproductionbyaboutonethird(Wiser,
2007).26ThecreditwasoriginallycreatedundertheEnergyPolicyActof1992,butithasexpired
andbeenextendedseveraltimessinceitsinception(Wiser,Bolinger,andBarbose2007,p.1‐2).27
Its lapses over the years are correlatedwith decreases inwind capacity additions and are often
blamedforthosedeclines(AWEA2005,p.4).Barradale(2010)findsthatuncertaintyinthefederal
PTC leads to investmentvolatility, asproducersdelayproduction innon‐PTCyearsand rampup
productionwhenthePTCisactive.
V.6.Discussion:EarlyAdopterStates
BecauseRPSmandatesdonotbecomeimmediatelybindingbutareimplementedthrougha
seriesofintermediategoalsleadinguptothefinalmandate,weonlyfocusedonearlyadopterstates
(i.e.,statesthatenactedRPSbetween1997and2000).Thisallowedussufficientpost‐intervention
years tocapture theeffectofRPS. Indeed,with twoexceptions,MassachusettsandCalifornia, for
statesthatpassedtheirRPSbetween2000and2008,theearliestintermediatemandateis2006.28
Whiletheavailablepost‐interventionperiodsmaynotbesufficientforcarryingoutSCMimpactsof
RPS inMassachusetts andCalifornia,wehave still carried out the estimates.Wedonot find any
impactofRPSonrenewablecapacityinMassachusetts.AsforCalifornia,weareunabletoestablish
apre‐interventionmatching.ThisisbecauseCaliforniahadbyfarthelargestrenewablecapacityfor
thepre‐interventionperiod(1990‐2002);Californiahadatleast8timestherenewablecapacityof
26ThePTCiscurrentlyworth$22perMWh(2011dollars).In2013,Texasaccountedfor22%ofthe167millionMWhoftotalpowergeneratedfromwindnationwide.Seehttp://www.eia.gov/todayinenergy/detail.cfm?id=8870(EIA‐PTC)andhttp://www.eia.gov/todayinenergy/detail.cfm?id=15851(EIA‐Texas).27ThePTCexpiredandwasextendedin2000,2002,2004,and2012.Itwasextendedin2010priortoexpiration.28Massachusetts,whichpasseditsRPSin2002,requiredrenewablegenerationof1percentofsalesin2003,increasingby0.5percentannuallythrough2009.California,whichpassed itsRPS in2003, includedarequirement thatutilitiesincreaserenewablegenerationannuallybyaminimumof1percentoftheirsales.
Page 22
21
anyotherstateforthisperiod.Asaresult,noweightedaverageofstatescanapproximatethepre‐
interventionrenewablecapacityofCalifornia.29
VI.Conclusion
Variation across states in their policy environment, electricity market structure, and
availabilityofrenewableenergyresourcessuggestthatempiricalidentificationoftheeffectofRPS
reliescruciallyontheaccuratedeterminationofthecontrolstates.WeemploytheSCMcasestudy
approachwhich,weargue,usesamoreappropriatecounterfactualforimpactevaluationcompared
to the approaches estimating average treatment effects. We find that RPS have heterogeneous
impactsonrenewablecapacitydevelopment.
Therenewablepolicyenvironmentacrossstatesisatacrossroads.Thisisparticularlytrue
forRPS in lightof therecent legaland legislativeefforts torepealorweakenRPS inanumberof
states including California, Colorado, Kansas,Massachusetts,Minnesota, andOhio (Plumer 2013;
Gallucci 2013). InMay 2014, Ohio legislators voted to halt the continued implementation of the
state’sRPS,whichwaspassedin2009(Cardwell2014).Similarbillshavealsobeenintroducedin
Wisconsin,WestVirginia,Minnesota andTexas.WhileRPS survived repeal bills early in 2014 in
KansasandNorthCarolina,theyareexpectedtobepickedupagainlaterintheyear.
On the backdrop of the previous findings that RPS are not contributing to renewables
development (Delmas and Montes‐Sancho 2011, Shrimali and Kniefel 2011, Hitaj 2013, and
Maguire2014),theserepealeffortsmaypickupsteam.Butthefindingsinthispapersuggestthat
the impact of RPS may not be generalized; instead, the success of a particular RPS may be
29Incontrast,considerTexas,forinstance.Texas’saveragenon‐renewablecapacityduringthepre‐interventionperiod(1990‐1998)fellbetweenthemedianandthe75thpercentileamongtheU.S.states.Asaresult,thefeasibilityoffindingaweightedaverageofcontrolstatesthatwouldmimicTexas’spre‐interventionnon‐renewablecapacitywasnotanissue.
Page 23
22
contingent on the features of the policy itself and the characteristics of the pertinent electricity
markets.
Page 24
23
ReferencesAbadie,Alberto,AlexisDiamondandJensHainmueller(2014).“ComparativePoliticsand
the Synthetic Control Method,” First published online: 23 APR 2014, in AmericanJournalofPoliticalScience.
Abadie, Alberto, Alexis Diamond, and Jens Hainmueller (2010). "Synthetic ControlMethods forComparativeCaseStudies:EstimatingtheEffectofCalifornia’sTobaccoControlProgram."JournaloftheAmericanStatisticalAssociation,vol.105:493‐505.
Abadie,Alberto,andJavierGardeazabal(2003)."TheEconomicCostsofConflict:ACase‐Control Study for theBasqueCountry."AmericanEconomicReview, vol. 93 (1):113‐132.
Adda, Jérôme,andRussellW.Cooper (2003).DynamicEconomics:QuantitativeMethodsandApplications:MITPress.
AWEA (2005). Economics of Wind Energy. Washington, DC: American Wind EnergyAssociation.
Barradale, M. J. (2010). Impact of public policy uncertainty on renewable energyinvestment: Wind power and the production tax credit.Energy Policy, vol. 38(12),7698‐7709.
Berry, David (2002). "The market for tradable renewable energy credits."EcologicalEconomics,vol.42(3):369‐379.
Bertrand,Marianne,EstherDuflo,andSendhilMullainathan(2004)."Howmuchshouldwe trust differences‐in‐differences estimates?" The Quarterly Journal of Economics,vol.119(1):249‐275.
Bohn,Sarah,MagnusLofstrom,andStevenRaphael(2014)."Didthe2007LegalArizonaWorkers Act Reduce the State’s Unauthorized Immigrant Population?" Review ofEconomicsandStatistics,vol.96(2),258‐269.
Buchmueller,ThomasC., JohnDiNardo,andRobertG.Valletta(2011)."TheEffectofanEmployerHealthInsuranceMandateonHealthInsuranceCoverageandtheDemandforLabor:EvidencefromHawaii."AmericanEconomicJournal:EconomicPolicy,vol.3(4):25‐51.
Cardwell, Diane (2014). "A Pushback on Green Power."TheNew York Times, May 28,2014.
Carley,S.(2009)."Staterenewableenergyelectricitypolicies:Anempiricalevaluationofeffectiveness."EnergyPolicy,vol.37(8):3071‐3081.
Cooper,RWandJCHaltiwanger(2006).“Onthenatureofcapitaladjustmentcosts,”TheReviewofEconomicStudies,vol.73(3),611‐633.
Daniel, Kate, Heather Calderwood, Ethan Case, Ben Inskeep, Autumn Proudlove, andAchyutShrestha(NCCleanEnergyTechnologyCenter).TechnicalAssistanceforU.S.DepartmentofEnergy.November2014.
Delmas,MagaliAandMariaJ. Montes‐Sancho(2011)."U.S.statepoliciesforrenewableenergy:ContextandEffectiveness."EnergyPolicy,vol.39:2273‐2288.
Page 25
24
Donald,StephenGandKevinLang(2007)."Inferencewithdifference‐in‐differencesandotherpaneldata."ThereviewofEconomicsandStatistics,vol.89(2):221‐233.
Doot, David T, Paul N Belval, and Lynn M Fountain (2007). "State Mandates MostEffectiveSoFarinRenewablePortfolioStandards."NaturalGas&Electricity.
Elliott,D.L.,L.L.Wendell,andG.L.Glower(1991).AnAssessmentoftheAvailableWindyLandAreaandWindEnergyPotentialintheContiguousUnitedStates.Richland,WA:PacificNorthwestLaboratory.
Fischer,Carolyn,andRichardG.Newell(2008)."EnvironmentalandTechnologyPoliciesfor Climate Mitigation." Journal of Environmental Economics andManagement, vol.55:142‐162.
Fischer,Carolyn(2010)."RenewablePortfolioStandards:WhenDoTheyLowerEnergyPrices?"EnergyJournal,vol.31(1):101‐119.
Gallucci, Maria (2013). Renewable Energy Standards Target of Multi‐Pronged Attack.InsideClimateNews,March19,2013.
Heeter,Jenny,andLoriBird.(2011)"StatusandTrendsinUSComplianceandVoluntaryRenewableEnergyCertificateMarkets(2010Data)."Contract,vol.303:275‐3000.
Hitaj, Claudia (2013). "Wind Power Development in the United States." Journal ofEnvironmentalEconomicsandManagement,vol.65:394‐410.
Hurlbut,David(2008)."ALookBehindtheTexasRenewablePortfolioStandard:ACaseStudy."NaturalResourcesJournal,vol.48:129‐161.
Keele,Luke,NeilMalhotra,andColinH.McCubbins(2013).“DoTermLimitsRestrainStateFiscalPolicy?ApproachesforCausalInferenceinAssessingtheEffectsofLegislativeInstitutions,”LegislativeStudiesQuarterly,vol.38:291–326.
Kneifel, Joshua (2007).Effectsof StateGovernmentPoliciesonElectricityCapacity fromNon‐HydropowerRenewableSources,DepartmentofEconomics,UniversityofFlorida.
Maguire, Karen. 2014. "What’s PoweringWind? The Effect of State Renewable EnergyPoliciesonWindCapacityintheUnitedStates(1994‐2012)."WorkingPaper.
Milbrandt,A.,"AGeographicPerspectiveontheCurrentBiomassResourceAvailabilityinthe United States". NREL/TP‐560‐39181, December 2005. National RenewableEnergyLaboratory,GoldenCO
Munasib, A. andD. Rickman (2015). “Regional Economic Impacts of the ShaleGas andTightOilBoom:ASyntheticControlAnalysis,”RegionalScienceandUrbanEconomics,vol.50,Jan2015:1–17.
NREL(2010).NewWindEnergyResourcePotentialEstimatesfortheUnitedStates.AWSTruwind,NationalRenewableEnergyLaboratory.
NREL(2012).Lopez,Anthony,BillyRoberts,DonnaHeimiller,NateBlair,andGianPorro.“U.S. Renewable Energy Technical Potentials: A GIS‐Based Analysis.” TechnicalReport.NREL/TP‐6A20‐51946.Golden,CO:NationalRenewableEnergyLaboratory.
Plumer, Brad (2013). "State renewable‐energy laws turn out to be incredibly hard torepeal."TheWashingtonPost,August8,2013.
Rubin,AlanM.2001."TheChallengeofWritingtheQuantitativeStudy."InHowtoPublishYourCommunicationResearch:AnInsider'sGuide,57.
Page 26
25
Shrimali, Gireesh, and Joshua Kniefel (2011). "Are Government Policies Effective inPromoting Deployment of Renewable Electricity Resources?"Energy Policy, vol. 39(4726‐4741).
Wiser, Ryan, Mark Bolinger, and Galen Barbose (2007). Using the Federal Production Tax Credit to Build a Durable Market for Wind Power in the United States. Berkeley, CA: Lawrence Berkeley National Laboratory.
Wiser,Ryan(2007).“WindPowerandtheProductionTaxCredit:AnOverviewofResearchResults.”Berkeley,CA:LawrenceBerkeleyNationalLaboratory.
Yi,H.,&Feiock,R.C.(2012).Policytoolinteractionsandtheadoptionofstaterenewableportfoliostandards.ReviewofPolicyResearch,vol.29(2),193‐206.
Yin,Haitao,andNicholasPowers(2010)."DoStateRenewablePortfolioStandardsPromoteIn‐stateRenewableGeneration?"EnergyPolicy,vol.38:1140‐1149.
Page 27
26
Figures
Figure1:FERCElectricPowerMarkets:NationalOverview
Source:http://www.ferc.gov/market‐oversight/mkt‐electric/overview.asp
Figure2:U.S.RPSandRenewableGenerationCapacity
Page 28
27
Figure3:SCMEstimatesoftheImpactofRPSonRenewablesCapacity
PanelA:Nevada
1990 1992 1994 1996 1998 2000 2002 2004 2006 20080.1
63.09
126.08
189.07
252.06
315.05
378.04
441.03
504.02
567.01
630
year
nam
epl
ate
ca
paci
ty
Nevada Renewable Nameplate CapacityNevada vs Control States Synthetic Control
ActualSynthetic
1 3 5 7 9 11 13 15 17 19
-629.9
-503.92
-377.94
-251.96
-125.98
0
125.98
251.96
377.94
503.92
629.9
year
nam
epl
ate
ca
paci
ty
Nevada Nameplate Capacity GapAnd Placebo Gaps in All Control States
PanelB:Connecticut
1990 1992 1994 1996 1998 2000 2002 2004 2006 20080.1
39.45
78.8
118.15
157.5
196.85
236.2
275.55
314.9
354.25
393.6
year
nam
epl
ate
ca
paci
ty
Connecticut Renewable Nameplate CapacityConnecticut vs Control States Synthetic Control
ActualSynthetic
1 3 5 7 9 11 13 15 17 19
-393.5
-314.8
-236.1
-157.4
-78.7
0
78.7
157.4
236.1
314.8
393.5
year
nam
epl
ate
ca
paci
ty
Connecticut Nameplate Capacity GapAnd Placebo Gaps in All Control States
PanelC:Maine
1990 1992 1994 1996 1998 2000 2002 2004 2006 20080.1
22.89
45.68
68.47
91.26
114.05
136.84
159.63
182.42
205.21
228
year
nam
epl
ate
ca
paci
ty
Maine Renewable Nameplate CapacityMaine vs Control States Synthetic Control
ActualSynthetic
1 3 5 7 9 11 13 15 17 19
-227.9
-182.32
-136.74
-91.16
-45.58
0
45.58
91.16
136.74
182.32
227.9
year
nam
epl
ate
ca
paci
ty
Maine Nameplate Capacity GapAnd Placebo Gaps in All Control States
Page 29
28
PanelD:NewJersey
1990 1992 1994 1996 1998 2000 2002 2004 2006 20080.1
63.59
127.08
190.57
254.06
317.55
381.04
444.53
508.02
571.51
635
year
nam
epl
ate
ca
paci
ty
New Jersey Renewable Nameplate CapacityNew Jersey vs Control States Synthetic Control
ActualSynthetic
1 3 5 7 9 11 13 15 17 19
-634.9
-507.92
-380.94
-253.96
-126.98
0
126.98
253.96
380.94
507.92
634.9
year
nam
epl
ate
ca
paci
ty
New Jersey Nameplate Capacity GapAnd Placebo Gaps in All Control States
PanelE:Texas
1990 1992 1994 1996 1998 2000 2002 2004 2006 2008-500
454.32
1408.64
2362.96
3317.28
4271.6
5225.92
6180.24
7134.56
8088.88
9043.2
year
nam
epl
ate
ca
paci
ty
Texas Renewable Nameplate CapacityTexas vs Control States Synthetic Control
ActualSynthetic
1 3 5 7 9 11 13 15 17 19
-9543.2
-7634.56
-5725.92
-3817.28
-1908.64
0
1908.64
3817.28
5725.92
7634.56
9543.2
year
nam
epl
ate
ca
paci
ty
Texas Nameplate Capacity GapAnd Placebo Gaps in All Control States
PanelF:Wisconsin
1990 1992 1994 1996 1998 2000 2002 2004 2006 20080.1
67.14
134.18
201.22
268.26
335.3
402.34
469.38
536.42
603.46
670.5
year
nam
epl
ate
ca
paci
ty
Wisconsin Renewable Nameplate CapacityWisconsin vs Control States Synthetic Control
ActualSynthetic
1 3 5 7 9 11 13 15 17 19
-670.4
-536.32
-402.24
-268.16
-134.08
0
134.08
268.16
402.24
536.32
670.4
year
nam
epl
ate
ca
paci
ty
Wisconsin Nameplate Capacity GapAnd Placebo Gaps in All Control States
Notes:(a)Outcomevariableisrenewablescapacityofgeothermal,biofuels,solar,andwind.(b)ThesearethepicturesoftheestimatesthatarefurtherdescribedinTable2.
Page 30
29
TablesTable1:SummaryStatistics(1990‐2008) Donorpool(26states) TreatmentStateMeans
Mean SD Min Max NV CT ME NJ TX WI
Renewablenameplatecapacity(MW) 79.90 166.81 1.00 1130.00 247.32 235.08 78.17 198.31 1162.30 97.23
Totalnameplatecapacitygrowth 37.99 28.49 0.35 117.41 7.15 164.52 13.07 249.06 34.46 25.62
Coalgenerationshare 0.58 0.29 0.00 0.99 0.53 0.13 0.03 0.16 0.39 0.70
Naturalgasgenerationshare 0.10 0.16 0.00 0.65 0.35 0.15 0.21 0.31 0.48 0.04
Realelectricityprice 7.29 1.74 4.48 13.57 7.99 12.50 11.43 11.79 8.11 7.01
Growthoftotalcustomer 1.15 0.13 0.90 1.65 1.56 1.06 1.11 1.08 1.22 1.15
RealPCpersonalincome($) 28709.57 4698.39 18152.14 45222.82 33078.03 43696.99 28482.82 40186.85 29529.13 30577.93GrowthofPCpersonalincome 1.22 0.16 0.98 1.82 1.19 1.21 1.20 1.19 1.24 1.23Percentofpopulationbelowpoverty 13.17 3.57 5.70 26.40 10.32 8.75 11.67 8.71 16.51 9.67
Shareofmfg.earnings 0.11 0.05 0.02 0.22 0.03 0.12 0.11 0.09 0.10 0.18
Windpotential(1991) 239.08 405.16 0.00 1210.00 50.00 5.00 56.00 10.00 1190.00 56.00Windpotential(2010) 227.74 325.94 0.00 952.37 7.25 0.03 11.25 0.13 1901.53 103.76Photovoltaicpotential(2010) 3168.83 2024.67 36.55 9005.30 3742.84 17.13 660.61 276.43 20565.29 3240.76
Biopower‐solidpotential(2010) 1.17 0.77 0.06 3.52 0.04 0.06 0.54 0.15 2.04 1.42
Geo‐&hydro‐thermalpotential(2010) 0.23 0.62 0.00 2.18 5.75 0.00 0.00 0.00 0.00 0.00
Januarymeanhoursofsunlight 147.13 25.34 105.14 197.64 200.00 161.75 156.25 152.24 182.59 133.54
Averagesummercoolingdegreedays 289.38 136.47 23.67 584.33 486.75 168.11 71.42 227.60 531.56 145.00
Averagesummerheatingdegreedays 24.49 36.12 0.00 212.00 13.21 13.14 62.96 5.46 0.00 47.23
Notes:(a)ForthedonorpoolsN=494,exceptfor1991windpotential,sunlightanddegreedaysthatareunavailableforAlaska(i.e.,N=475).(b)Renewablesincludegeothermal,biofuels,solar,andwind.(c)TotalnameplatecapacityismeasuresinMWhper100squaremiles.(d)Allmonetaryvaluesarein2005constantdollars.(e)Standardstatecodesused:Nevada(NV),Connecticut(CT),Maine(ME),NewJersey(NJ),Texas(TX),Wisconsin(WI),Massachusetts(MA).(f)YearoftheenactingRPS in the treatment states: Nevada (1997), Connecticut (1998), Maine (1999), New Jersey (1999), Texas (1999), andWisconsin (1999). (g) Following statesenactedRPSonorbefore2008andthereforeexcludedfromthedonorpool:Iowa(1983),Massachusetts(2002),California(2003),Colorado(2004),Hawaii(2004),Maryland (2004),NewYork (2004),Rhode Island (2004),Delaware (2005),DistrictofColumbia (2005),Montana (2005),Oregon (2005),Pennsylvania (2005),Washington(2006),Arizona(2007),Minnesota(2007),NewHampshire(2007),NewMexico(2007),Michigan(2008),Missouri(2008),NorthCarolina(2008).(h)1991windpotentialisinenergyunits(annual'000GWh),therestofthepotentialmeasuresaremeasuredaspower(GW).
Page 31
30
Table2:SCMEstimateoftheImpactofRPSonRenewablesCapacity Nevada Connecticut Maine NewJersey Texas WisconsinEstimationsummary
Pre‐interventiondifference(D1) 0.34 0.69 0.03 0.13 ‐0.54 ‐0.54Post‐interventiondifference(D2) ‐4.52 ‐46.59 47.65 1.78 2301.25 42.62DID=|D2|‐|D1| 4.19 45.91 47.62 1.65 2300.71 42.08P‐value:DID 0.89 0.56 0.59 0.96 0.00 0.59DIDrank 25 16 17 27 1 17W‐weightsAlabama 0.00 0.00 0.00 0.00 0.00 0.00Alaska 0.00 0.00 0.23 0.00 0.00 0.00Arkansas 0.00 0.00 0.00 0.00 0.00 0.00Florida 0.25 0.32 0.00 0.17 0.00 0.00Georgia 0.00 0.00 0.00 0.00 0.00 0.55Idaho 0.00 0.00 0.00 0.00 0.00 0.00Illinois 0.00 0.04 0.00 0.00 0.13 0.16Indiana 0.00 0.00 0.00 0.00 0.79 0.15Kansas 0.00 0.00 0.00 0.00 0.00 0.00Kentucky 0.00 0.00 0.00 0.00 0.00 0.00Louisiana 0.00 0.00 0.00 0.00 0.00 0.00Michigan 0.45 0.00 0.00 0.55 0.00 0.00Mississippi 0.00 0.00 0.03 0.00 0.00 0.00Missouri 0.00 0.00 0.00 0.00 0.00 0.00Nebraska 0.00 0.00 0.00 0.00 0.00 0.00NorthDakota 0.00 0.00 0.00 0.00 0.00 0.00Ohio 0.26 0.65 0.54 0.00 0.00 0.00Oklahoma 0.00 0.00 0.00 0.00 0.00 0.00SouthCarolina 0.00 0.00 0.00 0.00 0.00 0.00SouthDakota 0.00 0.00 0.00 0.00 0.00 0.00Tennessee 0.00 0.00 0.00 0.00 0.00 0.00Utah 0.04 0.00 0.13 0.00 0.00 0.00Vermont 0.00 0.00 0.04 0.28 0.00 0.00Virginia 0.00 0.00 0.03 0.00 0.08 0.14WestVirginia 0.00 0.00 0.00 0.00 0.00 0.00Wyoming 0.00 0.00 0.00 0.00 0.00 0.00ListofPredictors (a) Common set of predictors: Total nameplate capacity growth, coal generation share, natural gas generationshare, electricity price, growth of total customer, 2010 wind potential, 2010 photovoltaic potential, 2010biopower‐solid potential, 2010 geo‐ & hydro‐thermal potential, real PC personal income, growth in real PCpersonal income, poverty, share of manufacturing income. (b) 1990 to pre‐intervention renewables capacity(dependingontheyearofinterventionforeachtreatmentstate).Notes:(a)Outcomevariableisrenewablescapacityofgeothermal,biofuels,solar,andwind.(b)YearoftheenactingRPS:Nevada(1997),Connecticut(1998),Maine(1999),NewJersey(1999),Texas(1999),andWisconsin(1999).(c)Donorpool includesAlaska, therefore the set of predictors doesnot include the geographical variables and1991windpotential.However, 2010measure ofwindpotential is included. (d)Weights less than0.01 are reportedaszero.
Page 32
31
Table 3: SCM Estimate of the Impact of RPS on Renewables Capacity (Robustness Check withGeographicalVariables) Nevada Connecticut Maine NewJersey Texas Wisconsin
Estimationsummary Pre‐interventiondifference(D1) 0.35 0.61 ‐0.04 0.69 ‐0.74 ‐0.57Post‐interventiondifference(D2) ‐4.56 ‐45.00 ‐12.32 ‐9.67 2137.23 21.64DID=|D2|‐|D1| 4.20 44.38 12.28 8.97 2136.49 21.07P‐value:DID 0.92 0.62 0.85 0.88 0.00 0.85DIDrank 25 17 23 24 1 23W‐weights Alabama 0.00 0.00 0.00 0.00 0.00 0.00Arkansas 0.00 0.00 0.00 0.00 0.00 0.00Florida 0.25 0.32 0.00 0.24 0.00 0.00Georgia 0.00 0.00 0.00 0.07 0.00 0.00Idaho 0.00 0.00 0.00 0.00 0.00 0.00Illinois 0.00 0.03 0.00 0.06 0.18 0.15Indiana 0.00 0.00 0.00 0.00 0.00 0.60Kansas 0.00 0.00 0.24 0.00 0.63 0.00Kentucky 0.00 0.00 0.00 0.00 0.00 0.00Louisiana 0.00 0.00 0.00 0.01 0.00 0.00Michigan 0.45 0.00 0.00 0.14 0.00 0.00Mississippi 0.00 0.00 0.00 0.04 0.00 0.00Missouri 0.00 0.00 0.00 0.00 0.00 0.00Nebraska 0.00 0.00 0.00 0.00 0.00 0.00NorthDakota 0.00 0.00 0.00 0.02 0.09 0.12Ohio 0.26 0.65 0.54 0.00 0.00 0.00Oklahoma 0.00 0.00 0.00 0.01 0.00 0.00SouthCarolina 0.00 0.00 0.00 0.11 0.00 0.00SouthDakota 0.00 0.00 0.00 0.00 0.00 0.00Tennessee 0.00 0.00 0.00 0.00 0.00 0.00Utah 0.05 0.00 0.14 0.00 0.00 0.00Vermont 0.00 0.00 0.04 0.27 0.00 0.00Virginia 0.00 0.00 0.03 0.02 0.10 0.12WestVirginia 0.00 0.00 0.00 0.00 0.00 0.00Wyoming 0.00 0.00 0.00 0.00 0.00 0.00ListofPredictors (a) Common set of predictors: Total nameplate capacity growth, coal generation share, natural gas generationshare,electricityprice,growthoftotalcustomer,2010windpotential,2010photovoltaicpotential,2010biopower‐solidpotential,2010geo‐&hydro‐thermalpotential,realPCpersonalincome,growthinrealPCpersonalincome,poverty, share of manufacturing income, 1991 wind potential, January sunlight, summer cooling degree days,summer heating degree days. (b) 1990 to pre‐intervention renewables capacity (depending on the year ofinterventionforeachtreatmentstate).Notes:(a)Outcomevariableisrenewablescapacityofgeothermal,biofuels,solar,andwind.(b)YearoftheenactingRPS:Nevada(1997),Connecticut(1998),Maine(1999),NewJersey(1999),Texas(1999),andWisconsin(1999).(c)Geographicvariablesand1991windpotentialmeasurearemissingforAlaska;therefore,Alaskaisexcludedfromthedonorpool.(d)Weightslessthan0.01arereportedaszero.
Page 33
32
Table 4: SCM Estimate of the Impact of RPS on Renewables Capacity in Texas (AdditionalRobustnessChecks) (1) (2)Estimationsummary
Pre‐interventiondifference(D1) 1.34 13.30Post‐interventiondifference(D2) 2060.78 2090.37DID=|D2|‐|D1| 2059.44 2077.07P‐value:DID 0.00 0.00DIDrank 1 1W‐weightsAlabama 0.00 0.00Alaska 0.00 0.00Arkansas 0.00 0.00Florida 0.00Georgia 0.00Idaho 0.00 0.00Illinois 0.24Indiana 0.00 0.00Kansas 0.00Kentucky 0.00 0.00Louisiana 0.00 0.06Michigan 0.00Mississippi 0.00 0.00Missouri 0.00Nebraska 0.00 0.00NorthDakota 0.00 0.00Ohio 0.00Oklahoma 0.76 0.94SouthCarolina 0.00SouthDakota 0.00 0.00Tennessee 0.00 0.00Utah 0.00Vermont 0.00Virginia 0.00WestVirginia 0.00Wyoming 0.00 0.00ListofPredictors(a)Commonsetofpredictors:Totalnameplatecapacitygrowth,coalgenerationshare,naturalgasgenerationshare, electricity price, growth of total customer, 2010 wind potential, 2010 photovoltaic potential, 2010biopower‐solid potential, 2010 geo‐ & hydro‐thermal potential, real PC personal income, growth in real PCpersonalincome,poverty,shareofmanufacturingincome.(b)Column1includes1998renewablescapacityasthe only pre‐intervention outcome. (c) Column 2 includes 1990‐1998 renewables capacity as the pre‐interventionoutcome.Notes:(a)Tocheckifmatchingoncapacitiesisdrivingtheresults,incolumn(1)matchingindoneonlyon1998capacity. In column (2) donor pool includes states that are both non‐RPS and non‐deregulated states. (b)Outcomevariableisrenewablescapacityofgeothermal,biofuels,solar,andwind.(c)Yearofinterventionis1999(theyearRPSenactedinTexas).(d)ThecommonsetofpredictorsisthesameasthatinTable2.(e)Weightslessthan0.01arereportedaszero.
Page 34
33
AppendixA:RPSMandatebyStateandYearofImplementationState Yeareffective FinalMandate State Yeareffective FinalMandateArizona 2007 15%by2025 Montana 2005 15%by2015California 2003 25%by2016 Nevada 1997 25%by2025Colorado 2005 20%by2020 NewHampshire 2007 25%by2025Connecticut 1998 27%by2020 NewJersey 1999 22.5%by2021Delaware 2005 25%by2025 NewMexico 2004 20%by2020Hawaii 2004 40%by2030 New York 2004 29%by2015Illinois 2011 25%by2025 NorthCarolina 2008 12.5%by2021Iowa 1983 105MWby1999 Ohio 2009 12.5%by2024Kansas 2009 20%by2020 Oregon 2007 25%by2025Maine 1999 40%by2017 Pennsylvania 2005 18%by2020Maryland 2004 20%by2022 RhodeIsland 2004 16%by2019Massachusetts 2002 15%by2020 Texas 1999 10,000MWby2025Michigan 2008 10%by2015 Washington 2007 15%by2020Minnesota 2007 25‐30%by2020 Wisconsin 1999 10%by2015Missouri 2009 15%2021 Notes:(a)Statesinboldaretheearlyadopterstates.(b)AlthoughIowaadoptedanRPSin1983,theirimplementationpre‐datesthecapacitydataavailableandtheyarethereforenotanalyzed.(c)Thefinalmandatesofthepolicieshaveevolvedovertime,oftenbecomingmorestringent.Thelatestpolicyineffectduringthe1994‐2012periodislisted.(d)Inthe‘FinalMandate’column,thepercentagesindicatethepercentofelectricitytobegeneratedfromrenewableenergy.
Page 35
34
AppendixB:Proceduretoobtain W
Let )1( 0 T vector ),...,(01 TkkK define a linear combination of pre‐intervention
outcomesis
T
s si YkY 0
0
~ K . Define )~,...,~,( 11111 MYY KKZX as a )1( k vector of pre‐
intervention characteristics for the exposed statewhere Mrk .30 Similarly, define a
)( Jk matrix 0X that contains the same variables for the unexposed states. The thj
columnof 0X ,thus,is )~,...,~,( 1 Mjjj YY KKZ .
LetVbea )( kk symmetricpositivesemidefinitematrix.Then,
(4) 1and}1,...,2|0{)()(argmin 1
20101
J
j jj wJjwWXXVWXXWW
.
FollowingAbadieandGardeazabal(2003)andAbadie,DiamondandHainmueller(2010),
wechoose V amongpositivedefiniteanddiagonalmatricessuch that themeansquared
prediction error (MSPE) of the outcome variable is minimized for the pre‐intervention
periods.
As Abadie, Diamond and Hainmueller (2010) argue, it is important to note that
unlike the traditional regression‐based difference‐in‐difference model that restricts the
effectsoftheunobservableconfounderstobetime‐invariantsothattheycanbeeliminated
bytakingtimedifferences,SCMallowstheeffectsofsuchunobservablestovarywithtime.
More details of the synthetic control, the procedure to calculate W , and
permutation/randomizationtestsortheinferencecanbefoundinAbadieetal.(2010)or
obtainedfromtheauthorsonrequest.
30 For example, if )0,...,0,1(,2 1 KM and )1,...,0,0(2 K then ),,(
0111 TYYZX , that is the
outcomevaluesofTexasforthefirstyear(year2000)andtheyearbeforethepassingoftheRPS(year2004)
areincludedin 1X .