!! You!Say!You!Want!a!(Rose)!Revolution?!! The!Effects!of!Georgia s!2004!market!reforms!!!! Kevin!Grier! Texas!Tech!University!! Robert!Lawson*!

Transcription

1 YouSayYouWanta(Rose)Revolution? TheEffectsofGeorgia s2004marketreforms KevinGrier TexasTechUniversity RobertLawson* SouthernMethodistUniversity SamuelAbsher UniversityofOklahoma Abstract: AfterGeorgiadeclareditsindependencefromtheSovietUnion,itexperiencedcivil warandstrifeformorethanadecade.inlate2003,thepeacefulroserevolution installedanewgovernmentthatbeganaseriesofradicalmarketsliberalreforms. Theeffectivenessofthesereformsisamatterofcontroversy.Thisstudyshedslight onthedebatebyusingthesyntheticcontrolmethodtocreateacredible counterfactualtoexaminetheimpactoftheseliberalizationsongeorgia s subsequentsocioseconomicperformance.weconcludethatthereformsenacted aftertheroserevolutionledtosignificantimprovementsinthegeorgia ssocialand economicdevelopment,withonlytemporarysideeffectsandarguethatthiscase providessomesupportfortheideaofrapid,multidimensionalreform. Thisdraft,September,2017 *Correspondingauthor Keywords:Georgia,SyntheticControl,Liberalization,RoseRevolution,Reform JELCodes:P1,P2

2 I.#Introduction# Duringthelateeighteenthandearlynineteenthcenturies,theRussian EmpireannexedmuchoftheregionsouthoftheCaucasusMountains,includingthe KingdomofGeorgia.AfterabriefperiodofindependenceduringtheBolshevik Revolution,GeorgiafellbackintoRussianhandsin1921andeventuallywas organizedasthegeorgiansovietsocialistrepublic.ultimately,therussians controlledgeorgiaforthebetterpartoftwocenturies.georgiaachieved independenceonapril9,1991,justbeforethepoliticalcollapseofthesovietunion. 1 TheimmediatepostSSovietexperiencewaschaoticfeaturingabloodycoup d étatandviolentethnicconflictsinthegeorgianregionsofabkhazia,southossetia, andadjara.economically,thingswerelittlebetter.thebreakupofthesovietunion eliminatedtheestablished,ifeconomicallyirrational,sovietcentralizedeconomic plan.manystatesownedenterprisesnowownedbythegeorgiangovernment flounderedasthepatternsofspecializationandtradewithintheoldsovietunion collapsed.withoutstabletaxrevenueorstatesownedenterpriseearnings,the governmentresortedtoprintingmoney,whichresultedinperiodsofhyperinflation. Shortagesofbasicitemssuchasfood,fuel,andelectricitybecamethenormthanks topricecontrolsandfrequentsupplydisruptions. Inlate2003,aftermorethanadecadeofcivilwar,politicaluncertainty,and economiccollapse,theroserevolutionpeacefullydeposededuardshevardnadze, theformersovietofficialwhohadruledautocraticallyformostofthepostssoviet period.injanuary2004,acolumbiauniversityseducatedlawyer,mikheil 1SeeBurakovaandLawson(2013)foramoredetailedtreatmentofthepoliticaleconomyofGeorgia. 1

3 Saakashvili,waselectedtoleadthenewgovernment,andKakhaBendukidze,was tappedbysaakashvilitoleadthereformprocess.althoughtheissuesrelatedtothe separatistregionsandhostilerelationswithrussiaremained,theroserevolution ultimatelyusheredinaperiodofrapideconomicliberalization. Thepurposeofthisstudyistoexaminetheimpactofthesereforms.Inorder todothisweneedareasonablecounterfactual.thatis,weneedtoknowwhat wouldhavehappenedingeorgiaabsentthereforms.tothatend,weemploythe SyntheticControlMethod(SCM)thatallowsustocompareGeorgia ssocioseconomic performanceaftertheroserevolutionagainstaweightedaverageofcomparable countrieswithsimilarcharacteristics.wewilldescribethemethodindetailina latersection. Tosummarizeourfindingswefindlargepositiveandsignificanteffectsof reformonincomesandlargenegativeandsignificanteffectsoninfantmortality relativetoourstatisticalcounterfactual.inequalityincreasesandemployment declinesandthenrecoversbuttheeffectsarenotstatisticallysignificantwhen comparedtothetimepathofoursyntheticcontrol.weconcludethattherose RevolutiononbalancewasquitepositiveforthecountryofGeorgia. Inwhatfollowsbelow,weprovideabriefoverviewofGeorgia sexperience withliberalization,explainhowthescmmethodworks,andthenpresentour resultsontheeffectoftheroserevolutiononperscapitaincome,infantmortality, incomeinequality,andemployment.. II.#Liberalization#in#Georgia# 2

4 ThemarketSliberalreformsunleashedinGeorgiahaveearnedthecountrya reputationasoneofthefastestsreformingnationsintheformersovietbloc,ifnot theentireworld.simeondjankov,founderoftheworldbank sdoingbusiness projectandformerministeroffinanceofbulgaria,saidgeorgia sreformswere unprecedented, (USAID,2009:43)withtheclosestanaloguesbeingSingaporein the1960s,southkoreainthe1970s,irelandinthe1980s,andestoniainthe1990s. USAID(2009:3)claimsGeorgiahasmade thebroadest,deepest,fastestbusiness climatereformsofanycountryinthelastfiftyyears. Evidenceoftherapidpaceofreformcanbefoundinvariousmeasuresofthe policyandinstitutionalenvironment.figure1showsthe2004s2014ratingsfor GeorgiafortheEconomicFreedomoftheWorldIndex(Gwartney,Lawson,andHall, 2016),theDoingBusinessforEnforcingContracts(WorldBank,2017),andthe CorruptionPerceptionIndex(TransparencyInternational,2016).Eachdataseries beginswithastartingvalueofzero,andthedataareexpressedinstandardized units.theeconomicfreedomoftheworldratingincreasedby0.90standardunits. Theratingforenforcingcontractsimprovedby0.74.Finally,theCorruption PerceptionIndeximprovedbyanimpressive1.46standardunits.Allthree indicatorsagreethatgeorgiahasexperiencedsignificantimprovementinitspolicy environment. BurakovaandLawson(2013)listanddescribe28distinctreforminitiatives undertakenbetween2004and2007duringtheheydayofthereformprocessunder Saakashvili sandbendukidze sleadership.majortaxreformsincludedboth reducingthelevelsandnumberoftaxes.theincometaxwasinitiallysetataflat 3

5 12%butwaslatercombinedwiththepayrolltaxata20%flatrate.Privatizationof vastnumbersofstatesownedassetswasachievedmainlythroughopenand transparenthighestsbidderauctionswithverylittleapparentfavoritism.thenew laborcodewasonlyafewpageslongandofferedfewguaranteestoworkers; unions,thoughperfectlylegal,weregivennospeciallegalstatus thisofcoursewas amajordeparturefromtheoldsovietsystem.interestingly,georgiaabolishedantis monopolylawswhilesimultaneouslyopeningtoexternaltradeasmostgoodscan beimportedtarifffree.soscalled desbureaucratization effortsresultedin95% reductionsinexecutivebranchstaffingandtheeliminationof6entireministries and18agencies.tocombatcorruptionatthelocallevel,saakashvili sgovernment shockedthenationbyfiringthenation sentirepoliceforce,some30,000officers. ThereisvirtuallynoscholarlyliteratureontheeconomiceffectsoftheRose Revolutionorthemarketliberalreformsthatensued.Papava(2005,2013,and 2014)doesofferhighlycriticalaccountsofthepostSRoseRevolutionperiod, paradoxicallycallingit asymbiosisofneoliberalismandneobolshevism (Papava, 2013:51),butthereisessentiallynodataanalysisinthesepapers.Mostoftherest ofthescholarlyliteraturedealswiththepoliticalandinternationalrelations situation(seeforexample,mitchell,2008)andissimilarlyanecdotal. 2 Thesheerscaleandscopeofthesereformsandthespeedwithwhichthey wereimplementedwasquiteastounding.thequestionisdidtheymatter?thatis, 2TangiashviliandSlade(2014)doofferaninterestingandquitecriticaldataSdriven analysisofgeorgia szerostolerancepoliciesrelatedtoschoolviolenceinsaakashvili stwo termsinoffice. 4

6 weretheserapid,multidimensionalreforms, shocktreatment ifyouwill,agood thingforthecountry? # III.#Creating#the#Synthetic#Control# WewanttoestimatetheimpactofthemarketSliberalreformsunleashedby theroserevolutioningeorgiaonvarioussocioseconomicoutcomes.ideallywe wouldliketoobservewhathappenedtogeorgiaintheabsenceofthereforms.of course,thisisimpossibleasthereisonlyonegeorgiaanditexperiencedthe reforms,andthereisnogeorgiathatdidnotexperiencethereforms.instead,we usepresrevolutiondatatocreatea syntheticgeorgia,aweightedaverageof controlcountriessimilartogeorgia.thissyntheticcontrolisdesignedtotrackpres revolutiongeorgianoutcomesandalsotomatchactualgeorgiaonthevaluesof severalindicatorvariablesaswell.totheextentthesyntheticgeorgiaaccurately representstheactualgeorgia,itcanserveasthecounterfactualweneed. AbadieandGardeazabal(2003)developedthesyntheticcontrolmethodin ordertoidentifytheeffectofterrorisminthebasquecountryinspain.abadieetal. (2010)explainthetechniqueindetailandanalyzetheeffectofachangein California'stobaccoregulations.Abadieetal.(2015)furtherrefinethemethodand applyittotheeconomiceffectsofgermanreunification.gautieretal.(2009)and Montalvo(2011)provideadditionalapplicationstotheanalysisofterrorism. Syntheticcontrolmethodshavebeenusedtostudyaffirmativeaction(Hinrichs, 2012),compulsoryvoting(Fowler,2013),economicliberalization(Billmeierand 5

7 Nannicini,2013),naturaldisastersCavalloetal.(2013),andtheimpactofHugo ChavezinVenezuela(GrierandMaynard,2016). Themostimportantaspectsofthemethodaretheselectingthecountries,the indictorvariablesandwhatweightstoassigntothecountriescomprisingthe syntheticgeorgia.ourgoalistocreateasyntheticversionofgeorgiafromamonga listofpotentiallysimilarcountriesthatmimicstheconditionsingeorgialeadingup totheroserevolution.thus,inordertocreatethesyntheticcontrol,weneedto findasetofcountriesandcountrysweightsthatminimizesthepresroserevolution differencesinsocioseconomicconditionsbetweenactualgeorgiaandthesynthetic controlgeorgia. InadditiontosimilarsocioSeconomicconditionsleadingupto2004,wealso wantoursyntheticcontrolcountriestobeeconomicallyandstructurallysimilarto Georgia.Tothatend,wealsousevariousindicatorvariablestominimizethe differencebetweentheweightedaverageofthesevariablesforthesyntheticcontrol andgeorgia.inthefinalanalysis,countrieswithmoresimilarindicatorsandmore similarsocioseconomicoutcomeswillreceivehigherweightsincreatingthe syntheticcontrol.seetheappendixforthemathematicaldetailsoftheprocess. Oneissueeconomistsoftenhavewithcasestudiesisthelackofsignificance testsfortheresults.abadieet.al.addressthisissuebymeansofpermutation testing.specifically,wewanttoseewhetherthedeteriorationintheabilityofthe controltomatchgeorgiapost2004islargerelativetorandomdeviationsusingthe procedurewheretherewasnointervention.thatis,weconductasyntheticcontrol analysisasdescribedaboveforallthecountriesinourcontrolgroup.thesetests 6

8 aregenerallyreferredtoasplacebotests.sincetheothercountriesdidnotundergo theroserevolutionin2004,wewouldnotexpecttoseesignificantchangesinthe predictabilityoftheiroutcomes. WeusetheseplaceboteststocreateauniquepSvalueforeachpostS treatmentperiod,allowingustoseehowthesignificanceoftheestimatedeffect changesovertime.tocalculatethepsvaluesfortime,wecollecteachplacebo s estimatedtreatmenteffectforthethperiodanddividethembytheirrespective prestreatmentrmspe.thisdivisionminimizesthepotentialproblemofpoorpres treatmentfitdrivinganestimatedeffectinapoststreatmentperiod.sincewe conductatwossidedtest,itisalsonecessarytocalculatetheabsolutevalueofeach effect.oncethestatisticsarecollected,wegenerateperiod spsvalueestimateby countingthenumberofplaceboswithgreatereffectsthanthatofgeorgiaanddivide bythenumberoftotalcountries(donorsandgeorgia).theprocessisrepeatedfor allpoststreatmentperiods. 3 Weusethe Synth_Runner software(galiano& Quistorff,2016)toimplementthissignificancetestinginStata.# IV.#The#Experiments# WeconductfourquasiSexperimentsinthispaper.EachcomparesthepostS RoseRevolutionexperiencefrom2004onwardoftheactualGeorgiawithits 3WealsoreportresultfromasecondmethodthatyieldsasinglepSvaluefortheentirepost interventionperiod.theprocessforcalculationisnearlyidenticaltothebysperiodmethod. However,ratherthantakingeachperiod sestimatedeffect,thistechniqueusestheoverall poststreatmentrmspeforeachdonor,whichisdividedbyprestreatmentrmspe.weuse theabsolutevaluesforatwossidedtestliketheformertechnique,anduseanidenticalrank systemtocalculateapsvalue.thisproduceswhatcavalloetal(2013)call pseudots statistics 7

9 syntheticgeorgiacounterpart.theoutcomevariablestobeconsideredarepers capitagdp,infantmortality,incomeinequality,andtheemployment/population ratio.ineachseparatecase,asyntheticgeorgiawillbeconstructedthatmirrorsthe actualgeorgia spresroserevolutionrecordfrom1995s2003forthatoutcome variable. 4 Table1providesdescriptivestatisticsforboththeindicatorvariablesand theoutcomevariablesused.thefirstcolumnoftable2listsallthepotentialdonor countriesincludedinthestudy.thislistincludesformersovietrepublicsand CommunistSblocnations,Russiaitself,aswellasafewcountriesfromtheregion suchasturkey,israel,iran,andegypt.allofthesecountriesshareeither Communistbackgroundsorsimilarregionalinfluences. 5 Thelastfourcolumnsof Table2liststhecountryweightsgeneratedbythesyntheticcontrolmethodto createasyntheticgeorgiaforeachofthefourexperimentsweconsider. A:PerSCapitaGDPResults ThefirstexperimentlooksatperScapitaGDP.Sixindicatorvariables,foundin Table3,wereusedtominimizethedifferencesbetweenperScapitaGDPinGeorgia 4DataforGeorgiabeginin1991,butaseverhyperSinflationin 92S 93makeitimpossible forustofindasyntheticthatlooksatalllikegeorgiaintermsofmacroeconomicandsocial indicatorsforthoseyears.thuswebegintheanalysisin Wedonotclaimthatnoneofthesecountriesexperiencedanytypeofeconomicreform. WeareclaimingthattheextentandrapidityofGeorgia sreformexperienceisuniqueinthis groupoverthe2004s20014period.ofallthecountriesinourlist,georgia sexperienceon thethreereformindicatorslistedinthetextisthelargest.thenextclosestcountriesare PolandandSerbia.Interestingly,neitherofthesetwocountriesmakeitintoanyofthe estimatedsyntheticcontrolsinthepaper.totheextentthatcountriesinthesyntheticshave experiencedsimilarreformstogeorgia,ourresultsontheimpactofgeorgia sreformswill beunderstated. 8

10 andsyntheticgeorgia,whichiscomprisedof26%armenia,21%bosniaand Herzegovina,and54%Moldova.WhencomparingGeorgia sperscapitagdpwith thatofitssyntheticcontrolcounterpartduringthepresroserevolutionera,wefind armspeof thiscalculationhighlightsthevalueofcreatingasynthetic controlusingaweightedaverageofsimilarnationsinsteadofjustasingle comparisonnation.theclosestsinglecountrytogeorgia spresroserevolutionpers capitagdpperformancewasarmenia,butthermspebetweenarmenia sand Georgia sperscapitagdpwas duringtheprestreatmentperiod,nearly seventimeslargerthanitwasforthesyntheticcontrol. AsFigure2shows,thepreSRoseRevolutionactualGeorgiaandsynthetic Georgiaperformedverysimilarly,butaftertheRoseRevolution,perScapitaGDP grewsubstantiallyfasterinactualgeorgiarelativetoitssyntheticcontrol.this suggeststhattheimpactoftheroserevolutionandthemarketliberalreformsgave perscapitagdpaboostingeorgiaoverwhatwouldhavehappenedotherwise.and notjustasmallboost:bytheendofourstudyperiodactualincomeisover33% higherthanpredictedbythecontrolandremember,thecontrolwasveryaccurate inpredictinggeorgianincomebeforetherevolution. Figure3presentssignificancetestsforeachyearandshowsthatthelarge estimateddifferencesbetweengeorgia sactualpostsroserevolutionexperience anditssyntheticcontrolwerefrequentlystatisticallysignificant,especiallybythe endoftheperiod. 6 6Itistruethatthesignificancelevelsdonotoftenhit0.05,butthesizeoftheeffects coupledwiththeirmarginalsignificancemeansweshouldtakethemseriously.foragood discussionofeconomicvs.statisticalsignificance,seegoldberger(1991)ormccloskey& 9

11 B:InfantMortalityResults Thesecondexperimentlooksatinfantmortality.Sixindicatorvariables, foundintable4,wereusedtominimizethedifferencesbetweeninfantmortalityin GeorgiaandthesyntheticcontrolGeorgia.Forinfantmortality,thesynthetic Georgiawasmadeupof30%Albania,23%Armenia,33%Moldova,9%Slovenia, and4%ukraine.thepresroserevolutionrmspeofpredictinggeorgiabyits syntheticcontrolwasonlyabout1/25 th asgreatasthermspeforpredicting Georgiabyitsbestsinglecomparisoncountry,Turkey.Thesyntheticcontrolversion ofgeorgiacloselytracksgeorgia sinfantmortalitypriortotheroserevolution. AsFigure4illustrates,thepreSRoseRevolutionactualGeorgiaandsynthetic Georgiaperformedverysimilarly,butaftertheRoseRevolutioninfantmortalityfell substantiallyfasterinactualgeorgiacomparedwithitssyntheticcontrol.infant mortalityimprovedaftertheroserevolutionfasterthanitwouldhavewithoutthe intervention.again,theeffectissubstantial.attheendoftheperiod,infant mortalityisalmost25%loweringeorgiacomparedtoitssyntheticcontrol. Figure5presentsoursignificancetestsbyyearfortheeffectoftheRose Revolutiononinfantmortality.Allarestatisticallysignificantatconventionallevels. Similarly,theaggregatepSvalueis0.069,furtherdemonstratingthestatistical significanceoftheresults.thislendssupporttotheideathatthereductionininfant mortalityingeorgiawasinfactattributabletotheroserevolutionanditspolicy Ziliak(1996).TheoverallpSvalueascalculatedbyoursecondmethodis0.14fortheentire period.# 10

12 reforms.thebottomlineisthatweseealargeandsignificantimprovementin infantmortalitybeginningin2004ingeorgia,i.e.,aftertheroserevolutionrelative tothepredictionofthe businessasusual syntheticcontrol. C:IncomeInequalityResults Liberalizationsareoftenaccusedofsacrificingequityforgrowth.Thisthird experimentisdesignedtoassessthisargumentbyexaminingincomeinequalityas measuredbytheginicoefficientprovidedbysolt(2016). 7 Threeindicator variables,foundintable5,wereusedtominimizethedifferencesbetweenthegini ingeorgiaandthesyntheticcontrolgeorgia.fortheginicoefficientexperiment, syntheticgeorgiawascomprisedof27%macedonia,32%turkey,and41% Turkmenistan.Again,themethoddoesanexcellentjoboffindingasyntheticversion ofgeorgiathatmatchesgeorgia sincomeinequalitybefore2004.thermspe betweengeorgiaanditssyntheticcontrolwasjust0.228,whichislessthanones tenthasmuchasthermspewithturkmenistan,georgia sclosestsingle comparisonwithrespecttoincomeinequality. Figure6showsthatthepreSRoseRevolutionactualGeorgiaandsynthetic Georgiaperformedverysimilarly,butaftertheRoseRevolutiontheGINIincreased inactualgeorgiacomparedwithitssyntheticcontrol.itappearsincomeinequality worsenedaftertheroserevolutionfasterthanitwouldhavewithoutthe 7WeuseStata s ipolate commandtofillinmissingdata.thecommandusesthefollowing formulatoestimatemissingyvalues:y = x x + y,suchthatx < xandx > x, wherey andy areobserved. 11

13 intervention.weshouldnotethoughthattheestimatedeffectisverysmall.atits peakin2008,thedifferencebetweenthetwoserieswasonlyaround5%. Figure7presentsoursignificancetests,andshowsthattheeffectisnot significantinanyoftheyearsfrom2004on.theaggregatepsvalueisalso completelyinsignificant,at0.414.georgia spostsroserevolutionginiwasindeed slightlyhigherthanexpectedwhencomparedwithitssyntheticcontrol,butthe variationwasnotmuchdifferentthatwhatweseeinthecontrolcountries. Georgia sroserevolutiondidnotmateriallyaffecttheevolutionofgeorgianincome inequality. D:EmploymentResults Thefinalexperimentexaminesemployment,specificallythe employment/populationratio.westudyemploymentbecauseweknowtherose Revolutionhadadirectanddramaticimpactongovernmentemployment.Ina countryoffewerthan5millionpeople,95%forcereductionsinvariousgovernment ministriesandfiringtensofthousandsofpoliceofficersseemslikelytoimpact aggregateemploymentnumbers.inthiscaseourstatisticalprocedurechoosesa syntheticcontrolcomprisedofalbania(33%),azerbaijan(5%),israel(8%), Kazakhstan(34%),andRomania(20%).Table6showstheindicatorsvariablesused andthatthissyntheticcontrolagainoutperformstheclosestsinglecountry (Tajikistan)inmimickingGeorgia semployment/populationratioleadingupto 2004.After2004,wefindalargeinitialdropinGeorgianemploymentrelativetoits syntheticcontrolandthenasharpreboundaround2009asshowninfigure8. 12

14 Figure9containstheyearbyyearsignificancetestsandshowstheeffectoftheRose Revolutiononemploymentwasnotstatisticallysignificant. 8 Again,giventhesizeoftheseeffects,thereislittlequestionthatliberal reformscancauseshortrunlabormarketdisruptionsasemployeesinthe governmentitselfandinnewlyprivatizedstatesownedenterpriseslosetheirjobs. WeabsolutelyfindthisinthecaseofGeorgia sroserevolution.thegoodnewsis thatthenewliberallaborcodealongwiththeaccelerationineconomicgrowth appearstohaveallowedtheseworkerstheopportunitytofindnewemployment withinareasonabletimeframesothatthenegativeeffectsweretemporary. # V.#Conclusion# Byallaccounts,Georgiaexperiencedoneoftheworld sfastestanddeepest economicsliberalizationsduringtheyearsimmediatelyfollowingitsrose Revolutioninlate2003.However,itisincorrecttotryandmeasuretheeffectofthe reformssimplybylookingatchangesingeorgia ssocioseconomicperformance.the relevantcomparisonistowhatwouldhavehappenedifingeorgiaifthereforms hadnotoccurred. ThisstudyusestheSCMtoexaminehowGeorgiaperformedcomparedtoa syntheticcontrolversionofgeorgia,aweightedaverageofsimilarcountries,that didnotexperiencethesamereformprocess.theresultssuggestthattheperscapita 8AppendixTables2and3reporttheemploymentexperimentagain,butwiththeaddition ofperscapitagdpasapredictorvariable.thisinclusionworsensthecovariatebalance,but createsacloserprestreatmentfit.thegeorgianemploymentdeclinesaregreater,and statisticallysignificantinsomeyearsinthisspecification.qualitatively,though,thepicture issimilarwithemploymentreboundingbytheendoftheperiod. 13

15 GDPandinfantmortalityimprovementsthatGeorgiaexperiencedaftertheRose Revolutionweremuchgreaterthanwouldhavebeenexpectedintheabsenceofthe reforms.incontrast,thereisnoevidencesupportingthenotionthattherose Revolution smarketsliberalreformscausedanyincreaseinincomeinequality. Finally,theliberalizationsdoappeartohavecausedlargeemploymentlossesinthe yearsimmediatelyfollowingtheroserevolution,thoughthedislocatedworkers weremostlyabletofindnewemploymentbytheendoftheperiodperhapsthanks tothenewlyliberalizedlabormarket. Insum,theRoseRevolutiongeneratedlargeandsignificantincreasesinperS capitaincomes(33%)alongwithlargeandsignificantdeclinesininfantmortality(s 25%)relativetothesyntheticcontrol,withoutworseninginequality.The employmentratioinitiallyfell,butrecoveredbytheendofourperiodofstudy. Thesefindingshouldprovidereformadvocateswithsomehope.TheRose revolutiongreatlyimprovedeconomicandhealthoutcomesingeorgiawithfairly shortslastingnegativeemploymentsideeffectsandnosignificantexacerbationof inequality.shocktreatmentcanwork. # # # # # # # # # # # # 14

16 References Abadie,A.,Diamond,A.,Hainmueller,J.,2010.Syntheticcontrolmethodsfor comparativecasestudies:estimatingtheeffectofcalifornia stobaccocontrol program.journaloftheamericanstatisticalassociation105(490), Abadie,A.,Diamond,A.,Hainmueller,J.,2015.ComparativePoliticsandthesynthetic controlmethod.americanjournalofpoliticalscience59(2),495s510. Abadie,A.,Gardeazabal,J.,2003.Theeconomiccostsofconflict:Acasestudyofthe BasqueCountry.AmericanEconomicReview93(1), Burakova,LarisaandRobertLawson.2013.Georgia sroserevolution:howone CountryBeattheOdds,TransformedItsEconomy,andProvidedaModelforReformers Everywhere.(Guatemala,Guatemala:UniversidadFranciscoMarroquín/TheAntigua Forum). Billmeier,A.,Nannicini,T.,2013.Assessingeconomicliberalizationepisodes:A syntheticcontrolapproach.reviewofeconomicsandstatistics95(3), Cavallo,E.,Galiani,S.,Noy,l.,Pantano,J.,2013.Catastrophicnaturaldisastersand economicgrowth.reviewofeconomicsandstatistics95(5), Fowler,A.,2013.Electoralandpolicyconsequencesofvoterturnout:Evidencefrom compulsoryvotinginaustralia.quarterlyjournalofpolitcalscience8, Gautier,P.,Siegmann,A.,VanVuuren,A.,2009.Terrorismandattitudestoward minorities:theeffectofthetheovangoghmurderonhousepricesinamsterdam. JournalofUrbanEconomics65(2), Goldberger,A.S.(2000).Acourseineconometrics.Cambridge,Mass.:HarvardUniv. Press. Grier,KevinandNormanMaynard.2016.TheeconomicconsequencesofHugoS Chavez:Asyntheticcontrolanalysis.JournalofEconomicBehavior&Organization 125,1S21. Gwartney,James,RobertLawsonandJoshuaHall.2016.EconomicFreedomofthe World:2016AnnualReport.(Vancouver:FraserInstitute). Hinrichs,P.,2012.Theeffectofaffirmativeactionbansoncollegeenrollment, educationalattainment,andthedemographiccompositionofuniversities.reviewof EconomicsandStatistics94(3), McCloskey,D.andZiliak,S.1996.JournalofEconomicLiterature.TheStandardError ofregressions,34(1),97s

17 Mitchell,Lincoln.2008.UncertainDemocracy:U.S.ForeignPolicyandGeorgia srose Revolution.(Philadelphia:UniversityofPennsylvaniaPress). Montalvo,J.,2011.Votingafterthebombings:Anaturalexperimentontheeffectof terroristattacksondemocraticelections.reviewofeconomicsandstatistics93(4), Papava,Vladimir Georgia smacroeconomicsituationbeforeandafterthe RoseRevolution. ProblemsofEconomicTransition48(4),8S17. Papava,Vladimir EconomicAchievementsofPostrevolutionaryGeorgia. ProblemsofEconomicTransition56(2),51S65. Papava,Vladimir Georgia seconomy:thesearchforadevelopmentmodel ProblemsofEconomicTransition57(3),83S94. Solt,Frederick.2016."TheStandardizedWorldIncomeInequalityDatabase."Social ScienceQuarterly97.SWIIDVersion5.1,July2016. Tangiasshvili,NodarandGavinSlade ZeroStoleranceschooling:education policy,crime,anddemocracyinpostssovietgeorgia. PostWSovietAffairs30(5),416S 40. TransparencyInternational.2016.CorruptionPerceptionsIndex2016.(Washington, DC:TransparencyInternational). USAID.2009.Georgia:OpenedforBusinessGeorgiaBusinessClimateReformFinal Report(Washington,DC:USAID). WorldBank.2017.DoingBusiness:MeasuringBusinessRegulations.(Washington, DC:WorldBank). 16

18 TABLE 1: SUMMARY STATISTICS Variable Mean Standard Deviation Observations Description GDP per Capita 12, (7,874.97) 630 GDP per capita, PPP (constant 2011 international $) Infant Mortality (17.07) 630 Mortality rate, infant (per 1,000 live births) GINI (5.83) 630 GINI inequality index (SWIID) FDI 4.97 (5.84) 625 Foreign direct investment, net inflows (% of GDP) Employment Ratio (7.22) 630 Employment to population ratio, 15+ total (%) Urban (13.82) 630 Urban population (% of total) Investment (6.76) 622 Gross capital formation (% of GDP) Life Expectancy (3.97) 596 Life expectancy at birth, total (years) Military 2.26 (1.52) 586 Military expenditure (% of GDP) Inflation (80.13) 623 Inflation, GDP deflator (annual %) Note. This table contains summary statistics for all variables in all analyses. The statistics are calculated using all countries, donors and Georgia, for all periods of the analyses, There are 630 observations in the panel data. For the descriptions, we draw directly from the metadata of the respective sources. Source. World Bank's World Development Indicators, Standardized World Income Inequality (SWIID) and authors' calculations.

19 TABLE 2: ESTIMATED SYNTHETIC CONTROL WEIGHTS BY OUTCOME VARIABLE Outcome Variable Income Infant Mortality Inequality Employment Albania Armenia Azerbaijan Belarus Bosnia and Herzegovina Bulgaria Croatia Czech Republic Egypt Estonia Hungary Israel Kazakhstan Kyrgyz Republic Latvia Lithuania Macedonia Moldova Poland Romania Russian Federation Serbia Slovakia Slovenia Tajikistan Turkey Turkmenistan Ukraine Uzbekistan Note. Columns show the estimated weight for the synthetic Georgia. Each column represents an outcome variable, labelled at the top of the column. Values are in percentage points. Donors that receive a positive weight are in bold for the reader to more easily identify. Values are rounded, so the columns may not sum to one.

20 TABLE 3: AVERAGE INCOME PREDICTOR MEANS Variables Actual Georgia Synthetic Georgia Armenia Avg. GDP per Capita 3,181 3,180 2,920 FDI Urban Investment Life Expectancy Military Inflation RMSPE Note. This table shows the values of indicator variables for different comparison groups. In doing so, it illustrates the strenghts of the synthetic control, which better fits the behavior of actual Georgia in the pre-treatment period ( ). We compare the synthetic to Armenia, which is the best single-country comparison (identified by selecting the single country that minimized pre-treatment RMSPE). Variables are averaged across the pretreatment period. Please refer to table 1 for a description of the variables. The final row gives the root mean square prediction error for the unit of comparison.

21 TABLE 4: AVERAGE INFANT MORTALITY PREDICTOR MEANS Variables Actual Georgia Synthetic Georgia Turkey Infant Mortality GDP per Capita 3,181 3,186 12,542 Urban Investment Life Expectancy Military Inflation RMSPE Note. This table shows the values of indicator variables for different comparison groups. In doing so, it illustrates the strenghts of the synthetic control, which better fits the behavior of actual Georgia in the pre-treatment period ( ). We compare the synthetic to Turkey which is the best single-country comparison (identified by selecting the country that minimized pre-treatment RMSPE).. Variables are averaged across the pre-treatment period. Please refer to table 1 for a description of the variables. The final row gives the root mean square prediction error for the unit of comparison.

22 TABLE 5: AVERAGE GINI PREDICTOR MEANS Variables Actual Georgia Synthetic Georgia Turkmenistan GINI FDI Urban Investment RMSPE Note. This table shows the values of indicator variables for different comparison groups. In doing so, it illustrates the strenghts of the synthetic control, which better fits the behavior of actual Georgia in the pre-treatment period ( ). We compare the synthetic to Turkmenistan which is the best single-country comparison (identified by selecting the single country that minimized pre-treatment RMSPE). Variables are averaged across the pretreatment period. Please refer to table 1 for a description of the variables. The final row shows the root mean square prediction error for the unit of comparison.

23 TABLE 6: AVERAGE EMPLOYMENT RATIO PREDICTOR MEANS Variables Actual Georgia Synthetic Georgia Tajikistan Employment Ratio Age Dependency Ratio FDI Urban Inflation Trade RMSPE Note. This table shows the values of indicator variables for different comparison groups. In doing so, it illustrates the strenghts of the synthetic control, which better fits the behavior of actual Georgia in the pre-treatment period ( ). We compare the synthetic to Tajikistan which is the best single-country comparison (identified by selecting the single country that minimized pre-treatment RMSPE). Variables are averaged across the pre-treatment period. Please refer to table 1 for a description of the variables. The final row shows the root mean square prediction error for the unit of comparison.

24 Figure 1: Measures of Institutional Quality in Georgia Year Economic Freedom of the World Doing Business: Enforcing Contracts Doing Business: Starting a Business Corruption Perceptions Index Note. Following the Rose Revolution, Georgia's reform led to improvements in various institutional quality measures. This figure demonstrates the increases experienced by four different indeces, listed in the figure's key. Each data series begins with a starting value of zero, and the units are expressed in standardized terms. A one unit increase along the y-axis represents a one standard deviation in the respective index. Source. Gwartney, Lawson, and Hall, World Bank Group, Transparency International, Authors' calculations.

25 Per Capita GDP (2011 international $) Figure 2: Per Capita GDP Year Georgia Synthetic Georgia Note. This figure demonstrates the behavior of per capita GDP for Georgia and synthetic Georgia, pre- and post-treatment. The dashed vertical line indicates the treatment period.

26 Figure 3: The Effect of Rose Revolution on per Capita GDP Per Capita GDP Difference from Synthetic (in 2011 international $) Note. This figure shows the estimated treatment effect upon per capita GDP for each period following the Rose Revolution. Effects in orange are significant at the.04 level, effects in gold at the.10 level, in brown at the.13 level. Effects in grey are insignificant. Each period's p-value is indicated at the top of each bar. The p-value for the entire experiment is The method of p-value calculation for each method is described on page 9.

27 Figure 4: Infant Mortality Infant Mortality (per 1,000 live births) Year Georgia Synthetic Georgia Note. This figure demonstrates the behavior of infant mortality for Georgia and synthetic Georgia, pre- and post-treatment. The dashed vertical line indicates the treatment period.

28 Figure 5: The Effect of Rose Revolution on Infant Mortality Infant Mortality Difference from Synthetic (per 1,000 live births) Note. This figure shows the estimated treatment effect upon infant mortality for each period following the Rose Revolution. Effects in orange are significant at the.00 level, effects in gold at the.03 level, and brown at the.07 level. Effects in grey are insignificant. Each period's p-value is indicate at the bottom of each bar. The p-value for the entire experiment is The method of p-value calculation for each method is described on page 9

29 Figure 6: Gini Gini Coefficient (SWIID) Year Georgia Synthetic Georgia Note. This figure demonstrates the behavior of the Gini coefficient for Georgia and synthetic Georgia, pre- and post-treatment. The dashed vertical line indicates the treatment period.

30 Figure 7: The Effect of Rose Revolution on Inequality.31 Gini Index Difference from Synthetic (SWIID) Note. This figure shows the estimated treatment effect upon inequality (Gini index) for each period following the Rose Revolution. Effects in grey ar are insignificant. Effects in grey are insignificant. Each period's p-value is indicated at the bottom of each bar. The p-value for the entire experiment is The method of p-value calculation for each method is described on page 9.

31 Figure 8: Employment to Population Ratio Employment to Pop. Ratio (%) Year Georgia Synthetic Georgia Note. This figure demonstrates the behavior of employment ratio for Georgia and synthetic Georgia, pre- and post-treatment. The dashed vertical line indicates the treatment period.

32 Figure 9: The Effect of Rose Revolution on Employment Employment to Pop. Ratio Difference from Synthetic (in %) Note. This figure shows the estimated treatment effect upon employment for each period following the Rose Revolution. Effects in grey are are insignificant. Effects in grey are insignificant. Each period's p-value is indicated at the bottom of each bar. The p-value for the entire experiment is The method of p-value calculation for each method is described on page 9.

33 Appendix(A:(Synthetic(Control(Mathematics( Letj=0,,Jbethenumbercountriesinouranalysis,whereGeorgiaisj=0andJis thenumberofcountriesinthepotentialcontrolcountrypool(j=29;seetable1). Letk=1,,Kbethenumberofindicatorvariables(Kvariesbyexperiment). Letn=1,,NbethenumberofyearsinthepreKRoseRevolutionperiod(N=9since theprekroserevolutionperiodis1995k2003). LetX1representtheKx1vectorofthepreKRoseRevolutionindicatorvariablesfor GeorgiaandY1representGeorgia snx1vectorofprekroserevolutionoutcomes. LetX0betheKxJmatrixwhosecolumnsarevectorsofthepreKRoseRevolution indicatorvariablesforthepotentialcontrolcountriesandy0isthenxjmatrixwhose columnsarevectorsoftheirprekroserevolutionoutcomes. LetWrepresenttheJx1columnofcountryweights,withtheweightassignedto countryjequaltothej th individualelementofw.theprocessofselectingwisas follows: Step1.Selectweights,W,tominimizethedistancefunction: D=((X1 X0W) T V(X1 X0W)) ½, wherevisakxkpositivekdefinitediagonalmatrix,initiallysetasanidentitymatrix. ThedistanceKminimizingWvectoristhusafunctionoftheVmatrix,W(V). Step2.GiventheWmatrixfoundinStep1,calculatethemeansquaredprediction errorintheoutcomevariablebetweengeorgiaandthepotentialcontrolscountries overtheprekroserevolutionperiod: MSPE=(Y1 Y0W(V)) T (Y1 Y0W(V)) Step3.RedoSteps1and2usingallpossiblepositiveKdefinitediagonalmatricesofV. Ultimately,W * =W * (V * )isthewvectorthatgloballyminimizesthemspeofthe outcomevariablesfoundinstep2duringtheprekroserevolutionperiod.

34 Employment Ratio (Employment to Pop. Ratio (%)) Appendix B: Employment to Population Ratio Year Georgia Synthetic Georgia Note. This appendix demonstrates the behavior of employment ratio for Georgia and synthetic Georgia, pre- and post-treatment. The dashed vertical line indicates the treatment period. This specification includes GDP per capita as a predictor variable. The inclusion worsens covariate balance, but creates a closer pre-treatment fit, which is why we include the results here.

35 Appendix C: The Effect of Rose Revolution on Employment Employment to Pop. Ratio Difference from Synthetic (in %) Note. This figure shows the estimated treatment effect upon employment for each period following the Rose Revolution. Effects in grey are are insignificant. Effects in grey are insignificant. Each period's p-value is indicated at the bottom of each bar. The p-value for the entire experiment is The method of p-value calculation for each method is described on page 9.