The Impact of R&D Subsidies during the Crisis

(1)

Contents lists available atScienceDirect

Research Policy

j o u r n a l h o m e p a g e :w w w . e l s e v i e r . c o m / l o c a t e / r e s p o l

The impact of R&D subsidies during the crisis

Martin Hud

^a,b,∗

, Katrin Hussinger

^a,c,d

aZEWCentreforEuropeanEconomicResearch,Mannheim,Germany

bMaastrichtUniversity,TheNetherlands

cUniversityofLuxembourg,Luxembourg

dK.ULeuven,DepartmentofManagerialEconomics,StrategyandInnovation,Belgium

a r t i c l e i n f o

Articlehistory:

Received13April2014

Receivedinrevisedform8June2015 Accepted8June2015

JEL:

C14 C21 G01 H50 O38

Keywords:

R&D Subsidies Policyevaluation Economiccrisis Treatmenteffects

a b s t r a c t

ThisstudyinvestigatestheimpactofpublicR&DsubsidiesonR&Dinvestmentofsmallandmedium- sizedenterprises(SMEs)inGermanyduringthemostrecenteconomiccrisis.Ouranalysisisbasedon firm-leveldataoftheMannheimInnovationPanel(MIP)coveringtheperiod2006–2010.Whilewefind anoverallpositiveeffectofR&DsubsidiesonSMEs’R&Dinvestmentbehavior,thereisevidencefora crowdingouteffectforthecrisisyear2009.In2010,whentheGermaneconomystartedtorecover,the subsidyeffectissmallerthaninthepre-crisisyears,butpositiveandsignificant.Additionaltestsindicate thatthetemporarycrowdingouteffectwascausedbyreluctantinnovationinvestmentbehaviorofthe subsidyrecipientsratherthanbyGermany’scountercyclicalinnovationpolicyduringthecrisis.

1. Introduction

Theglobaleconomiccrisisof2008/2009hasseverelyaffected theOECDeconomies.Theunemploymentratehasreachedapost- warheightof8.5%inOctober2009,theGDPdeclinedby4%in2009 ascomparedto2008(OECD,2012a),andlong-terminvestments likeinnovationexpendituresdecreasedsigniﬁcantlyinarangeof countriesincludingCanada,Swedenand theUK(OECD, 2012b;

FilipettiandArchibugi,2011).

Privatesectorinnovationandresearchanddevelopment(R&D) activitiessubstantiallycontributetosustainablegrowth(Griliches, 1979;GrossmanandHelpman,1994;AghionandHowitt,2009;

DoraszelskiandJaumandreu,2013).Evenashort-termdeclineor stagnationoftheseactivitiescanhavedetrimentalconsequences inthelongrun.Policymakersarewellawareoftheimportanceof privatesectorR&DandalsoofthefactthatprivateR&Dspendingis lowerthansociallydesirable,eveninboomperiods.Forthatreason,

∗ Corresponding authorat:Centre forEuropeanEconomicResearch (ZEW), DepartmentofIndustrialEconomicsandInternationalManagement,L7,1,D-68161 Mannheim,Germany.Fax:+496211235170.

E-mailaddress:hud@zew.de(M.Hud).

publicsupportforR&Dactivitiesisparticularlyimportantintimes ofaneconomicdownturn.

R&D investment is risky and the returns are uncertain and long-term. During recessions, not only firms facing finan- cial constraints are likely to reduce their investment in R&D (Schumpeter,1939;Freemanetal.,1982).R&Dinvestmentsmight also be cut in response to a decreased demand in recession periods (Stiglitz, 1993; Aghion and Saint-Paul, 1998). Further- more,it hasbeenshown thatthe responsivenessofcompanies to policy initiatives is weaker in times of economic uncertainty (Bloom et al., 2007; Bloom, 2008). Uncertainty raises thereal option value ofinvestments, which makesfirms more cautiousconcerningtheirR&Dinvestmentdecisionsduringreces- sions.

Inorder topreventﬁrms fromreducingtheirR&D expenses andtomaintainthenationalR&Dcapacities,policymakersinmany industrializedcountries,includingAustria,DenmarkandSweden, reactedimmediatelytothemostrecentcrisisandincreasedthe publicR&Dbudgets(OECD,2012b).InGermany,theprivatesec- tor reduced R&D expensesby 2.9% (Fig. 1), while the German FederalMinistry forEducation and Research (BMBF)reactedto thecrisisbyincreasingitsbudgetby9%in2009ascomparedto 2008.

(2)

Fig.1.AnnualrealGDPgrowthandrealBusinessEnterpriseExpenditureonR&D (BERD)growthinGermanyandEU28between2000and2012,in%.

Source:OECDandOECD(2013).Owncalculations.RealGDPandBERDgrowthrates arecalculatedbasedonGDPandBERDatconstant2005USD.

ThispaperempiricallyexaminestheeffectsoftheBMBF’spub- lic R&D subsidyprogram on firms’R&D investment duringthe mostrecentcrisis.Ouranalysisisbasedonfirm-leveldataofthe Mannheim Innovation Panel (MIP) and onpublic R&D subsidy dataprovidedbytheBMBF.Wefocusonsmallandmedium-sized enterprises(SMEs)becausethesefirmsareexpectedtobemore vulnerableduringan economicdownturn ascompared tolarge enterprises.Oursamplecoverstheperiodof2006–2010,with2009 markingthepeakofthecrisisperiodinGermany(andtheEU)as revealedinFig.1.¹

Ourempiricalstrategyconsistsofseveralsteps.Werelyupon propensityscorematchinginordertoassesswhetherR&Dsubsi- diesstimulateadditionalR&Dinvestmentoverthesampleperiod.

Thematchingestimatoraccountsforthefactthatsubsidiesarenot randomlydistributedamongSMEs,butthatcompaniesself-select themselvesintothefundingscheme.Thisprocedureprovidesus withanestimateoftheaverageeffectofsubsidiesonthesubsi- dizedcompanies’R&Dinvestment.Inthenextstep,wecompare theeffectivenessofthetreatmentofthecrisisperiodtothepre- crisis(2006–2008)andpost-crisis(2010)periods.Intheﬁnalstep, wetestwhetherSMEssubsidizedduringthecrisisarelesspromis- inginnovatorsthanSMEssubsidizedbeforethecrisisorwhether theinvestmentbehaviorofsubsidyrecipientschangedduringthe crisis.Theformercouldbea consequenceofanexpanded sub- sidyprogramduringthecrisisthatcouldhaveloweredtheaverage

“quality”ofthepoolofsubsidyrecipients.

Our results show that R&D subsidies lead to an additional- ityeffectfortheoverallperiod.Onlyforthecrisisyear,wefind evidenceforacrowdingouteffectfromwhichSMEsarealready recoveringagaininthefirstpost-crisisyear.Thecrowdingouteffect ofthecrisisyearcanbeindicativeofreluctantR&Dinvestment behaviorofthesubsidyrecipientsorofthecountercyclicalinnova- tionpolicythatmayhavechangedtheaverage“quality”ofthepool ofsubsidyrecipientsduringthecrisis.Ourfurtherempiricaltests suggestthatthecrowdingouteffectiscausedbySMEs’reluctance toinvestinR&Dduringacrisisperiod.Wedonotfindsupportfor thehypothesisthatthecrowdingouteffectisrelatedtoalower average“quality”ofsubsidizedfirmsinthecrisisandpost-crisis yearsduetheexpandedsubsidyprogram.

Although ourmain ﬁnding is that theaverage additionality effect is negative for the crisis year 2009, the countercyclical

1Germanywasalreadyseverelyaffectedbythecrisisinthelastquarterof2008 reachingthepeakin2009.Thisdoesnotshowupintheaggregateannualdata though.Wethereforereferto2009asthecrisisyearintheremainderofthepaper.

innovationpolicyislikelytohavehadastabilizingeffectonthe economy.ItmayhavehelpedSMEstokeeptheirR&Dpersonnel andtomaintainnationalinnovationcapacities.²

Theremainderofthepaperisorganizedasfollows.Thenext sectionsurveysrelatedliterature.Section3presentstheempiri- calstrategy.ThedatasetisdescribedinSection4.Theresultsare discussedinSection5.Thelastsectionconcludes.

2. Literaturereview

2.1. TherationaleforpublicR&Dsubsidies

TheeconomicrationaleforR&Dsubsidiestotheprivatesec- toristhatthelevelofprivatelyﬁnancedR&Dintheeconomyis lowerthansociallydesirable.ThisisbecauseR&Dhasthecharac- teristicsofapublicgoodasitgeneratespositiveexternaleffects, whichcannotbeinternalizedbytheinnovatingcompanies(Arrow, 1962).Intheabsenceofpublicsubsidies,projectsthatwouldgen- erate positive beneﬁtsfor society but do not cover the private costs,wouldhencenotbecarried out.Thistype ofmarketfail- ureisthemainreasonforgovernmentstosubsidizeprivateR&D projects.Publicfundingreducesthepriceforprivateinvestorsso thattheotherwisetooexpensiveinnovationprojectsarecarried out.Thepolicymakers’objectiveistwofoldregardingR&Dsubsi- dies.Ontheonehand,theaimistomaintainnationalinnovation capabilities,thenationalR&Dandemploymentlevel,especiallyin recessionperiods,wheretypicallysubsidiestotheprivatesector arepreferredoverpublicprocurementbecauseofthepotentialof additionalityeffects.Ontheotherhand,thegovernment’sinterest istogeneratemoreinnovationoutcome.Publicsubsidiescanhelp theeconomyrecoveringfromacrisisbyfosteringthecreationof newinnovationsleadingtoeconomicgrowth.

ThepositiveeffectofR&D subsidiesonfirms’R&Dactivities, however,cannotbetakenforgranted.Inpractice,firmsalwayshave anincentivetoapplyforpublicR&Dsupportduetorelativelylow applicationcosts,eveniftheexpectednetreturnoftheprojectis positiveandalthoughtheR&Dprojectscouldbeconductedwith ownfinancialmeans.Oncetheapplicationwassuccessful,firms canusethepublicgranttoreplaceprivatewithpublicinvestment.

Thisiscalleda“crowdingout”effect(e.g.,Davidetal.,2000).If themajorityoffirmsactedthisway,publicR&Dsubsidieswould leadtheeconomytoalowergrowthpathinthelong-term.The likelihoodofcrowdingoutmaybeparticularlyhighduringreces- sionperiodsasfirmsfacedecliningsalesandfinancialmarketsthat hamperthefinancingofR&D.Firmsmayusetheadditionalrisk-free moneytoserviceshort-termdebtortomaintaintheirproduction capacities.

A vast empirical literature has investigated the question whetherR&Dsubsidyprogramsleadtoacrowdingouteffector stimulateR&Dactivitiesintheprivatesector.Themajorityofthe studiesfindthatR&Dsubsidiesleadtoanadditionalityeffect(see Zú ñiga-Vicenteetal.,2014,forarecentsurvey).Theearlyliterature uptotheyear2000–assurveyedbyDavidetal.(2000)andKlette etal.(2000)–iscriticizedfordisregardingapotentialselectionbias offirmsintoR&Dsubsidyprograms.Ontheonehand,companies withlargerR&DcapacityaremorelikelytoapplyforR&Dsubsi- dies.Ontheotherhand,thesecompaniesmaybemorelikelyto receivethepublicfundsifthegovernmentwantstomaximizethe returnstothesubsidyprogram.Asimplecomparisonofsubsidized andnon-subsidizedfirmshenceleadstobiasedresults.

Themorerecentliteraturewithfocusontheﬁrmlevelassur- veyedbyCerulliandPoti(2010)andZú ˜niga-Vicenteetal.(2014)

2AccordingtotheGermanR&Dstatistics,R&Dpersonnelisthelargestcostunit inaﬁrm’sR&Dprocess(Stifterverband,2013).

(3)

takestheselectionproblemintoaccount.Prominentmethodsthat havebeenusedtoaccountforsampleselectionarematchingmeth- ods(e.g.,CzarnitzkiandFier,2002;AlmusandCzarnitzki,2003;

Czarnitzkietal.,2007;forheterogeneoustreatments),instrumental variablesmethods(e.g.,Wallsten,2000),andselectionmodels(e.g., Busom,2000;Hussinger,2008).³Mostofthestudiesthataccount forselectionissuesreportapositiveeffectofthesubsidyonthe subsidizedﬁrms’R&Dexpensesaswell.⁴

With focus on Germany, prior literature shows exclusively positiveaverageeffects ofR&Dsubsidies.For thedirectproject subsidyprogramoftheBMBF,whichisalsosubjecttoourstudy, CzarnitzkiandFier(2002),AlmusandCzarnitzki(2003),Czarnitzki andHussinger(2004),Czarnitzkietal.(2007),Hussinger(2008), andAerts andSchmidt (2008)report thatthesubsidyleadson averagetoadditionalR&Dexpenses.

2.2. Privateinnovationexpenditureduringeconomicdownturns Theeconomicliteraturehasdevelopeddifferentviewsonthe impactofaneconomicdownturnoninnovationactivities.Oneline ofresearchadvocatescountercyclicalbehaviorofR&Dinvestment.

Intimesofaneconomicdownturn,proﬁtabilitydeclinesencour- ageﬁrmstoseekformeasurestoimproveproductivity.Atthesame time,opportunitycostsofreallocatingproductiveassetsfromman- ufacturingtoR&Darerelativelylowbecauseofalimiteddemand forgoodsandservices(Stiglitz,1993;AghionandSaint-Paul,1998) whichisinlinewiththeSchumpeteriannotionofcreativedestruc- tion.

The contrary perspective suggests that innovation behavior is procyclical. Innovation strongly depends ondemand so that there is noincentive tointroduce newproducts intothe mar- ketifdemandislow(Schmookler,1966;Shleifer,1986).Further, R&Disoftenfinancedbythefirm’sfreecashflowthat depends onthecompany’scurrentprofitsothatfinancialconstraintsdur- inganeconomicdownturnreduceinvestmentinR&D(Hall,1992;

HimmelbergandPetersen,1994;Harhoff,1998;RaffertyandFunk, 2008).

2.3. Privateinnovationexpenditureduringthemostrecent economiccrisis

The mostrecent crisiscaused an overall decline of innova- tionactivitiesacrossOECDcountries,wherebydifferentcountries havebeenaffectedtoadifferentdegree(OECD,2012b;Makkonen, 2013).SomeOECDeconomiesshowsigniﬁcantlyreducedR&Dﬁg- uresfortheprivatesectorduringthecrisis,amongthemCanada, the Czech Republic and the Netherlands. Others show slightly decreased R&D expenditures during the crisis such as Austria, Belgium,andGermany.InGermany,businessenterpriseexpen- ditureonR&D (BERD) decreased by 2.9% in 2009compared to 2008.Apossiblereasonisthat thefederalgovernmentadopted acountercyclicalpolicyand increaseditsgovernmentspending (governmentbudgetappropriationsoroutlaysforR&D,GBAORD) by9%between2007and2009(OECD,2012a;Makkonen,2013).

OthereconomiesincludingChina,EstoniaandHungaryshowan upwardstrendofprivateR&Dexpenditurethroughoutthecrisis (OECD,2012b).

3 Gonzalezetal.(2005)andTakaloetal.(2008)presentstructuralmodelstoaccess theeffectofsubsidiesonthesubsidizedﬁrms.

4 ExceptionsareBusom(2000),whofindsapartialcrowdingouteffectforSpain, andWallsten(2000),whoreportsasubstitutiveeffectofsubsidiesfortheU.S.SBIR program.Gelabertetal.(2009)finddifferencesintheeffectivenessofpublicsubsi- diesdependingonthelevelofappropriationinthefirm’sindustry.

Heterogeneousresponsestothecrisishavealsobeenobserved at the firm level. Overall, European firms’ innovationactivities declined during the crisis with a few exceptions comprising somenew,fastgrowingfirmsandsomehighlyinnovativefirms thatsustainedhighinnovationperformance throughoutthecri- sis(Archibugiet al.,2013a).In addition,a few smallfirms and newentrantsshowgreaterreadinessto“swimagainstthestream”

withregardstotheirinnovationstrategyafterthecrisis(Archibugi etal.,2013b).ForLatinAmerica,Paunov(2012)showsthatmany firmsstoppedongoingR&Dprojectsduringthecrisis.Withfocus onpublicR&Dsubsidies,Paunov(2012)findsthatthelikelihoodto stopprojectscorrelatesnegativelywiththereceiptofpublicfund- ingandconcludesthatpublicfundingschemesareanimportant meanstofostercountercyclicalinvestmentbehavior.Apotential selection biasof firms into subsidyprograms is nottaken into account.

Based on a survey, Kulicke et al. (2010) conclude that the majority of companiesin Germany would have been forcedto stop or postpone their R&D project during the crisis as well.

Rammer (2011) reports that the innovation expenditures of German ﬁrms in the research-intensive manufacturing sector decreasedby9.5%andby16.7%inothermanufacturingindustries (seeRammer,2011;Table2).⁵Especiallysmallﬁrmswereaffected.

About2000 SMEsstopped theirinnovationactivities (Rammer, 2011).

ArecentstudyinvestigatesthemacroeconomiceffectsofR&D subsidiesgrantedbythe“CentralInnovationProgramforSMEs”

(ZIM)programthatwaslaunchedin2008(Brautzschetal.,2015).

Brautzschetal.(2015)concludethattheZIMprogramstabilized production,valueaddedandemploymentduringthecrisiswhich preventedaGDPdeclineof0.5%.Ourstudycomplementstheirﬁnd- ingsbyprovidingmicroeconomicevidencefortheeffectsofR&D subsidiesonSMEs’R&Dinvestmentduringthecrisis.Incontrastto Brautzschetal.(2015),ourdataalsoallowsustocomparethecrisis behaviorofSMEswiththepre-crisisperiod.

3. Empiricalstrategy

Ourempiricalapproachhasthreedifferentparts.Theﬁrstpart evaluatestheeffectivenessofR&Dsubsidiesfortheentiresample period.Thesecondpartcomparestheeffectofsubsidiesinthecrisis yeartopre-andpost-crisisyears.Inthelastpartoftheanalysis,we investigatepossiblereasonsfora differenteffectofsubsidiesin crisisandnon-crisisyears.

3.1. EffectivenessofR&Dsubsidies

Theaimofourpolicyevaluationistoassesstheaverageeffect ofthepublicsubsidyontheR&DspendingofthesubsidizedSMEs.

Sinceasimplecomparisonbetweentreated(subsidized)andnon- treated(non-subsidized)SMEsislikelytobebiasedduetoselection problems,⁶ourempiricalapproachaimsatinvestigatingwhatﬁrms inthetreatedconditionwouldhavespentonR&Diftheyhadnot receivedthesubsidy,i.e.,thecounterfactualsituation.Thisaverage treatmenteffectonthetreated(ATT)isdeﬁnedasfollows:

ATT =E(Y₁−Y₀|S= 1)=E(Y₁|S= 1)−E(Y₀|S= 1) (1) whereY₁istheR&DspendingoftheSMEshavingreceivedthetreat- ment(=subsidy)andY0istheR&DspendingofSMEsthatdidnot receiveasubsidyandSdepictstheactualtreatmentstatus.This

5Rammer’s(2011)calculationsarebasedonthesamesurveythatweuseforthe empiricalanalysis.

6Foradetaileddiscussionoftheselectionproblemseee.g.,BlundellandCostas Dias(2000).

(4)

equationindicatesamissingdataproblem:whilewecanobserve (Y1|S=1),i.e.,theR&Dspending ofasubsidizedﬁrm,wecannot observewhatthesubsidizedﬁrmwouldhavespentonR&Dwithout thesubsidy(Y₀|S=1)

Constructing a valid proxy for the counterfactual situa- tion is the main issue in empirical policy program evaluation.

Econometrictechniquesservingtoovercomethisproblemcom- prisedifference-in-difference(DID)estimations,controlfunction approaches(selectionmodels),instrumentalvariable(IV)estimations,andmatchingtechniques.Duetohighlyunbalancedpanel dataandalackofreliableinstrumentsandexclusionrestrictions, wechooseamatchingapproach.Theintuitionbehindthematching approachistoproxythecounterfactualsituation,i.e.,theinvest- mentofatreatedcompanyintheabsenceofthetreatment,bythe investmentofthemostsimilarnon-treatedobservation.Functional formassumptionsanddistributionalassumptionsabouttheerror termsarenotrequired.Thedisadvantageofthematchingapproach is that it only accounts for selection based on observable ﬁrm characteristics.Priorstudiesfocusingontheeffectivenessofour subsidyprograminGermany,however,showcomparableresults independentofthemethodthatwasemployed(seee.g.,Almusand Czarnitzki,2003;Czarnitzkietal.,2007;Hussinger,2008).

Weapplyanearestneighborpropensityscorematching.This meansthatwematcheachsubsidyrecipientwiththesinglemost similarSMEinthecontrolgroupofthenon-subsidizedSMEs.The pairsarechosenbasedonthesimilarityintheestimatedprobabil- ityofreceivingasubsidy,i.e.,thepropensityscore.Matchingon thepropensityscoreavoidsa“curseofdimensionality”becauseall informationisbundledinthepropensityscorewhichisthenused asthesinglematchingargument(RosenbaumandRubin,1983).

In addition,we requirethat theselectedcontrol observationis observedinthesameyearasthetreatedobservation.Thisiscrucial forouranalysisbecauseweareinterestedincomparingtreatment effectsacrossyears.Wefurtherdemandthatthecontrolobserva- tionsarelocatedwithinthesamegeographicalareainGermany bydistinguishingbetweenEastandWestGermancompanies.We applythisdistinctionbecausethefundinglikelihoodaswellasthe infrastructureforinnovationdiffersbetweenthetworegions.

Thematchingestimator’smaindisadvantageisitsrelianceon theconditionalindependenceassumption(CIA).Thismeansthat theassignmenttotreatmentisrequiredtobeindependentofthe outcomes,inourcasetheR&Dinvestment,conditionalonasetof observablecharacteristics(Rubin,1977).TheCIAissatisfiedifall informationaffectingthetreatmentassignmentandtheoutcomeis includedinthesetofobservablecharacteristics.Ifso,theobserved non-treatedoutcomeE(Y₀|S = 0)isavalidproxyfortheunob- servablecounterfactualoutcomeE(Y0|S=1).Unfortunately,itisnot possibletoformallytesttheCIA.However,weareconfidentthat ourrichsetofcontrolvariablessuffices.⁷ Afurtherrequirement ofthematchingmethodis thatthere hastobesufficient over- lapbetweenthetreatedandthecontrolgroupintermsoftheir propensitytoreceiveapublicsubsidy(commonsupport).Inorder toguaranteecommonsupport,wecalculatetheminimumandthe maximumofthepropensityscoresofthepotentialcontrolgroup, anddeleteobservationsontreatedfirmswithprobabilitieslarger thanthemaximumandsmallerthantheminimuminthepotential controlgroup.IftheCIAandthecommonsupportarefulfilled,the

7Asimilarsetofcontrolvariableshasbeenusedinavarietyofstudiesthateval- uatetheeffectsofR&Dsubsidiesbasedonsimilardatasetsemployingamatching approach(e.g.,CzarnitzkiandFier,2002;AlmusandCzarnitzki,2003;Czarnitzki etal.,2007;CzarnitzkiandLopesBento,2013).

ATTwillbeidentiﬁedandconsistentlyestimatedbythefollowing equation^8,9:

ATT =(Y₁|X=x,S= 1)−E(Y₀|X=x,S=0) (2)

3.2. EffectivenessofR&Dsubsidiesincrisisandnon-crisisyears AftertheidentiﬁcationoftheATT,weinvestigatewhetherthe effectofsubsidiesdiffersinthecrisisandnon-crisisyearsbyrun- ninganOLSregressionoftheATTonasetoftimedummiesd.

ˆATTt=∝+ ˙

n=2007−2010

ˇndnt+ut (3)

Therearetwodifferentpossiblescenarios.Ontheonehand,one canexpectthatthesubsidiesaremoreeffectiveinthecrisisyear becausefirmsmayfacemoreseverefinancialconstraintssothat thesubsidyincreasesR&Dinvestmentsubstantiallyascomparedto thecounterfactualsituation.Ontheotherhand,alowertreatment effectcanoccurinthecrisisyeariffirmsmatchthepublicfunds withlessprivateinvestmentsthantheywouldhavemadeinnon- crisisyearsorreplaceprivateinvestmentsbypublicfunds.

3.2.1. ExplanationsforpossibledifferenteffectsofR&Dsubsidies incrisisandnon-crisisyears

Intheﬁnalpartoftheanalysis,weinvestigatepossibleexpla- nations forthepotentialdifferencesoftheeffectivenessofR&D subsidiesin crisisandnon-crisisyears. Such differencescan be motivatedby(a)analteredfundingpolicyin crisistimesor(b) achangeofbehaviorofgrantrecipientsincrisistimes:

(a)Duringthepastcrisis,directprojectfundinghasbeenincreased intermsofgrantedamountsperprojectandnumberofprojects funded.ThiscanreducetheATTbecauseifmoreprojectsare funded theaverage“quality”oftherecipientsislikelytobe lowerthaninnon-crisisyears.

(b)Subsidyrecipientsmayfacetighterbudgetconstraintsduring thecrisis.Inresponse,theymightinvestlessintothesubsidized R&Dprojectsthantheywouldhaveinvestedinnon-crisisyears.

Inorder toanalyzewhetherthechangein innovationpolicy affectedthe effectof R&D subsidiesduring thecrisis,we com- parefirst-timesubsidizedfirmsincrisisandnon-crisisyears.Ifa lowersubsidyeffectwouldbecausedbyalower“quality”offunded firmsduringthecrisisyears,weshouldfindasignificantdifference betweenthesegroupsoffirmsintermsofsuccesspredictorslike firmsizeorpatentstock.

InordertoinvestigatewhethercompanieschangedtheirR&D investmentbehavior,werepeattheanalysisdescribedintheSec- tion3.2forthesubsampleofSMEsthatreceivedsubsidiesbefore, duringandafterthecrisis.Ifwewouldﬁndthatthesecompanies reduce theirR&D investment we couldconclude that a poten- tialdecreased effectivenessofR&Dsubsidiesiscausedbyﬁrms’

reluctancetoinvestinR&Dduringthecrisisrather thanbythe countercyclicalinnovationpolicy.

8Thedetailsofourmatchingprocedure(matchingprotocol)areprovidedupon request.Very similarmatchingprotocolscanbe founde.g.,inCzarnitzkiand Hussinger(2004),AertsandSchmidt(2008)andCzarnitzkiandLopesBento(2013).

9AsimilarmatchingapproachhasbeenusedamongothersbyCzarnitzkiand Fier(2002),AlmusandCzarnitzki(2003),CzarnitzkiandHussinger(2004),Goerg andStrobl(2007),GonzalezandPazo(2008),CzarnitzkiandLopesBento(2013)and Hottenrottetal.(2014).

(5)

4. Data,variablesanddescriptivestatistics 4.1. Data

Forourempiricalanalysisofapotentialadditionalityeffectof R&Dsubsidies,weconstructa databasewhichconsistsofﬁrm- levelinformationandtheirsubsidyrecords.Theﬁrm-leveldatais providedbytheMannheim InnovationPanel(MIP),which isan annualsurvey conductedbytheCentre forEuropeanEconomic Research (ZEW) on behalf of the German Federal Ministry for EducationResearch (BMBF)since 1993.The MIPis theGerman contributiontotheEuropeanCommission’sCommunityInnova- tionSurvey(CIS)andis basedonthemethodologyproposedin theOsloManualforcollectinginnovationdata(OECDandEurostat, 2005).¹⁰

Informationonthe Federal Government’s projectfunding is takenfromBMBF’sPROFIdatabase,whichisavailabletousfrom 1992onwards.It contains information onallnon-military R&D projectsfunded bytheBMBF.TheBMBFprogram isthelargest sourceofpublicR&DfundsforthebusinesssectorinGermanyand accountsformorethan80%ofthetotalpublicR&Dfundingavail- abletothebusinesssector.Thedirectprojectfundingprogramis opentoallﬁrmslocatedinGermany.Theofﬁcialapplicationform requiresdetailedinformationonthecompanyanditsplannedR&D projects.Thereisapeerreviewprocessaccordingtowhichgrants areassignedas“matchinggrants”totheselectedprojects,which meansthatapplicantsthemselveshavetocontributeatleast50%to thesubsidizedprojects.Thegovernmentsponsorsatmost50%asis prescribedinthefundingguidelinesoftheEuropeanCommission (1996)andinGermanregulations(BMBFandBMWi,2001).

FurtherdatasourcescomprisetheEuropeanPatentOffice(EPO) providinguswithfirms’patentapplicationssince1979,thecredit rating agencyCreditreform and theFederal Statistical Office of Germany.Thatlatterofficeprovidesuswithameasureforthebusi- nesscycle.We constructedtheannual changeofindustry-level salesbased onthe4-digitindustryleveldata(NACE 2.0).¹¹ The databaseofCreditreformprovidesuswithfirmageinformationand acreditratingindicatorthatproxiesthefirms’financialfitness.

Thefinalsamplecoverstheyears2006–2010inordertocover thepre-crisisperiod(2006–2008),thecrisisyear(2009)andthe firstpost-crisisyear(2010).Thechoiceofthetimeframeforour studyismotivatedbythefactthatwewanttoexcludetherecession 2002/2003and thefollowing recoveringperiod. Contrarytothe mostrecenteconomiccrisis,therecession2002/2003wasexpected byGermancompanies(Rammeretal.,2003)sothatwewouldnot beabletoidentifycausaleffects.Withregardstotherecenteco- nomiccrisis,Rammer(2011)showsthatintheyears2006–2008the actualinnovationexpendituresofGermanfirmsofallsizeclasses werealmostequaltotheinnovationexpendituresplannedinthe previousyear.Onlyinthecrisisyear2009,agapbetweenplanned andactualinnovationexpensesarises(seeFig.1inRammer,2011).

Werestrictoursampletoﬁrmswithlessthan250employees sinceSMEsaremostsensitivetothebusinesscycle(OECD,2009).

Our sampleincludes manufacturingaswellas business related servicesectors.

Theﬁnalsampleconsistsof10,527ﬁrm-yearobservationsout ofwhich1016receivedanR&DsubsidyfromtheBMBF.

10 ForfurtherdetailsseeAschhoffetal.(2013)andPetersandRammer(2013).

11 Forafewcaseswehadtousethe3-digitlevel.

Table1

Overviewofsubsidizedandnon-subsidizedﬁrmsovertime.

2006 2007 2008 2009 2010 Total Notsubsidized 1,579 1,715 2,358 1,925 1,934 9,511

Subsidized 191 176 153 194 302 1,016

First-timesubsidizedafter2008 6 5 8 15 93 127 Note:Duetothepanelstructureofourdataset,observationsofﬁrmssubsidizedfor theﬁrsttimeafter2008canalsoappearin2006–2008.

4.2. Variables

4.2.1. Treatmentvariable

Wemeasuretreatmentwithabinaryindicatorthattakesonthe value1ifafirmhadbeensubsidizedbytheBMBFintherespective year.Theindicatortakesonthevalue0ifafirmhadnotreceived anyR&Dsubsidyatallintherespectiveyear,neitherfromtheEU norfromtheFederalGovernmentnorfromothersources.Thus, ourcontrolgroupsolelyconsistsofnon-subsidizedfirms,allowing ustoruleoutside-effectsfromothersubsidyprograms.Table1 showsthedistributionofsubsidizedcompaniesandunsubsidized companiesinoursampleovertime.Itbecomesevidentthatthe numberofsubsidizedfirmshasincreasedduringandafterthecrisis.

Thenumberofnewlyfundedcompaniesisstillsmallinthecrisis year2009,butincreasessigniﬁcantlyin2010.

4.2.2. Outcomevariables

Wetestpotentialadditionalityeffectsforsixoutcomevariables inordertoshowrobustnessoftheresultswithregardstodifferent definitionsofthedependentvariable.RDdepictsafirm’stotalR&D expenditure,whichismeasuredinmillionEUR.PRIVRDisdefined astheprivateR&Dinvestment,i.e.,RDminusthesubsidyreceived.

Sincethesevariablesaredistributedaskew,weemployRDINT(RD oversales)aswellasPRIVRDINT(PRIVRDoversales)inaddition.

Furthermore,wedeﬁneRDEMP (RDover numberofemployees) andPRIVRDEMP(PRIVRDovernumberofemployees)asalternative measuresfortheR&Dintensity.

4.2.3. Controlvariables

Ourcontrolvariablesencompassfirmsizeasmeasuredbythe logofthenumberoffulltimeemployees,Lemp.Weexpectthat R&Dexpenditurecorrelateswithfirmsize.Thus,largerfirmsare morelikelytoapplyforsubsidiesandtoreceiveagrantifthegov- ernmentwantstomaximizethelikelihoodofapositiveoutcome ofthefundedprojectbychoosingcompanieswithasuperiorinno- vationcapacity.Thelogarithmicspecificationischosenbecauseof theskewdistributionofthefirmsizevariable.

Ifafirmispartofanenterprisegroup,this membershipcan improvetheaccesstoinnovationcapacityandalsotoinformation ongovernmentalprograms.Thismayresultinahigherlikelihood toapplyforasubsidy.Further,governmentalevaluatorscouldbe pronetosubsidizefirmsthatbelongtoanetworkoffirms,being aware of potential knowledge spillovers within the enterprise groupduetothesubsidizedproject.Wecontrolforfirmsbelonging toafirmgroupwithabinaryvariable,Group.Firmgroupswithafor- eignheadquarter,Foreign,could,incontrast,belesslikelytoreceive fundingifthegovernmentwouldwanttoinduceeconomiceffects fortheowncountry. ThebinaryvariableEastindicateswhether afirmislocatedinEasternGermanyorintheWesternpartofthe country.EastGermanfirmscouldbemorelikelytoreceiveasubsidy asthisregionisstillinacatch-upprocesswithregardstoWest- ernGermany.Thelogoffirmage,Lage,coverspotentialfirmage effects.Firmscompetinginforeignmarketsaremoreinnovative thanothers(ArnoldandHussinger,2005).Therefore,wealsoexpect export-orientedfirmstoapplymorefrequentlyforR&Dsubsidies.

OurbinarydummyExportindicateswhetheraﬁrmhasexportsales

(6)

ornot.Furthermore,weaccountforafirm’sinnovationpotentialby pastinnovationsuccessintermsofthecompany’spatentstock.To constructthepatentstock,weusepatentapplicationsfrom1979 onwardswhichhavebeenfiledattheEPO.Thepatentstockintis calculatedasadepreciatedsumofallthesepatentapplicationsuntil t-1plusthe(non-depreciated)patentapplicationsint.Thedepre- ciationrateissetto0.15,asiscommonintheliterature(seee.g., Hall,1990;GrilichesandMairesse,1984).Duetocollinearitycon- cernswithfirmsize,thepatentstockisnormalizedbythenumber ofemployees,Patemp.

Toaccountforpotentialfinancialrestrictions,particularlydur- ingthecrisisperiod,weincludeCreditreform’screditratingindex, Credit.¹²Thisisanindexrepresentingafirm’ssolvency.Theindex rangesfrom100to600.Thelargertheindexvalue,theloweristhe creditratingandtheabilitytoattractdebtcapital.Firmsthathave moreproblemstoattractexternalfinancemightbemorelikelyto applyforsubsidies.

We alsocontrol forthereceipt of pastsubsidies.Firms that receivedsubsidiesinthepasthaveahighchancethatanewapplica- tionwillbeevaluatedpositively(Hussinger,2008;Aschhoff,2010).

Weuseadummyforpastsubsidyreceiptwithafouryearlagto accountforthefactthattheaveragesubsidizedR&Dprojectlasts forabout3years.

Anothercharacteristictobeconsideredisthebusinessclimate ofthefirm’sindustrysector.SMEsusuallyparticipateonlyinoneor afewproductmarkets.Incaseofeconomicdownturns,thesefirms maynothavetheopportunitytocompensateaseriousdecrease ofdemandinoneoftheirfewmarkets.Wecontrolforthecompa- nies’businessenvironmentbyincludinganindustry-specificsales growthrate,Salesgrowth.

Toavoidpotentialendogeneity,welagalltime-variantexplana- toryvariablesandconsiderGroup,Foreign,Eastastime-invariant andSalesgrowthandLageasexogenoustotheﬁrm.

4.3. Descriptivestatistics

Table2shows descriptivestatistics,comparingthevariables’

meanvaluesofnon-subsidizedandsubsidizedfirms.Significantt- testsindicatesystematicdifferencesbetweenthemeanvaluesof thevariablesforsubsidizedandnon-subsidizedfirms.Forexam- ple,subsidizedfirmsscorehigheroneachR&Dinputmeasure.They havemorepatentsperemployeeandmoreemployeesingeneral, andaremorelikelytobeexporters.Further,subsidizedSMEsare youngerandaremorefrequentlylocatedinEasternGermany.Sub- sidizedSMEs,onaverage,reveal ahigherbusiness indexwhich describesthattheindustriesofthesubsidyrecipientshaveexperi- encedalargerupswingascomparedtothepreviousperiodthanthe industriesofnon-subsidizedcompanies.ThissuggeststhatSMEs aremorelikely toapplyforand receivea subsidyiftheindus- tryisexperiencinganupswing.Further,thecreditratingbetween non-subsidizedandsubsidizedfirmsdoesnotdiffersignificantly.

Avalueofabout225meansa“goodﬁnancialstanding.”¹³

12 Wealsoemployamissingvaluecorrection.ThemissingvaluesofCreditareset tozero.Anadditionalbinarydummy,Creditmvd,thattakesonthevalue1ifCredit equalszero,isincludedintheestimations.

13 Similarresultsfortheexogenousvariables appearwhencomparingthem betweenthepre-crisisandthe(post-)crisisperiods.However,almostnosigniﬁcant differencesarefoundfortheoutcomevariables(seeTableA1).

Table2

Descriptivestatisticsforthewholesample,bysubsidystatus.

Unsubsidized ﬁrmsN=9511

Subsidized ﬁrmsN=1016

t-Test

Mean Std.dev. Mean Std.dev.

Covariates

Subt-4 0.006 0.079 0.338 0.473 ***

Patemp 0.001 0.015 0.027 0.085 ***

Lemp 3.165 1.093 3.460 1.091 ***

Foreign 0.047 0.213 0.068 0.252 **

Export 0.404 0.491 0.802 0.399 ***

Group 0.213 0.409 0.247 0.432 **

Lage 3.096 0.858 2.679 0.728 ***

East 0.337 0.473 0.410 0.492 ***

Salesgrowth 2.459 13.208 4.068 13.947 ***

Credit 225.487 65.334 226.054 52.911

Creditmvd 0.035 0.185 0.023 0.149 **

Controllingforperioddifferences

Subt-4period 0.003 0.054 0.148 0.355 ***

Patempperiod 0.000 0.005 0.014 0.066 ***

Lempperiod 1.263 1.680 1.651 1.851 ***

Foreign period 0.017 0.127 0.029 0.167 **

Exportperiod 0.156 0.363 0.390 0.488 ***

Groupperiod 0.080 0.271 0.118 0.323 ***

Lageperiod 1.275 1.632 1.305 1.432

Eastperiod 0.138 0.345 0.172 0.378 ***

Salesgrowthperiod -0.635 10.543 0.919 12.443 ***

Creditperiod 91.563 117.935 111.111 119.564 ***

Credit mvdperiod 0.014 0.119 0.010 0.099

Outcomevariables

RD 0.036 0.310 0.705 2.586 ***

PRIVRD 0.036 0.310 0.577 2.310 ***

RDINT 0.005 0.035 0.150 0.220 ***

PRIVRDINT 0.005 0.035 0.098 0.184 ***

RDEMP 0.001 0.004 0.013 0.023 ***

PRIVRDEMP 0.001 0.004 0.009 0.020 ***

Note:*p<0.1;**p<0.05;***p<0.01.Forreasonsofclarity,wedroppedtheinforma- tiononindustrydummiesandtheirperioddifferencesfromthistable.

5. Empiricalresults

5.1. Fundingpropensity

AsdescribedinSection3,we employamatchingmethodto identifythecausaleffectofthesubsidytreatmentonR&Dinvest- ment.Thus,wehavetoﬁndnon-subsidizedobservationswiththe mostsimilar characteristicsto thesubsidizedobservations. We determinetheso-callednearestneighborsbasedonthepropen- sityscore,i.e.,thelikelihoodofreceivingasubsidy.Inordertoget anestimateforthepropensityscore,weestimateaprobitmodelfor thereceiptofpublicsubsidies.Weallowthatourcontrolvariables haveadifferentimpactonthereceiptofsubsidiesinthepre-crisis periodandthereafterbyincludinginteractiontermsofallcontrol variableswithayeardummyfor2009and2010.Table3showsthe results.

Apartfromthedummyindicatingmembershipofaﬁrmgroup, thedummyindicatingaforeignheadquarterandthebusinesssit- uation,eachvariablerevealssigniﬁcanteffectswiththeexpected sign.

5.2. Averagetreatmenteffectonthetreated

In a secondstep, wedetermine“twin observations” ofnon- subsidizedSMEsfor eachsubsidizedSME observationbased on thepropensityscoreandtheadditionaltwomatchingarguments – locationin EasternGermany andyear ofobservation. Due to thecommonsupportcriterionwehavetodroptwoobservations forwhichwecannotdetermineappropriatecontrolobservations.

Table4showsthemeanvaluesfortreatedandcontrolobservations

(7)

Table3

Probitestimationofsubsidyreceipt.

Coefﬁcients Std.err.

Subt-4 2.187*** (0.118)

Patemp 4.568*** (0.700)

Lemp 0.149*** (0.032)

Foreign 0.050 (0.127)

Export 0.775*** (0.075)

Group −0.243*** (0.084)

Lage −0.273*** (0.042)

East 0.245*** (0.065)

Salesgrowth −0.003 (0.004)

Credit −0.002** (0.001)

Creditmvd −0.686*** (0.243)

Industries

Mining −1.130*** (0.294)

Low-techmanufacturing −0.984*** (0.098)

High-techmanufacturing −0.423*** (0.099)

Energy,water,recycling −1.012*** (0.163)

Wholesale −1.909*** (0.356)

Transportation −1.389*** (0.202)

ICT −0.171 (0.117)

Consulting,advertising −0.800*** (0.180)

Years

2007 −0.069 (0.074)

2008 −0.293*** (0.074)

2009 0.117 (0.398)

2010 0.328 (0.398)

Constant −0.645** (0.260)

Observations 10,527

Log-likelihood −1968

McFaddenR2 0.411

Timedummies–chi2(4) 24.26***

Industrydummies–chi2(8) 183.6***

Periodinteractions–chi2(19) 77.26***

Note: *p<0.1; **p<0.05; ***p<0.01; see Table A3 for additionally estimated coefﬁcients.

afterthematching.Therenolongerexistsignificantdifferences betweenthetreatedandthenon-treatedobservationswithregards tothecontrolvariablesindicatingthatourmatchingspecification isvalid.Significantdifferencesinthemeanvaluesoftheoutcome variablespersistandcanbegivenacausalinterpretationnowafter thematching.

ThesubsidizedfirmsshowahigherR&Dactivityindependent ofthedefinitionoftheoutcomevariable.Thus,wefindanover- allpositiveATTsignalingthatfirmsincreasedtheirR&Dspending duetothesubsidy.Wecanrejectacrowdingout.TheATTequals 0.370(0.242)millionEURintermsofR&D(privateR&D)expenditures.TheATTintermsofR&D(privateR&D)oversalescorresponds to11.7%(6.6%)points.ForR&DintensityasdefinedbyR&Dover employmenttheATTamountsto0.8%(0.5%)pointsforR&D(private R&D).

Table4 Matchingresults.

Unsubsidized N=998

Subsidized N=998

t-Test p-Values

Mean Mean

Covariates

Subt-4 0.347 0.333 0.720

Patemp 0.013 0.019 0.115

Lemp 3.539 3.461 0.416

Foreign 0.081 0.067 0.528

Export 0.804 0.802 0.951

Group 0.302 0.245 0.134

Lage 2.699 2.681 0.739

East 0.414 0.414 1.000

Salesgrowth 4.117 4.089 0.982

Credit 228.667 226.270 0.649

Credit mvd 0.027 0.022 0.704

Controllingforperioddifferences

Subt-4period 0.152 0.141 0.708

Patemp period 0.006 0.008 0.293

Lempperiod 1.755 1.626 0.428

Foreignperiod 0.041 0.027 0.379

Exportperiod 0.390 0.384 0.881

Groupperiod 0.139 0.114 0.374

Lageperiod 1.291 1.286 0.963

Eastperiod 0.171 0.171 1.000

Salesgrowth period 0.973 0.899 0.949

Creditperiod 105.868 109.789 0.684

Creditmvdperiod 0.014 0.009 0.595

Outcomevariables

RD 0.306 0.676 *** 0.001

PRIVRD 0.306 0.548 ** 0.021

RDINT 0.029 0.146 *** 0.000

PRIVRDINT 0.029 0.095 *** 0.000

RDEMP 0.004 0.012 *** 0.000

PRIVRDEMP 0.004 0.009 *** 0.000

Note:*p<0.1;**p<0.05;***p<0.01.Forreasonsofclarity,wedroppedtheindustry dummiesandtheirperioddifferencesfromthistable.Eachdifferenceofthemean valuesisnotsigniﬁcantlydifferentfromzero.

5.3. Averagetreatmenteffectsinthecourseoftime

Inthissub-section,weinvestigatepotentialchangesoftheaver- agetreatmenteffectonthetreatedovertime.Table5presentsthe resultsofOLSregressionsoftheATTonasetofyeardummies.

Thisspecification’sconstantrepresentstheATToftheyear2006, our benchmark year. Table 5 shows that the ATTs on average arepositiveandsignificantasexpected.RegardingR&Dexpenses andprivateR&Dexpenses,theATTin2009issignificantlylower andevennegative(0.763–1.006<0;0.599–0.942<0)indicatinga crowdingout effect.In 2010,theeffectis stilllower but larger thanzero(0.763–0.522>0;0.599–0.464>0),indicatingasmaller, yetpositiveadditionalityeffect.WithregardstotheR&Dintensity Table5

OLSresultsoftheaveragetreatmenteffectsontimedummies.

RD PRIVRD RDINT PRIVRDINT RDEMP PRIVRDEMP

2007 −0.341 −0.329 0.010 −0.000 −0.000 −0.000

(0.315) (0.268) (0.025) (0.022) (0.003) (0.002)

2008 0.053 0.088 −0.005 −0.004 0.002 0.002

(0.440) (0.397) (0.027) (0.025) (0.003) (0.003)

2009 −1.006*** −0.942*** −0.056** −0.055*** −0.006** −0.006**

(0.330) (0.295) (0.023) (0.020) (0.003) (0.002)

2010 −0.522* −0.464* −0.030 −0.033 −0.003 −0.003

(0.295) (0.254) (0.023) (0.020) (0.002) (0.002)

Constant 0.763*** 0.599** 0.135*** 0.087*** 0.010*** 0.007***

(0.285) (0.245) (0.019) (0.017) (0.002) (0.002)

Observations 998 998 998 998 998 998

R2 0.018 0.019 0.010 0.012 0.013 0.014

Note:*p<0.1;**p<0.05;***p<0.01;robuststandarderrorsinparentheses.

(8)

Fig.2. FundingdevelopmentforSMEsbetween2006and2010.

Source:BMBF’sPROFIdatabase.Owncalculations.

variablesweﬁnda smalleradditionalityeffectinthecrisisyear 2009andnosigniﬁcantchangeforthepost-crisisyear.

Table A2 in the Appendix A shows a robustness check. We presentresultsfrompooledOLSregressionsincludinginteraction termsforthesubsidydummyandtheyeardummies.OLSmodels donotaccountforselectionofSMEsintothesubsidyprogram.The resultsareneverthelessinlinewiththeﬁndingabovesuggesting thatthesubsidieswerelesseffectiveinthecrisisyearandpostcrisis years.

5.4. PotentialexplanationsforthelowerATT

Asmentioned inSection3,thesigniﬁcantlylowerATTcould beduetoachangedfundingpolicyduringthecrisisorduetoan alteredinvestmentbehavioroftheﬁrms.Fig.2depictstheallo- cationpolicyoftheBMBFforSMEsovertheobservedperiod.In 2006,1226projectsweregrantedwithanoverallamountofabout 221millionEUR.Duringthecrisis,thisamountcorrespondedto 435million EURfor 1437granted projects.Thisshowsthatthe BMBFincreasedthenumberofprojectsgrantedaswellasthetotal amountin2009.In2010,pre-crisislevelsareachieved.Thepeakof 2009couldimplythattheBMBFsubsidizedSMEsofloweraverage

“quality”thaninthepre-crisisyearsextendingthepooloffunded companies.

Inorder toprovidea formaltest forpotentialqualitydiffer- encesofsubsidyrecipientsbeforeand duringthecrisis,Table6 presentsaprobitregressionofourcontrolvariablesonabinary variable that takes onthe value 1 if theﬁrm has been subsidized bythe BMBF forthe ﬁrst time beforethecrisis of 2009.

Table6

Probitestimationofﬁrst-timefundedﬁrms.

Coefﬁcients Std.err.

Patemp 0.473 (0.813)

Lemp −0.053 (0.066)

Foreign −0.260 (0.246)

Export 0.177 (0.160)

Group 0.233 (0.157)

Lage −0.026 (0.082)

East 0.133 (0.120)

Salesgrowth −0.002 (0.005)

Credit 0.001 (0.002)

Creditmvd −0.343 (0.496)

– Industries

Low-techmanufacturing −0.421** (0.214)

High-techmanufacturing −0.467** (0.193)

Energy,water,recycling −1.162*** (0.301)

Wholesale −1.217* (0.652)

Transportation −0.185 (0.433)

ICT −0.352 (0.232)

Consulting,advertising −0.450 (0.297)

Years

2007 0.083 (0.274)

2008 −0.168 (0.253)

2009 −0.519** (0.249)

2010 −1.379*** (0.201)

Constant 2.213*** (0.572)

Observations 1,175

Log-likelihood −320.3

McFaddenR2 0.204

Industrydummies–chi2(7) 17.89**

Timedummies–chi2(4) 99.04***

Note:*p<0.1;**p<0.05;***p<0.01.

It takes on thevalue 0 if theﬁrm has beensubsidized by the BMBFfor theﬁrsttime afterthecrisisstarted(i.e.,after2008).

The sampleonly includesfirms thathave receivedsubsidiesat somepointintheirlifetime(1992–2010).Ifafirst-timesubsidy recipientbeforeandduringthe(post-)crisisperiodsdifferedin termsofsuccessindicators,wewouldseesignificantdifferences inthesefirmcharacteristics.Theregressionresults,however,do not uncover systematic differences between first-time subsidy recipients in both periods indicating that the “quality” of the subsidyrecipientsdidnotdiffer.Theonlysignificantdifferences are foundfor someindustry and time dummies.Therefore, we rejectthat the crowding out effectduring the crisisyear 2009 andtheloweradditionallyeffectin2010arecausedbysystem-

Table7

OLSresultsoftheaveragetreatmenteffectsontimedummies–subsampleofcompaniesthatreceivedR&Dsubsidiesbefore,duringandafterthecrisis.

RD PRIVRD RDINT PRIVRDINT RDEMP PRIVRDEMP

2007 −0.341 −0.329 0.010 −0.000 −0.000 −0.000

(0.315) (0.268) (0.025) (0.022) (0.003) (0.002)

2008 0.053 0.088 −0.005 −0.004 0.002 0.002

(0.441) (0.397) (0.027) (0.025) (0.003) (0.003)

2009 −1.070*** −1.008*** −0.056** −0.057*** −0.006** −0.006***

(0.334) (0.299) (0.024) (0.021) (0.003) (0.002)

2010 −0.502* −0.467* −0.028 −0.033 −0.003 −0.003

(0.303) (0.262) (0.025) (0.022) (0.002) (0.002)

Constant 0.763*** 0.599** 0.135*** 0.087*** 0.010*** 0.007***

(0.286) (0.245) (0.019) (0.017) (0.002) (0.002)

Observations 896 896 896 896 896 896

R2 0.019 0.021 0.010 0.012 0.013 0.015

Note:*p<0.1;**p<0.05;***p<0.01;robuststandarderrorsinparentheses.

(9)

aticdifferencesbetweensubsidyrecipientsbeforethecrisisand afterwards.¹⁴

Thealternativeexplanationforthecrowdingouteffectisthat itisinducedbyfirms’alteredinvestmentbehavior.Firmshadto copewiththenegativeconsequencesoftheeconomiccrisis.They mayhaveallocatedthefundsthattheywouldhavespentonR&D projectstobusinessareasthattheydeemedmoreimportantdur- ingthecrisis.AccordingtoRammer(2011),firmsthatweremost severelyaffectedbythecrisishavebeenR&Dactivefirms.Wetest whetherfirms’changedtheirR&Dinvestmentbehaviorduringthe crisisbyfocusingonthesubsampleofSMEsthatreceivedsubsidies beforeandduringthecrisis.Werepeattheregressionspresented inTable5.

Table7showstheresults.Itappearsthatweﬁndacrowdingout effectforthecrisisyear2009andasmaller,butagainpositivetreat- menteffectforthepostcrisisyear2010forR&Dinvestment.Thisis inlinewiththepresumptionthatSMEschangedtheirinvestment behaviorduringthecrisisratherthanthealteredinnovationpolicy causingthereducedtreatmenteffectin2009.

6. Conclusion

OurstudyexaminestheeffectsofR&DsubsidiesonSMEs’R&D spendingduringthemostrecenteconomiccrisis.Westartbyinves- tigatingtheaveragetreatmenteffectonthetreated(ATT)ofdirect R&DsubsidiesinGermanyduringtheperiod 2006–2010.Using propensityscorenearestneighbor matching,we findan overall positiveATTfortheobservedperiod.Furtherresultsshowthatthe ATTwasnegativeinthecrisisyear2009indicatingacrowdingout effect.Inthefirstpost-crisisyear,however,wefindagainevidence forapositive,butstillsmalleffectofR&Dsubsidiesonthesubsi- dizedSMEs’R&DinvestmentbehaviorsignalingthatGermanSMEs recoveredfastafterthecrisis’peak.

Insearchforanexplanationforthecrowdingouteffectin2009, weinvestigatetwopotentialcauses.Weanalyzewhetherthecoun- tercyclicalinnovationpolicyduringthecrisisisresponsibleforthe crowdingouteffectorwhetheradifferentinvestmentbehaviorof subsidyrecipientsinthecrisisyearwasthecause.Theexpansion ofthesubsidyprogramduringthecrisiscouldhavedecreasedthe average“quality”ofthesubsidyrecipientsascomparedtothepre- crisisyears.This,inturn,couldhaveloweredtheaveragesuccess ofthesubsidyprogram.Inordertoprovideaformaltest,wecom- parefirst-timesubsidyrecipientsbeforeandduringthecrisis.The resultsshowthattherearenosystematicdifferencesbetweenthese groupsoffirmsintermsofinnovationsuccessindicators.Inorderto investigatewhetherSMEs’investmentbehaviorhaschangeddur- ingthecrisis,wefocusonthesubsampleofcompaniesthatreceived subsidiesbefore,duringandafterthecrisisandfindthatthesecom- paniessignificantlyreducedtheirR&Dexpensesin2009.Hence,we concludethatthecrowdingouteffectisnotcausedbytheR&Dsub- sidyprogramexpansioninthecrisisyearsbutthatitisduetofirm behavior.Inordertocopewithnegativeeffectsofthecrisis,SMEs seemtohaveshiftedfundsduringthecrisisthattheywouldhave spentonR&Dprojectsinnon-crisisyears,tootherbusinessareas, suchaskeepingtheirstockofemployees.

ThecountercyclicalinnovationpolicyoftheBMBFislikelyto havehad a stabilizing effectfor SMEshelpingthem topaythe wagesofR&Dworkersandtostartnewprojects.Thecrowdingout effectlastedonlyforoneyearafterwhichSMEsrecoveredfromthe

14 Additionally,wecomparefirmsthatreceivedsubsidiesbeforethecrisiswith firmsthatreceivedsubsidiesafterthecrisisstartedintermsofothereconomic performancecharacteristics,likelaborproductivityandinnovationsales,aswas suggestedbyareviewer,butdonotfindasignificantdifferencethereeither.The resultsarepresentedinTableA4intheAppendixA.

ﬁrstshockofthecrisisandreturnedtotheirpre-crisisR&Dinvest- mentbehavior.Thelikelihoodofnegativelongruneffectsforthe economyishencelow.

Alimitationofourstudyis thatwe canonlyfocusonaddi- tionality effects of R&Dsubsidies onR&D inputrather than on additionalityeffectsintermsofR&Doutput,asecondmeasureof highinterestforpolicymakers.¹⁵Whileitisfeasibletofocuson inputadditionalitiesofR&Dfundingsincetheyoccurintheimme- diatefundingyears,outputeffectsaredistributedoveraperiodof yearsafterthefundingwasreceiveddependingonthecomplex- ityoftheproject.Thismeansthattheoutputeffectsofpre-crisis fundingarevisibleinthepre-crisisperiodaswellasinthecrisis andinpost-crisisyears.Sincewecannotlinkparticularinnovation projectstotheirsuccess,acomparisonofoutputeffectsofprojects fundedbeforeandduringthecrisisisnotpossiblewiththedataat hand.

ItisalsoworthmentioningthatGermanyisoneofthebiggest R&D spenders in theworld. It is heavilyspecialized in manu- facturingindustriesandhasahighproportionofbusinessR&D.

Germany’sR&Dpolicyduringthecrisismaynot workforother countrieswherebusinessR&Dasastockislowerandmorevolatile.

Itwouldbeofgreatpolicyinteresttocomparedifferentinnovation policiesduringthecrisisandtheireffectivenessforasetofcountries withdifferentcharacteristics.

Furthermore,itwouldbeofhighinteresttotesttheeffectiveness ofresearchversusdevelopmentsubsidiesduringthepasteconomic crisis.Unfortunately,ourdatasetisnotdetailedenoughtoprovide suchaninvestigation,butothersurveydatasuchastheFlanders R&Dsurveywouldallowforsuchananalysis(seeHottenrottetal., 2014).Anotherimportantfactorthatwecannottakeintoaccount duetodatalimitationsistheeffectofcollaborationontheeffec- tivenessofR&Dsubsidiesduringthecrisis(Czarnitzkietal.,2007).

Thisisanotherimportantavenueforfutureresearch.

Lastly, the limitations of the matching approach should be acknowledged. Most important is the fact that the matching methodonlyaccountsforselectionbasedonobservablecharacter- istics.Inlightofthepreviousliteraturethatﬁndssimilarresultsfor thedirectR&DsubsidyprograminGermanyindependentofthe methodthathasbeenemployed(seee.g.,AlmusandCzarnitzki, 2003;Czarnitzkietal.,2007;andHussinger,2008),webelievethat thisisaminorissueinourparticularcontext.

Acknowledgements

WethankBenjaminBalsmeier,MartinCarree,DirkCzarnitzki, ClemensFuest,GeorgLicht,JeongsikLee,CindyLopesBento,Sadao Nagaoka,HiroyukiOkamuro,BettinaPetersandChristianRammer forhelpfulcommentsandPiaNeuerforproofreading.Furthermore, wewouldliketothanktheparticipantsoftheCISSSummerSchool 2013,thePhDWorkshop2014jointlyorganizedbyKULeuven,Uni- versityofTurin,CopenhagenBusinessSchoolandZEW,theDRUID AcademyConference2014,theHitotsubashiUniversity’sIOLabour Workshop2014,theEARIE2014,theAPIC2014andtheParisSchool ofEconomics’IOWorkshop.Furthermore,wewanttothankthree anonymousreviewersfortheirvaluablecomments.

AppendixA.

SeeTablesA1–A4.

15SeeBrautzschetal.(2015)forastudyexaminingR&Doutputeffectsduringthe mostrecentcrisisinGermany.