the effect of mergers and acquisitions on market...
TRANSCRIPT
TheEffectofMergersandAcquisitionsonMarketPowerandEfficiency
BruceA.Blonigen* JustinR.Pierce# UniversityofOregon FederalReserveBoard
NationalBureauofEconomicResearch
August2015PreliminaryandIncomplete
Abstract:Afundamentalquestionintheanalysisofmergersandacquisitions(M&As)isthe potential tradeoff between increased market power and efficiency gains. Theevidenceonthistradeoffintheliterature,however,isfarfromconclusive.CasestudiesofM&Afindmixedevidence,andestimationinmoregeneralsettingsishamperedbythedifficultyofseparatelyidentifyingproductivityfrommarkups.Inthispaper,weusenewtechniquestoseparatelyestimateproductivityandmarkupsacrossheterogeneousindustries using firm‐ or plant‐level data, and then employ these estimates in adifferences‐in‐differencesframeworktoexaminetheaverageimpactofM&Aactivityonproductivityandmark‐ups.WeemploydataonM&AtransactionscombinedwithU.S.Censusplant‐leveldatafortheentireU.S.manufacturingsectoroverthe1997to2007period.Robusttoavarietyofspecifications,wefindsignificantevidenceofincreasedmarkupsafterM&A,butnoevidenceofanyaverageimpactonplant‐levelproductivity.Moreover,wefindthattheincreaseinmarkupsgeneratesoverstatedeffectsofmergersontraditionalrevenueproductivitymeasures.Keywords:Markups,productivity,mergers.JELCodes:D24,G34,L41Acknowledgements:ThispaperhasbenefittedfromconversationswithPeterSchottandNicholasSly,aswellasparticipantsofaseminarattheLondonSchoolofEconomics.We thank Brian Hand and Dominic Smith for outstanding research assistance. Anyopinions and conclusions expressed herein are those of the authors and do notnecessarilyrepresenttheviewsoftheU.S.CensusBureau,theBoardofGovernorsoritsresearchstaff.Allresultshavebeenreviewedtoensurethatnoconfidentialinformationisdisclosed.*DepartmentofEconomics,1285UniversityofOregon,Eugene,OR,97403‐1285;541‐346‐4680;[email protected]#BoardofGovernorsoftheFederalReserveSystem,20thandCSt.NW,Washington,DC20551,202‐452‐2980;[email protected]
1
1.Introduction
Merger and acquisition (M&A) activity by firms is an important phenomenon in the
globaleconomy,involvingseveraltrillionsofdollarsofworldwideassetsannually.1
Asa result there is significant change inownershipof assetsannually, especially for
advancedeconomies.Forexample,MaksimovicandPhillips(2001)findthatabout4%
oflargemanufacturingplantsintheUnitedStateschangeownershipeveryyear.
Relatedly,cross‐borderM&Aactivityisaprimarymodebywhichmultinationalfirms
engageinforeigndirectinvestment(FDI).2
Fundamental questions in finance and industrial organization concern the
motivationforandeffectsofM&Aactivity.Andperhapsthemostfundamentalquestion
isthepotentialtradeoffbetweenincreasedmarketpowerversusefficiencygainsinthe
wakeofaM&Atransaction.Increasedmarketpowercreatesnetwelfarelosses,while
efficiency gains lower costs and create net welfare gains. While the theory is
straightforward,estimatingtheseeffectsempiricallyisdifficult.Yet,understandingthe
actualeffectsofM&Aactivity inoureconomyisnotonly important fortheacademic
literature,butalsoakeyconcernforsocialwelfare.
Therehavebeenanumberofapproachestoestimatethesepotentialeffectsof
M&Aactivity,eachassociatedwiththeirownstrengthsandweaknesses.Inthe1980s
and 1990s, a finance literature developed that used stock market event studies to
examine the impact of a variety of phenomena on firms’ profitability, including the
1ArecentForbesarticlevaluedM&Adealsin2014ataround$3.5trillion.(http://fortune.com/2015/01/05/2014‐was‐a‐huge‐year‐for‐ma‐and‐private‐equity/)2SeetheWorldInvestmentReportpublishedannuallybytheUnitedNation’sConferenceonTradeandDevelopment.
2
impactofM&Aactivity.Thesestudiesexaminechangesinreturnstofirmshareprices
after an announced M&A, generally finding that M&A activity leads to greater firm
profitability,withthebulkoftheprofitgainsaccruingtoshareholdersofthetargetfirm.
(seeRavenscraftandScherer,1987, foranoverview) Themethodologyofempirical
event studies is simple to implement consistently across a wide range of settings.
However,thisapproachisunabletoidentifywhetherthesourceofprofitabilitychanges
fromM&Aactivityisduetochangesinmarketpower,costefficiencies,orsomeother
factor.
Giventheseconcerns,morerecentanalysesoftheeffectsofM&Aactivityhave
taken primarily a case study approach,where the researcher can explicitly examine
morecloselytheparticularfeaturesofthefirmsandmarketwheretheM&Atakesplace.
Ashenfelter, Hosken and Weinberg (2014) document 49 such studies, which have
mainly focused on a few key sectors (airlines, banking, hospitals, and petroleum),
primarilybecausethesearethesectorswhereresearcherscanfinddetailedfirm‐and
product‐levelpricedata.3Mostofthesestudiesfocusonpriceandmarketsharechanges
toinfermarketpowereffects,typicallyfindingevidenceforincreasedmarketpowerby
the firms involved in the M&A activity with the exception of M&A activity in the
petroleumsector. While thesestudiescontributetoourunderstandingoftheeffects
M&Aactivity,theydosoinanincrementalfashion.Theytypicallyfocusonhighprofile
acquisitions, making it more likely that their results suffer selection bias and are
3Someprominentexamplesincludeairlines(Borenstein,1990;KimandSingal,1993,KwokaandShumilkina,2010),appliances(Ashenfelteretal.,2013),banking(FocarelliandPanetta,2003),andcottonspinning(Braguinskyetal.2013).
3
thereforenotgenerallyrepresentativeofM&Aeffectsonmarketpower.Additionally,as
mentioned, their data and analyses are specific to the particularmarket they study.
Finally,withonlyacoupleexceptions,theydonothavethedatatoestimateefficiency
effectsofM&Aactivity.4Asaresult,wedonothaveageneralpictureoftheeffectsof
M&Aactivityonmarketpowerandefficiencyintheeconomyfromthispriorliterature
usingacasestudyapproach.
Incontrast,therehavebeenafewanalysesoftheaverageeffectsofM&Aactivity
onproductivityandmarketpowerusingmicro‐leveldataforabroadsetoffirms(or
plants)acrosstheeconomy,includingMcGuckinandNguyen(1995),Maksimovicand
Phillips (2001), Gugler (2003), and Bertrand (2008). With the exception of Gugler
(2003),thesepapersusedetailedplant‐orfirm‐leveldataonthemanufacturingsector
toestimatetheeffectofM&Aactivityontotal factor(or labor)revenueproductivity,
findingthatM&Aactivitypositivelyimpactstheseproductivitymeasures.Onechallenge
facedinthesestudies,whichwefindtobeimportantinoursettingaswell,arisesfrom
theuseofrevenueasaproxyforoutput. Inparticular,whenestimatingtheeffectof
M&Aontraditionalrevenueproductivitymeasures,themarketpowereffect–operating
through output prices–makes it impossible to identifywhether changes in observed
revenueproductivityareduetochangesintrueproductiveefficiencyormarketpower.5
Gugler et al. (2003) examinesM&A transactionsworldwide using COMPUSTAT firm
data.Theytakeareducedformapproachandconductadifference‐in‐difference(DID)
4TheexceptionsofwhichweareawareareJaumandreu(2004)whichfoundsomeefficiencyeffectsfromM&AactivityintheSpanishbankingindustryandBraguinskyetal.2013.Kulick(2015)findspositiveeffectsonbothpricesandefficiencyresultingfromhorizontalmergersinthecementindustry,usingoutputdatameasuredinphysicalunitsofquantity.5BertrandandZitouna(2008)alsoexaminetheimpactofM&Aactivityonprofits,butuseaccountingdataonearnings,whichcanbeaffectedbychangesinaccountingpracticesafteranM&Atransaction.
4
analysisoftheextenttowhichsalesand/orprofitschangesignificantlyforafirmafter
anacquisition.Perhapsduetodatalimitations,theygenerallyfindmixedresults.
Thispaperprovidesanumberofnewcontributionstothisliteratureaddressing
theimportantquestionofM&Aeffectsonefficiencyandmarketpower.Thefirstisthe
applicationofrecenttechniquesdevelopedbyDeLoeckerandWarzynski(2012),where
one can separately estimate productivity and market power effects for firms with
minimalstructuralassumptions–onlycostminimizingbehaviorbyfirmsisneeded.We
useU.S.Censusplant‐leveldataforthemanufacturingsectorandmatchthatwithdata
onM&AactivityfromtheSDCPlatinumdatabasemaintainedbyThomsonFinancialto
examinetheeffectsattheplantlevelafteranacquisitionoverthe1997to2007period.6
Thisapproachhasimportantadvantagesoverthepriorstudies.Unlikethecasestudy
evidence,wecanexamineM&Aeffectsacrossabroadrangeofindustriesinaconsistent
framework. Unlike the prior studies estimating general effects ofM&A on revenue‐
basedmeasureofTFP,wecan identify theseparateM&Aeffectsonproductivityand
marketpowerinourestimates.
Asecondimportantcontributionofourpaperisthat,unlikemanystudiesinthe
prior literature, we address the difficult issue of endogenous selection due to the
possibilitythatthedecisiontoengageinM&Aactivitymayberelatedtocurrentand
expectedchangesinproductivityandmarketpower.ThiswouldbiasestimatesinaDID
frameworkthatuses“allotherfirms”asthecontrolgroup.First,oneadvantageofour
datainthisrespectisthatitisplant‐level,whereasM&Adecisionsaretypicallyatthe
6Around50%ofallU.S.M&Asareinthemanufacturingsectoroverrecentdecades,includingoursampleperiodfrom1997to2007.UseofCensusdataalsomeansthatoursampleincludebothprivateandpublicfirms.
5
firm‐level. Many of the firms in our sample aremulti‐plant,whichmakes theM&A
decisionmore independent of the performance of any one plant. While some prior
studieshaveemployedplant‐leveldata,mosthavenot.Beyondthis,weexploreother
strategiestoconstructaplausiblecontrolgroup.
Themainalternativeweuseinthispreliminarydraftistouseplantswherean
acquisitionisannounced,butisnevercompleted,asourcontrolgroup.Suchplantshave
alltheattributesnecessarytoleadtoanannouncedacquisition,eliminatingaportionof
potentialsampleselectionbias.Thus,theidentifyingassumptionisthatthereasonfor
non‐completionoftheM&Ais independentof futureproductivityandmarketpower.
Belowweprovidefurtherdiscussionforwhythismaybeareasonableassumption.In
future drafts, we also plan to examine the robustness of our results when using
propensityscorematchingtechniquestoidentifyacontrolgroup. Whilethisisnota
novelstrategy,ithasnotoftenbeenemployedinstudiesofM&Aeffects,especiallywhen
examining effects of M&A on productivity and market power.7 Finally, we plan to
examine another more‐novel strategy in future drafts – using plants that will be
acquiredinsubsequentyearsinoursampleasacontrolgroupforcurrentlyacquired
plants. This isavalidstrategy if theattributesthatmakeaplantmore likely tobea
targetoftenexistforafewyearsbeforeasuccessfulmatchwithanacquirerismade.
Likepriorstudies,wefindthatM&Aactivitysignificantlyincreasestraditional
productivitymeasuresthatuserevenue‐relateddata.However,usingourestimatesthat
separatelyidentifytruetechnicalproductivityfrommarkupsusingthemethodsofDe
Loecker andWarzynski (2012), we find new evidence for an important divergence.
7ExceptionsincludeFresard,HobergandPhillips(2015),amongothers.
6
M&Assignificantly increasemarkupsonaverage,buthaveno statistically significant
effectonproductivity.Themagnitudeofthemarkupincreaseiseconomicallysignificant
aswell,rangingfroma16%to85%increaseoftheaveragemarkupinthesample.These
resultsarerobusttowhetherwedefinethecontrolgroupasallothernon‐M&Aplants
orthemuchsmallergroupofplantsthatwereinvolvedinanannouncedM&Athatwas
ultimatelywithdrawn. They are robust to2SLS estimates,wherewe instrument for
M&Aselection,aswellastheinclusionofanumberoffixedeffects,includingplantfixed
effects.Theyarealsorobusttowhetherweexaminefive‐yearintervals,usingthefull
censusofU.S.manufacturingplants,orestimateusingannualsurveydataofonlythe
largestplantsfromtheAmericanSurveyofManufactures. Wealsopresentaplacebo
testtoverifythatsecular(pre‐)trendsarenotdrivingourresults,whichisacommon
concernwithDIDestimation.
On a final note, our focus is on the effect ofM&Aonplant‐level productivity,
consistentwithmanyprior studies ofM&A. Thereare, however, additionalways in
which M&A can affect overall firm‐level efficiency. First, an acquiring firm may
reallocate resources across the firm tomore efficient plants and/or shut down less‐
efficientplants. Inthisway,anM&Acouldincreaseefficiencyacrossthefirmevenif
plant‐levelproductivitywasunaffected.Weplantoexaminethisinfuturedrafts,but
havenotaddressedthissourceofefficiencygainsinthisversionofthepaper.Asecond
sourcemayberealizationofeconomiesofscalewithheadquarterservicesofthefirm
(management,marketing,advertising,etc.)afteranM&A.
Therestofthepaperproceedsasfollows.Inthenextsection,wediscussthedata
weusetoexaminetheeffectsofM&Aonmanufacturingplants.Wethendescribeour
7
two‐stageestimationprocessthatbeginswithestimationofplant‐levelproductivityand
markupmeasuresandisthenfollowedbyemployingtheseasdependentvariablesina
DIDframework.Wethenpresentanddiscussourresultsbeforeconcluding.
2.Data
Wemake use of two datasets for this analysis. First, to calculate productivity and
markupsforU.S.manufacturingfirms,weemployconfidentialdatafromtheU.S.Census
Bureau’sCensusofManufactures(CM).TheCMcollectsplant‐leveldataforeveryU.S.
manufacturer including, for example, total value of shipments, value added, cost of
materials,employment,investmentandthebookvalueofcapital.Thisanalysisemploys
datafromthe1997and2007CMs,focusingonlongdifferencesinplant‐levelmeasures
of productivity and markups. 8 Lastly, because the CM contains data for all U.S.
manufacturers,weareabletocalculateseveralusefulcontrolvariables includingthe
Herfindahlindexforeachindustry,aswellaseachfirm’sshareofindustryoutput.
Ourseconddataset,ThomsonFinancial’sSDCPlatinumdatabase(SDC),contains
informationonmergerandacquisitiontransactionsinvolvingbothpublicly‐tradedand
privatefirms.Foreachtransaction,SDCprovidesdataforthetargetandacquiringfirms
includingthename,addressandmajorindustryofboth,withadditionalinformationfor
the firms’ corporate parents if applicable. SDC also reports a variety of detailed
informationaboutthetransaction,includingthedatesthemergerwasannouncedand
completed, the share of the target that was purchased and the share owned after
completionofthemerger.Moreover,SDCcontainsinformationformergersthatwere
8TheCMiscollectedeveryfiveyears,inyearsendingin2and7.
8
announcedbutlaterwithdrawn.Weusethesetoffirmsinvolvedinthesewithdrawn
mergersasacontrolgroupinsomeportionsoftheanalysis.
Forpurposesofthispaper,wefocusonthesetofmergertransactionsinwhicha
U.S.manufactureristhetarget,sincethesearethefirmsforwhichCMproductionand
inputdataareavailable.Intermsoftimeperiod,weusemergertransactionsthatwere
completedorwithdrawnfrom1998to2006.Thistimeframeensuresthatweareable
toobserveeachtargetfirmbothbeforeandaftertheyareacquired.Itisalsoaperiod
that includes both periods of high M&A activity in the late 1990s and less activity
connectedwith a general slowdown of theworld economy in the early 2000s. The
sample period ends just before the start of the Great Recession. We also limit our
analysis tomerger transactions inwhich the entire target firm is purchased by the
acquirer. Withoutthisrestriction, itwouldbeimpossibletodetermineintheCensus
datawhichportionofthetargetfirmwasacquiredinthemerger.
Themergertransactiondata inSDCare linkedtotheCMdataviaanameand
addressmatchingprocedure. Suchmatchingisfarfromperfect,butwefindthatour
resultsarerobusttoanumberalternativematchingcriteriasuchaslimitingthesample
toonlytheobservationswhereboththefirmnamesandaddressesmatchtosamples
wherewerequireamatchinonlyoneofthesedimensions.
3.EmpiricalFramework
Our empirical analysis proceeds in two major steps. We first estimate plant‐level
markupsandproductivityfollowingthemethodsofDeLoeckerandWarzynski(2012),
9
whichwebrieflydescribebelow. Wethenusetheseestimatesinasecond‐stageDID
frameworktoassesstheimpactofM&Aonplant‐levelmarkupsandproductivity.
3.1.Estimationofmarkupsandproductivity
Weclosely followthe frameworkdevelopedbyDeLoeckerandWarzynski (2012) to
separately identifyaplant’smarkupfromitsproductivity.9 Usingtheirnotation,we
beginwithaproductionfunction
,..., , , ,(1)
where , … , aretheVvariable inputchoicesbyplant i intimeperiodt; is the
plant’scapitalstock;and isaproductivityparameter.Assumingcostminimization,
wecanwritetheassociatedLagrangian,
,..., , , , ∑ . ,(2)
where , … , and arethevariableinputpricesandcostofcapital,respectively.
Thefirst‐orderconditionforanygivenvariableinput(V)is
.
0.(3)
Rearrangingthefirstconditions,wecanwrite:
. (4)
9Thiscontrastswithliteraturethatmakesmorespecificassumptionsonconsumerpreferencesandmarketstructuretostudyaparticularindustry.Perhapsthemostwell‐knownexampleistheseminalworkofBerryetal.(1995)andGoldberg(1995)tomodelandestimatestructuralparametersofmarketbehaviorintheautomobilemarket.
10
Define the markup as ≡ ⁄ , where is the marginal cost of production.
Substitutingintheexpressionforthemarkup,yields
. (5)
Theleft‐handsideofequation(5)istheelasticityofoutputwithrespecttoavariable
input(whichwedenoteas ),while theratioon therighthandside is theshareof
expendituresonthevariableinputintotalsalesofthefirm(whichwedenoteas ).As
aresult,wecanexpresstheplant’smarkupasasurprisinglysimplyfunctionofthese
twoelements:
(6)
In order to obtain consistent estimates of the production functionparameters,DLW
followthemethodsproposedbyAckerberg,Caves,andFrazier(2006).Fortractability,
werestrictattentiontoproductionfunctionswithaHicks‐neutralscalarproductivity
termandassumecommontechnologyparametersforplants(withinthesameNAICS3‐
digitindustry):
,..., , ; exp .(7)
Takinglogsandassumingarandomerrorterm,wecanexpresstheproductionfunction
as:
, ; .(8)
Productivityandshockstoproductivityareunobservedtotheeconometrician,but
maybeendogenouswithinputchoicesmadebytheplant.Thisishandledthrougha
controlfunctionapproach.Inparticular,weassumethattheplant’scurrentchoiceof
materialsdependsonthecurrentlevelofanydynamicvariables(here,capitalstock),
11
productivity,andanyotherobservablevariablesthatcouldaffectoptimalmaterial
demand: , , .Invertingthisfunctionweobtain,
, , ,(9)
whichservesasaproxyindexingaplant’sproductivity,providedthatmaterialdemand
is monotonic in productivity after conditioning on a plant’s capital stock and other
observablesinthevector, . 10
Wenowassumethatproductivityfollowsasimplelawofmotion:
.(10)
Usinglaborasourvariableinputandassumingatranslogproductionfunction,wecan
derivecurrentproductivityasafunctionofourparametersviaequation(8):
.(11)
Using this,we can derive an expression for the unobserved productivity shock as a
functionofourproductionfunctionparameters.Lastperiod’sinputdecisionsshouldbe
highlycorrelated,butindependent,ofthisperiod’sinputdecisions.Thus,wecanuse
theseasinstrumentsandformthefollowingmoments
0.(12)
and then use General Method of Moment (GMM) estimation techniques to recover
consistentestimatesofourproductionfunctionparameters.Withtheseinhand,wecan
constructconsistentestimatesofourindexofproductivityandmarkups.
10ThemonotonicityofmaterialdemandinproductivityisaconditionthatholdsforabroadclassofmodelsofimperfectcompetitionasdiscussedinDLW.
12
Afewobservationsareworthnoting.First,were‐iteratethatthesemethodsare
generalenough toapplyacrossabroadrangeofheterogeneous industries. Weonly
require assumptions of cost minimization and some basic functional forms for the
productionfunction.However,followingpreviousstudiesusingthistechnique,wedo
notassumecommonproductionfunctionparametersacrossallplantsinoursample,but
estimateseparateparameters foreachNAICS3‐digit sector. Themethodabovealso
assumesproductionofa singleproduct. A recentpaperbyDeLoeckeretal. (2012)
highlights the complications when applying these techniques to multiproduct firm
becauseoneoftenhasonlyinformationonafirm’stotalinputusage,notinputusageby
eachproductitproduces.Here,wehaveplant‐level,ratherthanfirm‐leveldata,which
allowsustolargelyavoidtheissue.Thevastmajorityofplantsaresingle‐productor
veryclose,andourresultsarerobusttowhetherweexcludethesmallshareofplants
thathavesubstantialproductioninmorethanoneproduct.Afinalnoteisthatourmain
samplehasdataat five‐year intervals,whichcould lead toaweakcorrelationofour
(lagged)instruments.However,asdescribedbelow,wealsofindourresultstoberobust
tousingasmallersamplewherewehaveannualobservations.
3.2.DIDestimationofmarkupandproductivityeffectsofM&A
Inoursecondstageofestimation,weidentifytheeffectsofM&Aactivityonplant‐
levelproductivityandmarkupmeasuresthroughaDIDstrategy.Thereareanumberof
featuresofourdatathatwemustaddressinhowwedeviseourparticularDIDstrategy.
Themainaspectisthatourdataarerepeatedobservationsonplantsatthreefive‐year
intervals:1997,2002and2007.Inordertoensurethatwehaveatleastoneobservation
13
of acquired plants before the acquisition and one after the acquisition,we focus on
acquisitionsthatoccurredbetween1998and2006.Asaresult,thelengthoftimesince
the acquisition has occurred can vary across observations. We show an example in
Figure1,whereanacquisitionoccursintheyear2000.Plantsinvolvedintheacquisition
willbecodedasnotsubjecttoanM&Ainthefirstyearofoursample1997,butthen
codedassubjecttoM&Ain2002and2007.Asonecansee,thisstructuremeansthatwe
areestimatingM&Aeffectsfortheaverageperiodsincetheplantwasacquired.11
WeestimateourDIDspecificationwithtwocontrolgroups.Inordertoprovide
abaseline, the first issimplyallotherplants in thesamplethatarenotacquired. Of
course, there is the real concernof sample selectionbiaswith this relativelygeneric
controlgroup.Asmentioned,usingplantdatamaysubstantiallymitigateselectionbias
becauseM&Adecisionsaremadeatthe(multi‐plant)firm‐level,nottheplantlevel.Past
studies of M&A effects have often controlled for this sample selection bias using
propensityscoremethods.12Weleavethisanalysisforfuturedraftsofthispaper,but
alsonotethatthisstrategyiseffectiveonlytotheextentthatonecaneliminatesuchbias
throughobservables.
Instead,weestimateM&Aeffectsusinganovelcontrolgroup–plantsthatwere
partofanannouncedM&Adealthatwasnotultimatelycompleted.Suchplantsshare
allthesameattributes–observedandunobserved–thatleadthemtobetargetedforan
M&Atransaction,arelativelyinfrequentevent. Andthenumberoffaileddealsisnot
11Wehaveexperimentedwithestimatingdifferentlaglengths,butdonotfindmuchvariationineffectsovertimesubsequenttotheacquisition.12SuchstudiesincludeHeyman,Sjöholm,andTingvall(2007),BertrandandZitouna(2008),ArnoldandJavorcik(2009),BandickandGörg(2010),andFresard,Hoberg,andPhillips(2013).
14
trivial.Forexample,BranchandYang(2003)showthatabout11%oftheover1000U.S.
mergerstheyevaluateoverthe1991‐2000periodfailtocomplete.Akeyworrywith
thiscontrolgroupstrategyisthattherearealsofactorsunobservedtousthatbecome
observedtotheinvolvedfirms(especiallywithrespecttothetargetfirm)thatleadtoa
M&A deal failure. There is a small literature that evaluates failure of M&A deal
completion.13Mostofthecovariateswithexplanatorypoweraresolelyrelatedtothe
acquiring firm, including the type of financing it uses to fund the deal, its size, and
measures of their attitude toward completing the deal, which Baker and Savasoglu
(2002)indicateisthebestsinglepredictorofdealcompletion.Thereisn’tanyobvious
correlationbetweenthesefactorsrelatedtotheacquiringfirmandthefuturemarket
powerand/orproductivityof the targetplants,which is the focusofourstudy. The
media often reports that disputes betweenmanagers of the two firms–often termed
“socialissues”–canleadtofailedM&Adeals.14
4.Results
4.1.First‐stageestimatesofmarkupsandproductivity
WebeginbyusingthemethodsinDLWasdescribedinsection3.1toestimateamarkup
andameasureofproductivityforeachofthe187,100manufacturingplantsinourfull
sample.Table1providesthemeanandstandarddeviationoftheseDLWmeasuresof
markupandproductivity.ThemeasureofDLWproductivityissimplyanindexfroma
13AdditionalstudiesbeyondthoselistedinthetextincludesMitchellandPulvino(2001),Officer(2007),andBranch,Wang,andYang(2008)14TheinherentproblemisthattherearetwosetsofallseniormanagerscomingintoanM&Aandthisduplicationmustbeeliminated.Willis,A.(2001)“’Socialissue’maybekeytobankmergers,”TheGlobeandMail(Canada),August28,p.B17isanexamplearticleinthebusinesspressonthisissue.
15
translogproductionfunction.Thus,itslevelisnotmeaningfulperse,butchangesinthe
indexwillreflectpercentagechangesintheplant’sproductivity.Asonecansee,there
issubstantialvariationinthisproductivitymeasureacrossplants.
Themarkupsmeasurehowmuchmoreisthepricechargedbythefirmthanits
marginalcost.Wefindquitehighmarkups,withanaveragemarkup(theratioofprice
tomarginalcost)ofaround5.5withastandarddeviationover7. Therangeofthese
markups ismuchhigherthanwhatDLWfindforSlovenianfirms,andabouttwiceas
highonaverageasthatfoundbyDeLoeckeretal.(2012)forIndianfirms.However,an
importantdifferenceisthatweestimatethesemarkupsattheplant‐level,notthefirm‐
level.Whilethepriceiswhatthefirmcancharge,the(marginal)costsarespecifictothe
plant. Inputs used and costs incurred by the firm for headquarter services, such as
advertising,distribution,centralmanagementandR&D,willtypicallynotbeaccounted
forintheseproductionplants.Inotherwords,weshouldexpectlargermarkupsatthe
plant‐level because the price has to only cover the plant’s costs, but the firm’s non‐
productioncostsaswell.Therefore,likeourDLWproductivitymeasure,wearenotas
interestedinthelevelperse,buthowthemarkupchangeswithanM&A.
For comparison purposes,wewill be examining the impact ofM&A onmore
traditional revenue productivity measures; specifically, the log of total factor
productivity,whichwecalculateusing thesamemethodologyasFoster,Haltiwanger
and Syverson (2008), and a simplemeasure of log labor productivity. Table 1 also
providesthemeanandstandarddeviationofthesemeasures.
Forthepurposesofcontrollingforpotentialsampleselectionbias,weemploya
sampleofonlytheplantswhereaM&Adealwasannounced.Thisreducesoursample
16
substantiallyto4,200plants. ThelasttwocolumnsofTable1providethemeanand
standarddeviationofourmarkupandproductivityparametersforthissampleofplants.
There are not large differences in these descriptive statistics between this reduced
sample and the full sample, though both averagemarkup and all three productivity
measuresareslightlyhigherinthesmallersampleofplantssubjecttoanannounced
M&Adeal,suggestingthatthereistargetingoffirmswithplantsthathavehigherthan
averagemarkupsandproductivity.
4.2.Second‐stageDIDestimatesofM&Aeffects:Baselineestimates
Wenowtakeourfirst‐stageestimatesanduseaDIDspecificationtoexaminetheimpact
ofM&Asonthese threemeasuresofproductivityandtheDLWestimateofmarkups.
FollowingastandardDID,ourindependentvariableofinterestisaninteractionbetween
anindicatorvariablethataplanthasbeen“treated”bybeingsubjecttoanM&Aandan
indicatorthatitistheperiodaftertreatmenthastakenplace.Welabelthisvariableas
the“TargetFirminthePost‐M&APeriod.”Wealsoincludetheindicatorvariableforthe
“Post M&A Period” to control for any secular trends, year fixed effects to capture
macroeconomic shocks affecting all plants identically in a particular year, and plant
fixed‐effects,whichcontrolforanytimeinvariantfactors(observedorunobserved)that
affectanindividualplant’sproductivityormarkup.
Table2providesresultsforourfullsampleofplants.Forthefullsample,weonly
lookatthelongdifferenceofoursample, includingonlytheyears1997and2007,to
avoid concerns about ambiguity of definingwhen a control plant is in a “postM&A
period.”ThefirsttwocolumnsofTable2provideestimatesoftheimpactofM&Aon
17
traditional revenue‐based measures of productivity, log revenue TFP and log labor
productivity. Weestimateapositive,butstatistically insignificant,M&Aeffecton log
revenueTFP,butapositiveandstatisticallysignificantM&Aeffectonourmeasureof
labor productivity. As mentioned, these traditional revenue‐based measures can
confoundchangesinmarketpowerwithchangesintrueproductivity.
To address this, the second two columnsofTable2provide resultswhenwe
apply our DID exercise to our first‐stage estimates where we estimate separate
estimatesofameasureofproductivityandmarkupforeachplant.Nowweseeasharp
divergence. There is a negative, but statistically insignificant, effect on productivity
whenaplanthasbeenacquired,butastatisticallysignificantincreaseinmarkup.The
magnitudeofthemarkupeffectissizeableaswell,withthe1.167effectcorresponding
toroughlya20%increaserelativetothemeanmarkup.
4.3.Second‐stageDIDestimatesofM&Aeffects:AnnouncedM&Asample
Our baseline estimates do not control for sample selection issues, other than
through inclusion of plant‐level fixed effects, which capture time‐invariant
characteristics. However, plants of firms targeted for M&A may have unobserved
attributes that will affect changes in future markups and productivity that could
spuriously show up as M&A effects. To address this potential for selection on
unobservablesweconstructanovelcontrolgroup.
Specifically, this alternate control group consists of plants in firms thatwere
announcedastargetsofM&A,butforwhichthemergerwasultimatelywithdrawn.By
definition,theseplantspossesstheattributesnecessarytoleadtoanannouncedmerger,
18
eliminatingaportionofpotentialsampleselectionbias.Oneconsequenceofrestricting
the control group to the setof firmswithwithdrawnmergers is that it substantially
reducesoursamplesizefromover187,000observationstojust4,200observations.
Estimates obtained with this alternate control group are qualitatively very
similartothoseobtainedwiththebaselinesample.Asshowninthefirsttwocolumns
ofTable3,ourrevenue‐basedproductivitymeasurescontinuetoshowapositiveM&A
effect.Coefficientestimatesforthemergereffectvariablearestatisticallysignificantfor
bothrevenue‐basedmeasures,thoughtheeffectsareoffairlymodestmagnitudes–less
than5%oftheaverageproductivity.Moreover,weobtainthesamequalitativeresults
fortheDLWmeasuresonproductivityandmarkupswiththisannouncedM&Asample
andthefullsample–thereisnodiscernibleM&Aeffectontheproductivityofacquired
plants,butsubstantialM&Aeffectsontheirmarkups. Here, thecoefficientestimates
suggestamuchlargermarkupeffectofsomethingaround60%oftheaveragemarkup
in the sample. Moreover, the results indicate that even after our initial control for
sampleselection,wecontinuetofindthattheeffectofmergersonrevenueproductivity
measures isoverstated for theaverageplant in themanufacturingsector,due to the
increaseinmarkups.15
Weconductseveraladditionalrobustnessexerciseswiththissampleofplants
thataresubjecttoanannouncedM&Adeal. First,toconfirmthatourresultsarenot
driven by spurious matches between the Thomson Financial and Census data, we
construct a sample composed only of firms with an exact match between the two
15Kulick’s(2015)examinationofhorizontalmergersinthecementindustryalsofindsevidenceofover‐estimatedeffectsofmergersonrevenueproductivity,usinganentirelydifferentestimationprocedure.
19
databases(i.e.matchingonfirmname,address,cityandstate).Requiringthestricter
criterionofanexactmatchgivesmorecertaintyofthematchquality,butalsoreduces
oursamplebyafairamount.InTable4weshowresultswhenwelimitourannounced
M&Asampletoonlythoseobservationsthatmeetastrictmatchcriterion.Thislimits
thenumberofobservationsevenfurther from4,200to just1,900. Nevertheless, the
estimatesarequalitativelyidenticaltothoseinTable3withthelessstrictmatchcriteria.
Thesecondrobustnesscheckinvolvesasecondtechniquetocontrolforsample
sectionbiasviaa2SLSstrategythatthatinstrumentsforwhetheranannounceddealis
completedornot.Inparticular,weinstrumentforM&Acompletionbyinteractingthe
“PostM&APeriod” variablewith indicator variables for the year theM&A dealwas
announcedfortheplant. Theideaisthatseculartrends,suchasbusinesscyclesand
changesinantitrustenforcementintheyearthattheM&Adealisannouncedcouldaffect
whetherthedealiscompleted.Theresultsofthese2SLSestimates,reportedinTable5,
arequalitativelyverysimilartotheOLSestimatedM&Aeffects.
In Table 6, we explore a possible source of heterogeneity in our results. In
particular,onemightthinkthattheM&Aeffectonmarkupswouldbegreaterfortargets
thatalreadyhavesignificantmarketsharefortheproducttheyproduce.Weexamine
thisby interactingthetargetfirm’smarketshareintheplant’sproductwithourDID
regressors. As Table 6 shows, there is an estimated positive coefficient on the
interactionof“TargetFirminPostM&APeriod”with“FirmMarketShare”inthemarkup
equation, but it is not statistically significant. These interactions are statistically
insignificantintheotherequationsexceptforthelogrevenueTFPequation,wherethis
20
samefocusinteractiondoeshaveapositiveandstatisticallysignificantcoefficient.The
weakstatisticalresultsmaybeduetolowpowerfromarelativelysmallsamplesize.
AfinalrobustnesscheckisexaminationofonlytheplantsintheannouncedM&A
sample for which we have a balanced panel of annual data. While the Census of
Manufacturesdataisgatheredonfive‐yearintervals,theAnnualSurveyofManufactures
(ASM)gathersdataonlargerplantsannually.UseofannualdatamaybehelpfulforDLW
estimationmethods, as noted above, and also allows formore precise coding of the
timing ofM&A announcements in our data. Because the sample frame for theASM
changeseveryfiveyears,werestrictattentiontothesetofplantsthatarepresentin
everyyear,toavoidspuriousinstancesofentryandexit.Thisrestrictionreducesthe
announcedM&Asamplefrom4,200to3,300.Table7providesresultsforthissample
ofannualobservationsonlargerplants. Onceagain,weobtainqualitativelyidentical
estimates.
In summary, our results are robust to a number of alternative specifications
meant to address various concerns onemay have with our estimates. They give a
consistentmessage that acquiredplantsdonot see statistically significant effects on
productivity,butdoexperiencepositivesignificanteffectsonmarkups.Wethinkthese
resultsaresurprisinglyrobustgiventhesmallnumberofobservationsinourannounced
M&A sample. Future drafts will explore a couple other alternative control‐group
strategiestoprovidefurtherrobustnessexercises.
4.4.Second‐stageDIDestimatesofM&Aeffects:Aplacebotest
21
Afinalconcernwithouranalysisistheworrythatpre‐existingseculartrendsfor
the treatment and control group could be driving spurious correlations for our
estimatedM&Aeffects.Toaddressthisconcernweconstructthefollowingplacebotest.
WeadddataforCensusyears1987and1992toourannouncedM&Asampleandthen
addindicatorvariablesthatflagourtargetedplantsin1992and1997.Theseareyears
when these targeted plants were not yet acquired; hence, these are the placebo
treatments. Iftherearesignificantcoefficientsontheseplacebotreatments,itwould
castsubstantialdoubtonthevalidityoftheestimateeffectweareobtainingforthetrue
treatmentvariable. Relatedly, if significant, thesevariableswould indicatepre‐trend
differences in our treated and control groups,which is amajor concern for anyDID
analysis. In other words, this particular placebo test is also a test for pre‐trend
differences.
AsresultsinTable8show,ourestimateseasilypassthisplaceboandpre‐trend
test. Thefirstfourcolumnsshowtheresultswhenweincludetheplacebovariables,
“Target Firm in 1992” and “Target Firm in 1997.” Estimated coefficients on these
variablesare statistically insignificant forall threemeasuresofproductivityandour
markupmeasure.Thissampleofobservationsissomewhatdifferentfromourprevious
samplesdiscussedaboveduetotheadditionalyearsof1987and1992.Socolumns5
through 8 of Table 8 verify that our base specification yields qualitatively identical
results to our previous analysis showing no M&A effects on DLW productivity, but
significant,positiveM&Aeffectsonmarkups.
5.Conclusion
22
This paper estimates the effects of mergers and acquisitions on productivity and
markupsofplantsacrossallU.S.manufacturingindustries.Ouranalysismakesuseof
high‐qualityU.S.CensusBureaudatacoveringtheuniverseofU.S.manufacturingplants,
whicharematchedtothesetofprivateandpublicmergersandacquisitionstrackedby
ThomsonFinancialintheirSDCPlatinumdatabase.Usinganoveltechniquedeveloped
byDeLoeckerandWarzynski(2012),wefindthatcontrollingformarketpowereffects
isimportantwhenestimatingtheeffectofM&Aonfirmandplant‐levelproductivity.In
particular,wefindthatwhileM&Aisassociatedwithastatisticallysignificantincrease
inrevenue‐basedproductivitymeasures,thereisalsopositiveandsignificanteffecton
markups. We findnoeffectofM&Aonaproductivitymeasure thatcontrols for this
changeinmarketpower.
23
ReferencesAckerberg,D.A.,K.Caves,andG.Frazer(2006).StructuralIdentificationofProductionFunctions.Mimeo,UCLA.
Arnold, J. M., & Javorcik, B. S. (2009). Gifted kids or pushy parents? Foreign directinvestment and plant productivity in Indonesia. Journal of International Economics,79(1),42‐53.Ashenfelter,O.C.,Hosken,D.S.,&Weinberg,M.C.(2013).ThePriceEffectsofaLargeMerger of Manufacturers: A Case Study of Maytag‐Whirlpool. American EconomicJournal:EconomicPolicy,5(1),239‐61.Ashenfelter,O.C.,Hosken,D.,&Weinberg,M.C.(2014).“DidRobertBorkUnderstatetheCompetitive Impact of Mergers? Evidence from Consummated Mergers,” NationalBureauofEconomicResearchWorkingPaper19939.Baker, M., & Savaşoglu, S. (2002) "Limited arbitrage in mergers and acquisitions."JournalofFinancialEconomics64(1),91‐115.Bandick, R., & Görg, H. (2010). Foreign acquisition, plant survival, and employmentgrowth.CanadianJournalofEconomics,43(2),547‐573.Bertrand,O.,&Zitouna,H. (2008).Domesticversus cross‐borderacquisitions:whichimpactonthetargetfirms’performance?AppliedEconomics,40(17),2221‐2238.Borenstein,S.(1990).AirlineMergers,AirportDominance,andMarketPower.AmericanEconomicReview,82(2),400‐4.Braguinsky,S.,Ohyama,A.,Okazaki,T.&Syverson,C.(2013).Acquisitions,productivityandprofitability:EvidencefromtheJapanesecottonspinningindustry.Mimeo.Branch, B., & Yang, T. (2003). Predicting successful takeovers and risk arbitrage.QuarterlyJournalofBusinessandEconomics,3‐18.Branch, B., Wang, J., & Yang, T. (2008). A note on takeover success prediction.InternationalReviewofFinancialAnalysis,17(5),1186‐1193.DeLoecker,J.,&Warzynski,F.(2012).MarkupsandFirm‐LevelExportStatus.AmericanEconomicReview,102(6),2437‐71.DeLoecker,J.,Goldberg,P.K.,Khandelwal,A.K.,&Pavcnik,N.(2012).Prices,markupsandtradereform(No.w17925).NationalBureauofEconomicResearch.Focarelli,D.,&Panetta,F.(2003).AreMergersBeneficialtoConsumers?EvidencefromtheMarketforBankdeposits.AmericanEconomicReview,1152‐1172.
24
Foster, L., Haltiwanger, J., & Syverson, C. (2008). Reallocation, Firm Turnover, andEfficiency:SelectiononProductivityorProfitability?AmericanEconomicReview,98(1),394‐425. Fresard, L.,Hoberg,G.,&Phillips,G. (2013).The incentives for verticalmergers andverticalintegration.ResearchPolicy,2,452‐466.Fresard,L.,Hoberg,G.,&Phillips,G.(2015).InnovativeActivitiesandtheIncentivesforVerticalAcquisitionsandIntegration.Mimeo.Gugler,K.,Mueller,D.C.,Yurtoglu,B.B.,&Zulehner,C.(2003).Theeffectsofmergers:aninternationalcomparison.InternationalJournalofIndustrialOrganization,21(5),625‐653.Heyman,F.,Sjöholm,F.,&Tingvall,P.G.(2007).Istherereallyaforeignownershipwagepremium?Evidence frommatched employer–employeedata. Journalof InternationalEconomics,73(2),355‐376.Jaffe, S., & Weyl, E. G. (2012). The first‐order approach to merger analysis.Mimeo,HarvardUniversity,Cambridge,MA,USA.Kim,E.H.,&Singal,V. (1993).Mergersandmarketpower:Evidence fromtheairlineindustry.TheAmericanEconomicReview,549‐569.Klette,T.J.(1999).Marketpower,scaleeconomiesandproductivity:estimatesfromapanelofestablishmentdata.TheJournalofIndustrialEconomics,47(4),451‐476.Kwoka,J.,&Shumilkina,E.(2010).ThePriceEffectofEliminatingPotentialCompetition:EvidencefromanAirlineMerger.JournalofIndustrialEconomics,58(4),767‐793.Kulick,RobertB.(2015).HorizontalMergers,Prices,andProductivity.Mimeo.Maksimovic,V.,&Phillips,G.(2001).Themarketforcorporateassets:Whoengagesinmergersandassetsalesandarethereefficiencygains?JournalofFinance,2019‐2065.Malmendier, U., Moretti, E., & Peters, F. (2013). Winning by Losing: Evidence onOverbiddinginMergers.Mimeo.McGuckin,R.H.,&Nguyen,S.V.(1995).Onproductivityandplantownershipchange:Newevidencefromthelongitudinalresearchdatabase.TheRANDJournalofEconomics,257‐276.Mitchell,M.,&Pulvino,T. (2001).Characteristicsofriskandreturn inriskarbitrage.JournalofFinance,2135‐2175.
25
Officer,M.S.(2007).Areperformancebasedarbitrageeffectsdetectable?Evidencefrommergerarbitrage.JournalofCorporateFinance,13(5),793‐812.Weinberg,M.C.,&Hosken,D.(2013).Evidenceontheaccuracyofmergersimulations.ReviewofEconomicsandStatistics,95(5),1584‐1600.Werden, Gregory J. "Demand elasticities in antitrust analysis."Antitrust Law Journal(1998):363‐414.
26
FIGURE1:DataandCodingofM&ATreatment
1997 2002 2007
Acquisition(Year=2000)
CensusYears
Pre‐AcquisitionObservation
Post‐AcquisitionObservations
27
Table1:First‐stagemeasuresofmarkupsandproductivity
FullSample
SampleWithOnlyAnnouncedM&A
Deals
Variable MeanStandardDeviation Mean
StandardDeviation
Logrevenuetotalfactorproductivity 4.07 0.60 4.38 0.63Loglaborproductivity 4.33 0.65 4.69 0.69DLWproductivitymeasure ‐0.74 1.87 ‐1.67 2.57DLWmarkup 5.49 7.97 7.20 11.27Notes:Wecalculatetotalfactorproductivityatthe3‐digitNAICSlevelusingthesamemethodologyasinFoster,Haltiwanger,andSyverson(2008).TheDLWmarkupandproductivitymeasuresareestimatedatthe3‐digitNAICSlevelusingthetechniquesinDeLoeckerandWarzynski(2012).Thereare187,100observationsinthefullsampleand4,200observationsinthesamplethatonlyincludesplantsthatwerepartofannouncedM&Adeals.
28
Table2:BaselineResultswithFullSampleofPlants
Variables
LogRevenueTFP
LogLaborTFP
DLWProductivity
DLWMarkup
TargetFirminPostM&APeriod 0.022 0.044* ‐0.063 1.167*** (0.016) (0.024) (0.060) (0.215)Post‐M&APeriod 0.067*** 0.161*** ‐0.322*** 0.557*** (0.002) (0.002) (0.004) (0.025) Constant 4.031*** 4.235*** ‐0.556*** 5.154*** (0.001) (0.002) (0.003) (0.018) Observations 187,100 187,100 187,100 187,100R‐squared 0.89 0.82 0.93 0.88Notes:Robuststandarderrorsinparentheses. Estimatesforplantandyearfixedeffectsaresuppressed.P‐valuesof0.01,0.05,and0.10aredenotedby***,**,and*respectively.
29
Table3:BaselineResultsWhereControlPlantsarefromWithdrawnM&ADeals(ReducedSample)
VariablesLogRevenue
TFPLogLabor
TFPDLW
ProductivityDLW
Markup
TargetFirminPostM&APeriod 0.150*** 0.123* ‐0.078 4.638** (0.057) (0.070) (0.112) (2.320)Post‐M&APeriod ‐0.158*** ‐0.125* 0.156 ‐4.237* (0.058) (0.071) (0.121) (2.314) Constant 4.358*** 4.590*** ‐1.477*** 6.442*** (0.009) (0.013) (0.028) (0.156) Observations 4,200 4,200 4,200 4,200R‐squared 0.85 0.73 0.90 0.84Notes:Robuststandarderrorsinparentheses. Estimatesforplantandyearfixedeffectsaresuppressed.P‐valuesof0.01,0.05,and0.10aredenotedby***,**,and*respectively.
30
Table4:ReducedSamplewithStricterMatchCriteria
VARIABLESLogRevenue
TFPLogLabor
TFPDLW
ProductivityDLW
Markup
TargetFirminPostM&APeriod 0.283*** 0.181* ‐0.011 2.696*** (0.076) (0.097) (0.153) (0.626)Post‐M&APeriod ‐0.320*** ‐0.265*** 0.108 ‐1.928*** (0.078) (0.101) (0.163) (0.684) Constant 4.364*** 4.562*** ‐1.474*** 5.752*** 0.014 0.019 0.037 0.146 Observations 1,900 1,900 1,900 1,900R‐squared 0.83 0.66 0.90 0.79Notes:Robuststandarderrorsinparentheses. Estimatesforplantandyearfixedeffectsaresuppressed.P‐valuesof0.01,0.05,and0.10aredenotedby***,**,and*respectively.
31
Table5:2SLSEstimateswithReducedSample
VariablesLogRevenue
TFPLogLabor
TFPDLW
ProductivityDLW
Markup
TargetFirminPostM&APeriod ‐0.079 0.292 ‐0.183 6.222** (0.164) (0.241) (0.558) (2.996)Post‐M&APeriod 0.054 ‐0.281 0.252 ‐5.699** (0.153) (0.225) (0.520) (2.793) Constant 4.357*** 4.591*** ‐1.478*** 6.452*** (0.009) (0.012) (0.029) (0.155) Observations 4,200 4,200 4,200 4,200Notes:Robuststandarderrorsinparentheses. Estimatesforplantandyearfixedeffectsaresuppressed.P‐valuesof0.01,0.05,and0.10aredenotedby***,**,and*respectively.
32
Table6:InteractionswithFirmMarketShare
VariablesLog
RevenueTFPLogLabor
TFPDLW
ProductivityDLW
Markup
TargetFirminPostM&APeriod 0.115* 0.095 ‐0.138 4.515* (0.061) (0.075) (0.119) (2.392)Post‐M&APeriod ‐0.127** ‐0.100 0.195 ‐4.153* (0.062) (0.076) (0.127) (2.383)TargetFirminPostM&APeriod×FirmMarketShare 1.283*** 1.031 2.229 4.611 (0.461) (0.715) (1.762) (10.479)Post‐M&APeriod×FirmMarketShare ‐1.123** ‐0.94 ‐1.425 ‐3.051 (0.439) (0.671) (1.688) (10.298) Constant 4.358*** 4.590*** ‐1.477*** 6.443*** (0.009) (0.013) (0.028) (0.156) Observations 4,200 4,200 4,200 4,200R‐squared 0.85 0.73 0.90 0.84Notes:Robuststandarderrorsinparentheses. Estimatesforplantandyearfixedeffectsaresuppressed.P‐valuesof0.01,0.05,and0.10aredenotedby***,**,and*respectively.
33
Table7:ResultsfromReducedSamplewithAnnualData
Variables
LogRevenueTFP
LogLaborTFP
DLWProductivity
DLWMarkup
TargetFirminPostM&APeriod ‐0.047 ‐0.084 ‐0.432*** 2.815*** (0.070) (0.077) (0.118) (1.091)Post‐M&APeriod 0.046 0.053 0.377*** ‐2.408** (0.069) (0.075) (0.134) (1.078) Constant 4.554*** 4.887*** 3.881*** 4.969*** (0.015) (0.023) (0.125) (0.194) Observations 3,300 3,300 3,300 3,300R‐squared 0.90 0.83 0.99 0.76
Notes:Robuststandarderrorsinparentheses.Estimatesforplantandyearfixedeffectsaresuppressed.P‐valuesof0.01,0.05,and0.10aredenotedby***,**,and*respectively.
33
Table8:PlaceboRegressionsUsingDatafrom1987through2007
Variables
LogRevenueTFP
LogLaborTFP
DLWProducti‐vity
DLWMarkup
LogRevenueTFP
LogLaborTFP
DLWProducti‐vity
DLWMarkup
TargetFirminPostM&APeriod 0.102* 0.114 ‐0.196** 2.249** 0.103* 0.102 ‐0.262*** 2.425** (0.062) (0.076) (0.092) (1.122) (0.057) (0.071) (0.082) (0.988)PostM&APeriod ‐0.137** ‐0.097 0.220** ‐1.680 ‐0.138** ‐0.088 0.276*** ‐1.822* (0.062) (0.077) (0.098) (1.106) (0.058) (0.072) (0.095) (0.984)TargetFirminYear1992 0.052 0.062 0.124 ‐1.397 (0.061) (0.086) (0.091) (0.996)TargetFirminYear1997 ‐0.044 ‐0.010 0.125 0.541 (0.052) (0.070) (0.104) (0.933)Constant 4.205*** 4.115*** ‐0.416*** 4.013*** 4.205*** 4.115*** ‐0.416*** 4.017*** (0.011) (0.016) (0.032) (0.105) (0.011) (0.016) (0.032) (0.105) 6,200 6,200 6,200 6,200 6,200 6,200 6,200 6,200 0.81 0.71 0.89 0.75 0.81 0.71 0.89 0.75Notes:Robuststandarderrorsinparentheses. Estimatesforplantandyearfixedeffectsaresuppressed. P‐valuesof0.01,0.05,and0.10aredenotedby***,**,and*respectively.