Credit and Labor Markets during the Great Depression

CreditandLaborMarketsduringthe
GreatDepression
FelipeBenguria1
UniversityofVirginia.
August,2013.
“2013MeetingoftheEconomicHistoryAssociation”
Abstract
Financialcrisesdisruptlabormarketsseverely.ThemostdramaticexampleinU.S.history
is the Great Depression. This paper uses a new dataset drawn from the Census of
Manufactures to study the relationship between financial conditions and labor market
outcomes during this period. I exploit the geographic variation in the supply of credit
across states finding a large effect of bank lending on employment and wages. The
relationshipbetweenbanklendingandemploymentisfoundonlywhencontrollingforthe
industrialcompositionofstates.
1GraduateStudent,DepartmentofEconomics,UniversityofVirginia.Email:[email protected]
1 I.Introduction.
The recent “Great Recession” and the subsequent “jobless recovery” that we
experience have renewed the interest of economists and the public on the relationship
between financial conditions and labor markets. As current events highlight, financial
crises disrupt labor markets severely. The most dramatic example of this relationship in
U.S.historyistheGreatDepression,whentheunemploymentratereachedapeakof25%,
and remained above pre‐recession levels for a decade. Whereas a large literature
pioneeredbyBernanke(1983)hasstudiedtheroleofcreditduringtheGreatDepression,
thereismuchlessresearchontherelationshipbetweencreditandlabormarketoutcomes
during this period. The goal of this paper is to examine the relationship between bank
lendingandemploymentandwagesinU.S.manufacturingduringtheDepression.Forthis
purpose,Iassembleanewdatasetonemploymentandwagesdisaggregatedbystateand
by 350 narrow industry categories for 1927 and 1937. The data are drawn from the U.S.
CensusofManufactures.Toidentifytheeffectofcreditonlabormarketoutcomes,Iexploit
thegeographicvariationinbanklendingacrossstates.Thisnewdatashowswidevariation
in the performance labor market outcomes during the Depression across industries and
states. Differences in industrial composition across states explain a large fraction of the
variationinemploymentandwages.
Two econometric issues are taken into account. First, the fact that bank loans
depend partly on the demand for credit raises concerns about the endogeneity of credit
growth.AsinBernankeandLown(1991)andCalomirisandMason(2003),Iuseindicators
of bank’s conditions at the before the Depression for subsequent growth in loan supply.
Second, the fact that the data are disaggregated by industry allows me to control for the
variationinthecompositionofeconomicactivityacrossstates.Toclarifytherelevanceof
thisissue,considerastatethatspecializesinmotorvehiclesandastatethatspecializesin
textiles.Alargerdeclineinemploymentintheformermaybeduetothefactthatthecar
industryismoresensitivetofinancialconditionsthanthetextileindustry,andnottothe
variation in the supply of credit between states. Without controlling for the industrial
2 composition of states, estimates of the effect of bank lending on employment would be
biased.
Ialsoexplorethedifferentialimpactoffinancialconditionsontheemploymentand
wagesofskilledandunskilledworkers.Recentresearch(Larrain,2013)suggeststhatthe
availability of credit can have different effects on workers of different skills. If unskilled
workers are more complementary to capital, a credit crunch is more likely to affect this
typeofworkers.
My findings indicate that the effect of bank lending on employment was
economicallylargeduringtheGreatDepression.Controllingfortheindustrialcomposition
is crucial. The effect of credit on employment is not significant in aggregate, state‐level
regressions that ignore the fact that states specialized in very different industries. I also
findarelevanteffectofcreditonwages.
This paper is related to Bernanke and Lown (1991), who study the relationship
between credit and employment during the 1990‐1991 recession in the U.S. As in their
paper,Iusethevariationinbanklendingacrossstatestoidentifythisrelationship.More
recently, Chodorow‐Reich (2013) identifies the effect of bank lending on employment
during the recent 2008‐2009 financial crises using detailed data that matches lending
banks to borrowing firms. Calvo et al (2012) examine the relationship between financial
crisesandlabormarketoutcomesacrosscountries.
Several authors have studied the behavior of labor markets in the U.S. during the
Great Depression, looking for explanations for the high unemployment and the slow
recovery.Hanes(2000)studiesnominalwagerigidityduringdifferentepisodesintheU.S.,
including the Great Depression. Hatton and Thomas (2010) provide an insightful
description of the behavior of labor markets in the U.K. and the U.S. during the period.
Cole and Ohanian (2004) argue that New Deal cartelization policies where partly
responsiblefortheweakrecoveryinemploymentintheU.S.RosenbloomandSundstrom
(1999) explore the role of industrial composition and geography in the variation of
employment,outputandwagesduringtheGreatDepressionintheU.S.
3 This paper is complementary to those which focus on the effect of financial
conditionsonoutput.CalomirisandMason(2003)andMladjan(2011)studytheeffectof
banklendingonoutputduringtheGreatDepression.Ziebarth(2013)usesthefactthatthe
stateofMississippiwasdividedintotwoFederalReservedistrictstocomparetheoutcome
of manufacturing plants in these two areas, where different policies where followed.
Ziebarth(2012)studiesthedeclineofproductivityduringtheGreatDepressionintheU.S.
In the following section I describe the data used in the paper and present some
preliminary observations on the behavior of labor markets during the Depression. In
section3Istudytherelationshipbetweenbanklendingandemploymentandwages.
II.TheData.
II.A:The“CensusofManufactures”
Tostudytheeffectofbanklendingonemploymentandwageoutcomesduringthe
Great Depression, I construct a new dataset based on the United States’ “Census of
Manufactures”. The data consists of observations on employment and wages by industry
andbystatefor1927and1937.Thereareabout350industriesineachyear.Theindustry
classification in both years is quite similar, but not exactly the same, and I develop a
concordancetomatchthem.MyworkextendsthatofRosenbloomandSundstrom(1999),
whohavecreatedadatasetwiththeinformationfor20majorindustriesfor1919‐1937.
The employment and wage figures correspond to “wage earners” (production
workers).Iamalsointerestedthedifferentialimpactoffinancialconditionsonskilledand
unskilledworkers,soforasubsetofindustries,Iassembledataonthewagesofboth“wage
earners”and“salariedemployees”(non‐productionworkers).
4 II.B:The“AnnualReportoftheComptrolleroftheCurrency”
Data on bank balance sheets by state come from the “Annual Report of the
Comptroller of the Currency”. The measure of bank lending I use is “loans and discounts
(including rediscounts)” by all active banks (national, state (commercial), savings, and
privatebanks)intheU.S.,whicharereportedseparatelybystate.AsinBernankeandLown
(1991)Icomputegrowthinlendinginnominalterms.Ialsousedataonbanks’conditions
at the beginning of the time frame under study which will serve as instruments for the
growthinbanklending.Thesearecapital‐assetratiosandtotalassetsin1927.
II.C:InitialObservations
A first observation from the employment and wage data comes from decomposing the
variationinwagesintotheshareexplainedbyindustriesandbystates.Inboththe1927
and the 1937 cross‐sections independently, I find that industrial composition is explains
roughly twice as much of the variation in both wages and employment than state fixed
effects.Moreimportantly,thevariationinemploymentbetween1927and1937islargely
industrialcompositionratherthangeography.Icomputetheuncorrelatedvarianceshare
(see Gibbons et al 2012) finding that the share of the overall variance in employment
growth explained by state fixed effects is only 3.5%, while industry fixed effects explain
41.6%ofthevariance.Geographyisonlyslightlymorerelevantinthecaseofthevariation
inwages.Figure1showsanexampleofthedistributionofwagesacrossindustriesinfour
states.Theleftsidesshowstherawdata,andtherightside, theresidualsafterremoving
industryfixedeffects.
5 0
0
k d en sity w a g e1 9 2 7 U S D
.0 0 0 5
.0 0 1
kdensityRESwage1927USD
.001
.002
.0 0 1 5
.003
Figure1:Distributionofwagesacrossindustriesinfourstates.
0
1000
x
California
Maryland
2000
3000
‐1000
‐500
0
x
California
Maryland
NewYork
Alabama
500
1000
NewYork
Alabama
Thefollowingfiguresshowthevariationinemploymentandnominalwagesbetween1927
and1937.Moststatesobservedadeclineinbothvariables,andthefiguresshowalargetail
ontheleftsiderepresentingstateswithaverylargeinemploymentandwages.
Figure2:Distributionofthe1927‐1937ChangeinEmployment
andWagesacrossStates.
1927‐37VariationinWagesAcrossStates
0
0
.5
2
1
D ensity
4
Density
1.5
6
2
8
2.5
1927‐37VariationinEmploymentAcrossStates
‐1
‐.5
kernel=epanechnikov,bandwidth=0.0701
Dlogemp
0
‐.4
.5
6 ‐.3
‐.2
kernel=epanechnikov,bandwidth=0.0231
Dlogwage
‐.1
0
.1
III.TheEffectofBankLendingonLaborMarkets.
In this section I study the relationship between bank lending and labor market
outcomes. I use the geographic variation in the growth of bank loans across states to
identify its effect on employment and wages in manufacturing industries. I begin by
estimating equations with aggregate state‐level employment and wages as dependent
variables:
∆
∆
(1)
, where represents the different labor market outcomes (employment, wage bill, and
wages) and ∆
log
log
isthegrowthinthesevariablesbetween1927and
log
1937. The independent variable ∆
log
is the
variationinbankloansduringthesameperiod.
Animportantconsiderationisthedemandforloanswillbeaffectedbytheeconomic
condition in each state. The objective is to identify the effect of the supply of credit on
employmentandwages.Forthispurpose,oneneedstoinstrumentforthegrowthinloans
withvariableswillhaveaneffectonlendingbutnotonemploymentandwages.Thisissue
hasbeenacknowledgedintheliterature,andfollowingtheBernankeandLown(1991)and
Calomiris and Mason (2003) I use capital‐asset ratios and total bank assets in 1927 as
instrumentsfor∆
.
Table1showstheresultsfortheOLSestimatesofequation(1).Ifindapositivebut
small and statistically insignificant relationship between growth in bank lending and
growthinemployment,wagebillsandwages.Ineverycolumn,Icontrolfortheinitiallevel
ofemploymentinthestateandincludefixedeffectsbygeographicregion(ofwhichthere
are six). Table 2 shows the IV results. In this case the effect of bank lending on wages is
positive and economically large. The effect on employment is large but not statistically
significant.
7 Table1:EffectofGrowthinBankLendingonLaborMarketOutcomes.State‐levelregressions.
OLSEstimates.
DependentVariable
∆
∆
∆
Growthinbankloans1927‐1937
LogEmployment1927
NumberofObservations
R‐squared
RegionFixedEffects
0.0938
(0.0897)
0.0744***
(0.0251)
45
0.439
Yes
8 0.0985
(0.104)
0.0900***
(0.0290)
45
0.441
Yes
0.00472
(0.0468)
0.0155
(0.0131)
45
0.170
Yes
Table2:EffectofGrowthinBankLendingonLaborMarketOutcomes.State‐levelregressions.
IVEstimates.
DependentVariable
Growthinbankloans1927‐1937
LogEmployment1927
NumberofObservations R‐squared
RegionFixedEffects
∆
0.504
(0.321)
0.0447
(0.0360)
45
0.123
Yes
∆
∆
0.878*
0.374* (0.472)
(0.219) 0.0334
‐0.0113 (0.0529)
(0.0246) 45
45
‐
‐
Yes
Yes
Notes:Thegrowthinbankloans1927‐1937isinstrumentedbythecapital‐assetratioin1927and
(log)totalbankassetsin1927.
Next,Itakeintoaccountthattheindustrialcompositionvariedsubstantiallyacross
states,asIdiscussedintheprevioussections.Industryfixedeffectsexplainamuchlarger
shareofthevarianceinwagesthanstatefixedeffects,andthesameistrueforthevariation
in labor market outcomes during the 1927‐1937 period. I estimate equations with
observationsdefinedattheindustryandstatelevel.
∆
∆
(2)
The subindex “i” stands for an industry (there are 350 categories). By including industry
fixedeffects
,Icomparetheeffectofcreditonlabormarketoutcomeswithinindustries
acrossstates.Thatis,Icontrolforthedifferentindustrialcompositionofthevariousstates.
Table3showsthenewresults.Inthiscase,theeffectofbanklendingonbothemployment
andwagesisstatisticallysignificantandofrelevantmagnitudes.
9 Table3:EffectofGrowthinBankLendingonLaborMarketOutcomes.Industry‐and‐Statelevel
regressions.IVEstimates.
DependentVariable
∆
Growthinbankloans1927‐
1937
LogEmployment1927
∆
∆
0.839***
(0.163)
0.710***
(0.165)
0.140*
(0.0763)
‐0.130***
(0.0108)
‐
0.116*** ‐0.00101
(0.0110) (0.00463)
NumberofObservations
2,445
2,445
2,445
R‐squared
0.416
0.456
‐
IndustryFixedEffects
Yes
Yes
Yes
RegionFixedEffects
Yes
Yes
Yes
Financialconditionsandwageinequality.
Tobewritten.
10 V.References
Bernanke,B.,(1983),“NonmonetaryEffectsoftheFinancialCrisisinthePropagationofthe
GreatDepression”,AmericanEconomicReview,73.
Bernanke,B.andGertler,M.(1990),“FinancialFragilityandEconomicPerformance”,
QuarterlyJournalofEconomics105.
Bernanke,B.,andLown,C.,(1991):“TheCreditCrunch”,BrookingsPapersonEconomic
Activity,1991‐2.
Calvo,G.,Coricelli,F.,andOttonello,P.,(2012):“Thelabormarketconsequencesoffinancial
criseswithorwithoutinflation:Joblessandwagelessrecoveries”,NBERWorkingPaper
18480.
Chodorow‐Reich,G.(2013):“TheEmploymentEffectsofCreditMarketDisruptions:Firm‐
levelEvidencefromthe2008‐09FinancialCrisis”,WorkingPaper.
Cole,H.,andOhanian,L.,(2004):“NewDealPoliciesandthePersistenceoftheGreat
Depression:AGeneralEquilibriumAnalysis”,JournalofPoliticalEconomy,112(4).
Friedman,M.andSchwartz,A.(1963),“AMonetaryHistoryoftheUnitedStates:1867–
1960”,Princeton,N.J.:PrincetonUniversityPress.
Hanes,C.,(2000):“NominalWageRigidityandIndustryCharacteristicsintheDownturnsof
1893,1929,and1981”,TheAmericanEconomicReview,90‐5.
Gibbons,S.,Overman,H.,Pelkonen,P.,(2012):“Thedecompositionofvarianceinto
individualandgroupcomponentswithanapplicationtoareadisparities”,WorkingPaper.
Larrain,M.,(2013):“CapitalAccountLiberalizationandWageInequality”,WorkingPaper.
Reinhart,C.andRogoff,K.(2009),“TheAftermathofFinancialCrises”,AmericanEconomic
Review99.
Rosenbloom,J.andSundstrom,W.,(1999),“TheSourcesOfRegionalVariationInThe
SeverityOfTheGreatDepression:EvidenceFromU.S.Manufacturing,1919‐1937",Journal
ofEconomicHistory,59.
Ziebarth,N.(2012):“MisallocationandProductivityduringtheGreatDepression”,Working
Paper.
Ziebarth,N.(2013):“Identifyingtheeffectsofbankfailuresfromanaturalexperimentin
MississippiduringtheGreatDepression”,AEJ:Macroeconomics5‐1.
11 Data:
“AnnualReportoftheComptrolleroftheCurrency”,TreasuryDeparment,1927and1937.
“Biennialcensusofmanufactures”.BureauoftheCensus,UnitedStates.Years1927,1937.
12