Committing to grow: Privatizations and firm dynamics in East Germany

Akcigi , U uk; Alp, Ha un; Diegmann, And é; Se ano-Vela de, Nicolas
Wo king Pape
Commi ing o g ow: P i a iza ions and i m dynamics in
Eas Ge many
IAB-Discussion Pape , No. 01/2024
P o ided in Coope a ion wi h:
Ins i u e o Employmen Resea ch (IAB)
Sugges ed Ci a ion: Akcigi , U uk; Alp, Ha un; Diegmann, And é; Se ano-Vela de, Nicolas (2024) :
Commi ing o g ow: P i a iza ions and i m dynamics in Eas Ge many, IAB-Discussion Pape , No.
01/2024, Ins i u ü A bei sma k - und Be u s o schung (IAB), Nü nbe g,
h ps://doi.o g/10.48720/IAB.DP.2401
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/283019
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by-sa/4.0/deed.de
IAB‑DISCUSSIONPAPER
A icles on labou ma ke issues
01|2024 Commi ing o G ow: P i a iza ions and
Fi m Dynamics in Eas Ge many
U uk Akcigi , Ha un Alp, And é Diegmann, Nicolas Se ano‑Vela de
ISSN 2195‑2663
Commi ing o G ow: P i a iza ions and
Fi m Dynamics in Eas Ge many
U uk Akcigi (Uni e si y o Chicago, Halle Ins i u e o Economic Resea ch (IWH)),
Ha un Alp (Fede al Rese e Boa d),
And é Diegmann (Halle Ins i u e o Economic Resea ch (IWH), Ins i u e o Employmen
Resea ch (IAB), Cen e o Eu opean Economic Resea ch (ZEW)),
Nicolas Se ano‑Vela de (Bocconi Uni e si y, IGIER)
Mi de Reihe „IAB‑Discussion Pape “ will das Fo schungsins i u de Bundesagen u ü
A bei den Dialog mi de ex e nen Wissenscha in ensi ie en. Du ch die asche
Ve b ei ung on Fo schungse gebnissen übe das In e ne soll noch o D ucklegung
K i ik ange eg und Quali ä gesiche we den.
The “IAB‑Discussion Pape ” is published by he esea ch ins i u e o he Ge man Fede al
Employmen Agency in o de o in ensi y he dialogue wi h he scien i ic communi y.
The p omp publica ion o he la es esea ch esul s ia he in e ne in ends o
s imula e c i icism and o ensu e esea ch quali y a an ea ly s age be o e p in ing.
Con en s
1 In oduc ion...............................................................................6
2 Ins i u ional Backg ound .............................................................. 11
3 A Model o Fi ms wi h Employmen Commi men s ................................. 13
3.1 S a ic En i onmen ..................................................................... 13
3.2 Dynamics ................................................................................ 15
3.3 Taking S ock............................................................................. 18
4 Da a and Desc ip i e S a is ics........................................................ 19
4.1 Con ac Da a ........................................................................... 19
4.2 Ma ching Con ac s o Fi ms .......................................................... 23
5 Empi ical Analysis o Labo Commi men s and Fi m Dynamics ................... 24
5.1 Iden i ica ion S a egy.................................................................. 24
5.2 Labo Commi men s and Fi m G ow h .............................................. 30
5.3 Labo Commi men s and P oduc i i y G ow h ..................................... 34
5.4 Labo Commi men s and Ma ke Exi ................................................ 35
6 Quan i a i e Analysis................................................................... 38
6.1 Calib a ion ............................................................................... 38
6.2 Calib a ion Resul s and Goodness o Fi ............................................ 39
6.3 Coun e ac uals ......................................................................... 41
7 Conclusion .............................................................................. 42
Re e ences .................................................................................. 44
Supplemen a y Appendix.................................................................. 50
A Fu he Empi ical Resul s.............................................................. 50
B Da a Addendum – ISUD Da a En i onmen .......................................... 65
C Da a Addendum – Me ging Con ac s o Mannheim En e p ise Panel Da a ....... 72
D Da a Addendum – T euhand Fi m Su ey Da a...................................... 79
D.1 Cons uc ing Fi m‑Le el Capi al S ock and TFP Measu es ........................ 79
D.2 Me ging Con ac s o T euhand Fi m Su ey Da a ................................. 84
Lis o Figu es .............................................................................. 88
Lis o Tables ............................................................................... 89
IAB‑Discussion Pape 01|2024 3
Abs ac
We de elop a labo demand model ha encompasses p e‑ma ch hi ing cos a ising
om igh labo ma ke s. Th ough he lens o he model, we s udy he e ec o labo
ma ke igh ness on i ms’ labo demand by applying no el Ba ik ins umen s o he
uni e se o adminis a i e employmen da a on Ge many. In line wi h heo y, he IV
esul s sugges ha a 10 pe cen inc ease in labo ma ke igh ness educes i ms’
employmen by 0.5 pe cen . When accoun ing o sea ch ex e nali ies, we ind ha he
indi idual‑ i m wage elas ici y o labo demand educes om ‑0.7 o ‑0.5 a he
agg ega e le el. Fo he 2015 minimum wage in oduc ion, he elas ici ies imply only
modes disemploymen e ec s mi o ing empi ical ex‑pos e alua ions. Mo eo e , he
doubling o igh ness be ween 2012 and 2019 led o a signi ican slowdown in
employmen g ow h by 1.1 million jobs.
Zusammen assung
Dieses Papie un e such eine wi scha spoli ische Maßnahme, die da au abziel , die
Beschä igung wäh end de P i a isie ung os deu sche Un e nehmen nach dem Fall des
Eise nen Vo hangs zu siche n. Die neuen Eigen üme n de Un e nehmen e p lich en
sich zu A bei spla zzusagen, wobei Sank ionen bei Nich einhal ung e aglich
implemen ie wa en. Mi einem dynamischen endogenen Wachs umsmodell zeigen wi
d ei Wege au , wie sich A bei spla zzusagen au Un e nehmen auswi ken.
A bei spla zzusagen üh en (i) zu eine e ze e Un e nehmensg öße, (ii) s eige n die
P oduk i i ä und (iii) e höhen die Ma k aus i swah scheinlichkei . Anhand eines
Ins umen al a iablenansa zes und un e Ve wendung on einziga igen Ve agsda en
we den die Modellaussagen bes ä ig . Das Ins umen de A bei spla zzusagen üh zu
eine um 22 P ozen punk e höhe en jäh liche Wachs ums a e de Beschä igung, eine
um 14 P ozen punk e höhe en jäh liche Wachs ums a e de P oduk i i ä und eine um
3,6 P ozen punk e höhe e Wah scheinlichkei des Ma k aus i s. Das kalib ie e
Wachs umsmodell zeig , dass ohne diese Zusagen die Gesam beschä igung nach 10
Jah en um 15 P ozen nied ige gewesen wä e. Da übe hinaus e weis sich eine
al e na i e wi scha spoli ische Maßnahme de In es i ionssub en ionen zu S eige ung
de P oduk i i ä als eue und wenige e ek i in de ku zen F is .
4 IAB‑Discussion Pape 01|2024

JEL
D22, D24, J08, L25
Keywo ds
indus ial policy, p i a iza ions, p oduc i i y, size‑dependen egula ions
Acknowledgemen s
We hank ou discussan s Chang‑Tai Hsieh, B ian Via d, S e an Obe nbe ge , Ricca do
Zago, and he semina and con e ence pa icipan s a No a Business School, Bocconi
Uni e si y, Halle Ins i u e o Economic Resea ch, NBER Summe Ins i u e, Cen e o
Eu opean Economic Resea ch (ZEW), Fede al Rese e Boa d, Richmond FED, The Socie y
o Labo Economis s, UVA, Ba celona School o Economics Summe Fo um, CSEF‑IGIER
Symposium on Economics and Ins i u ions, EDHEC, FEP School o Economics and
Managemen Uni e si y o Po o, Hi o subashi Uni e si y, Lisbon Mac o Wo kshop, and
AIEA‑NBER Con e ence. Fo p o iding aluable suppo o da a access and da a
expe ise, we hank Sand a Go schalk (Mannheim En e p ise Panel), Alexande Gieble
(ISUD da a) and Ch is Be hold, An je Klünde and Jana Michaelis (Ge man Fede al
A chi es). Akcigi g a e ully acknowledges inancial suppo om he Max Planck
Humbold ‑Resea ch Awa d 2019. The iews in his pape a e solely he esponsibili y o
he au ho s and should no be in e p e ed as e lec ing he iews o he Boa d o
Go e no s o he Fede al Rese e Sys em o any o he pe son associa ed wi h he
Fede al Rese e Sys em.
5 IAB‑Discussion Pape 01|2024
1 In oduc ion
Indus ial policy is o en designed du ing imes o signi ican s uc u al change caused
by shocks ela ed o compe i ion (e.g., China shock), dis up i e inno a ions (e.g., IT
e olu ion), o poli ical u bulence (as seen in he pos ‑So ie e a in Eas e n Eu ope).
Du ing hese pe iods, policymake s conside he immedia e cos s o labo ma ke
dis up ion and how o mi iga e hem by in oducing policies cen e ed a ound
employmen conside a ions. Howe e , he e emains a lack o e idence on how such
in e en ions dynamically a ec ealloca ion and i m beha io .
In his pape , we s udy a no el policy designed o p ese e employmen du ing he
p i a iza ion o Eas Ge man i ms ollowing he all o he I on Cu ain. The
p i a iza ion p ocess ep esen ed a pe iod o in ense s uc u al change and aised
signi ican conce ns abou he social cos s associa ed wi h high unemploymen . In
esponse, policymake s equi ed ha new owne s o Eas Ge man i ms commi o
employmen a ge s, wi h penal ies imposed o alling below he commi ed
employmen le el. In o al, hese labo commi men s we e applied o o e 18,000
p i a iza ion con ac s, co e ing mo e han 900,000 wo ke s in Eas Ge many.
Ou analysis p oceeds in h ee s eps. Fi s , we in oduce a dynamic model whe e i ms
ope a e unde employmen a ge s. An impo an ea u e o he model is ha i ms
in es esou ces o imp o e hei p oduc i i y, allowing us o s udy he endogenous
esponse o p oduc i i y o such a ge s. The model highligh s h ee channels h ough
which a i m is a ec ed by an employmen a ge ha is binding i.e., in which he
a ge is highe han he cu en employmen le el. The i s channel s ems om he
i m’s s a ic labo decision, which induces an upwa dly dis o ed employmen choice.
The second channel a ises dynamically as binding a ge s induce highe p oduc i i y
g ow h. This is because mo e p oduc i e i ms hi e mo e wo ke s in ou model, and
his s uc u e c ea es addi ional incen i es o in es in p oduc i i y imp o emen s o
a oid he penal ies. These wo channels imply ha i ms wi h binding employmen
a ge s expe ience highe employmen and p oduc i i y g ow h. The hi d channel
ope a es h ough he ex ensi e ma gin choice o he i m o exi . Fi ms wi h binding
employmen a ge s a e mo e likely o exi as binding a ge s in oduce a ixed‑cos ‑like
s uc u e in he cash low o he i m.
In he second s ep, we ake hese p edic ions o he da a. The empi ical analysis elies
on a no el da ase om he Ge man a chi es ha con ains all he documen a ion
p oduced by he T euhandans al (THA), he go e nmen agency esponsible o he
p i a iza ion p ocess. Ou da a con ains de ailed con ac ‑le el in o ma ion on
6 IAB‑Discussion Pape 01|2024
employmen a ge s and deadlines, as well as he da es and esul s o each on‑si e
audi o he employmen commi men . To measu e i m‑le el p oduc i i y, we me ge
ou con ac ‑le el in o ma ion wi h da a om he Mannheim En e p ise Panel (MUP)
and he SOESTRA su ey o Eas Ge man i ms.
The empi ical iden i ica ion o he link be ween employmen a ge s and i m dynamics
is a challenging ask. The eason is ha employmen a ge s a e no andomly
alloca ed and migh hus bias ou empi ical es ima es. In he spi i o he li e a u e on
judge leniency (Bhulle e al., 2020; Dobbie/Song, 2015; Be ns ein e al., 2019), we
de elop an ins umen al a iable (IV) app oach ha exploi s he e ogeneous p e e ences
o p i a ize s and hei quasi‑ andom assignmen o i ms. To do so, we es ima e he
p opensi y o a p i a ize o equi e binding labo commi men s. We show ha : (i) he
p obabili y o ecei ing a binding con ac inc eases con inuously along he labo
p e e ence measu e, (ii) hese p e e ences a e he e ogeneous ac oss p i a ize s, and (iii)
hey a e pe sis en ac oss ime. Impo an ly, we also p o ide e idence consis en wi h
he quasi‑ andom assignmen mechanism o i ms o p i a ize s. To do so, we use
in o ma ion om he balance shee s o i ms be o e hei p i a iza ion. Consis en wi h
anecdo al e idence abou he o ganiza ion o THA, we ind no e idence o an
economically o s a is ically signi ican co ela ion o ou ins umen wi h a wide ange
o sec o al cha ac e is ics, employmen and e enue measu es, and o he indi idual
cha ac e is ics o he p i a ize s.
Consis en wi h he model’s p edic ions, we ind ha binding labo a ge s a e
associa ed wi h highe employmen and p oduc i i y g ow h, as well as inc eased i m
exi o e he labo commi men pe iod. Ou IV es ima es e eal a 22 pe cen poin s
highe annual employmen g ow h a e o i ms wi h binding labo con ac s compa ed
o hose wi hou . Binding labo con ac s also lead o an addi ional yea ly p oduc i i y
g ow h o app oxima ely 14 pe cen poin s. Addi ional e idence based on i ms’
pa en ing ac i i y du ing he commi men pe iod also suppo s hese indings.
Fu he mo e, i ms wi h binding con ac s exhibi , on a e age, a 3.6 pe cen poin s
highe p obabili y o exi ing by he end o he commi men pe iod. Rela i e o he
baseline exi a e o 5.5 pe cen , his ep esen s an economically sizable inc ease in he
exi ma gin. We show ha hese esul s a e obus o al e na i e speci ica ions in e ms
o he measu emen o he dependen a iables, he cons uc ion o he ins umen al
a iable, and he inclusion o addi ional con ac ual cha ac e is ic as con ol
a iables.
In he las s ep o ou analysis, we calib a e ou model o he da a and un se e al
coun e ac ual scena ios in o de o quan i a i ely assess he impo ance o he
di e en channels on i m beha io . To iden i y he pa ame e s o he model, especially
he penal y o no mee ing he a ge , we ma ch he e ec s o binding employmen
7 IAB‑Discussion Pape 01|2024
commi men s on i m ou comes unco e ed in ou empi ical analysis. The calib a ed
model is able o ep oduce he main pa e ns in he da a well. Impo an ly, he model
eplica es i m‑le el g ow h pa e ns ac oss he employmen commi men dis ibu ion
as well as he pos ‑commi men employmen dynamics, which a e no a ge ed in he
calib a ion p ocess.
We s udy h ee coun e ac ual economies. We i s simula e an economy wi hou
employmen a ge s and ind ha agg ega e employmen would be 15 pe cen poin s
lowe pe manen ly a e 10 yea s. Nex , we decompose he impac o employmen
a ge s on o al employmen in o i s “s a ic” and “dynamic” componen s by shu ing
down i s impac on p oduc i i y imp o emen s. Ou calib a ed model a ibu es
one‑ hi d o he employmen g ow h o dynamic e ec s in he sho un. In he long
un, he en i e pe manen inc ease in employmen is d i en by he dynamic e ec s.
Las ly, we conside an al e na i e policy o subsidizing in es men in o p oduc i i y. We
calib a e he subsidy a e o achie e he agg ega e employmen g ow h in he da a
du ing he commi men pe iod. The implied cos o such a policy is high, amoun ing o
5 pe cen o he ou pu . While his policy esul s in highe pe manen employmen
le els ela i e o he employmen a ge policy, he inc ease is mo e g adual o e he
commi men pe iod. In o he wo ds, he subsidy policy is less e ec i e in he sho un
o p ese e employmen .
The pape con ibu es o he ecen li e a u e e isi ing he me i s and cos s o
indus ial policies. Lane (2022) and Choi/Le chenko (2021) use his o ical da a o s udy
he dynamic impac o he Sou h Ko ean hea y and chemical indus y d i e om 1973
o 1979. Lane (2022) shows ha his empo a y d i e shi ed Ko ean manu ac u ing in o
mo e ad anced ma ke s, c ea ing du able indus ial change. Choi/Le chenko (2021) link
he associa ed i m‑le el subsidies o pe sis en e ec s on i m size due o a
combina ion o lea ning‑by‑doing and inancial ic ions. Kaloup sidi (2018) and
Ba wick/Kaloup sidi/Zahu (2021) s udy he Chinese in e en ion in he shipbuilding
indus y. Kaloup sidi (2018) es ima es ha policy in e en ions educed shipya d cos s
by 13‑20 pe cen and ealloca ed in e na ional ma ke sha es.
Ba wick/Kaloup sidi/Zahu (2021) disen angle he a ious subsidies du ing he
in e en ion and es ima es hei impac . Acemoglu e al. (2018) demons a e ha
s a egic indus ial policies ha e signi ican in luence o e i ms’ composi ion, wi h he
po en ial o ha ness he economy’s i m selec ion p ocess o ampli y o e all
p oduc i i y gains. Liu (2019) embeds indus ial policy in a p oduc ion ne wo k se ing
and applies i o in e en ions in Sou h Ko ea in he 1970s and mode n‑day China.
Finally, Gio celli/Li, 2021 es ima e he long‑ e m e ec s o echnology and know‑how
ans e s on China’s s uc u al ans o ma ion using da a om he Sino‑So ie alliance
8 IAB‑Discussion Pape 01|2024
employmen a ge s. Finally, low‑p oduc i i y i ms, z˜ < z˜∗∗, ind i oo cos ly o ope a e
a he a ge le el o employmen , bu s ill hei labo choices a e dis o ed owa ds he
a ge le el. These labo choices unde line he i s channel h ough which i ms wi h
binding con ac s i.e., ha ing an employmen a ge la ge han he op imal
employmen le el unde no a ge expe ience a highe employmen g ow h h ough he
con ac pe iod, as hei employmen is simply dis o ed upwa d owa ds he a ge . We
e e o his channel as he “di ec ” e ec o binding labo con ac s on employmen
g ow h.
Figu e 2 illus a es key implica ions o he exis ence o employmen a ge s on i m
p o i s. The le panel plo s o al p o i s wi h espec o i m‑le el p oduc i i y. The
black line p o ides he benchma k o i ms wi h no commi men , while he dashed ed
line plo s p o i s o i ms unde commi men . Dashed e ical lines show he h eshold
p oduc i i y le els, z∗ and z∗∗. The plo shows ha dis o ed i ms ha e lowe p o i s.
As Equa ion 2 cla i ies, hese lowe p o i s a e a ibu able o he p esence o penal ies
ha in oduce a ixed‑cos ‑like s uc u e. These penal ies inc ease in magni ude as he
le el o he a ge ises. This will ha e impo an dynamic implica ions o he exi
decision, which will be discussed in he nex subsec ion.
The igh panel o Figu e 2 plo s ma ginal p o i s ac oss i m p oduc i i y and shows
ha dis o ed i ms ha e a highe ma ginal p o i wi h espec o p oduc i i y. This is
in ui i e: inc easing p oduc i i y no only inc eases p o i s bu also educes he amoun
o dis o ion (i he i m is bunching) o penal y paid (i he i m is no bunching) o
hose i ms wi h binding con ac s. In a dynamic se ing, which will be in oduced la e ,
his implies ha he inc ease in p o i s om p oduc i i y imp o emen s will be highe
o dis o ed i ms ela i e o undis o ed ones i.e., dis o ed i ms would be mo e
willing o in es in p oduc i i y imp o emen s. Since he labo choice is posi i ely
co ela ed wi h he le el o p oduc i i y, his highe p oduc i i y g ow h by
binding‑con ac i ms cons i u es he second channel o highe employmen g ow h
h ough i s dynamic implica ions on p oduc i i y g ow h.
3.2 Dynamics
Nex , we desc ibe he dynamic decisions o he i ms. A any poin , he owne decides
whe he o s ay in he economy o exi . I she decides o exi , she needs o pay an exi
cos , ne o ou side op ion, which we pa ame e ize wi h Ce.9 I she s ays in he
9 This cos e lec s no only he penal ies om missing he a ge as employmen becomes ze o upon
exi ing, bu also any o he , implici o explici , cos s due o impai ed ela ions be ween he acqui e
o he i m and he go e nmen .
15
IAB‑Discussion Pape 01|2024

Figu e 2: P o i s ac oss Fi m P oduc i i y
12345678
P oduc i i y
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0.55
P o i s
No commi men
Wi h commi men
12345678
P oduc i i y
0.08
0.085
0.09
0.095
0.1
0.105
0.11
0.115
0.12
0.125
0.13
Ma ginal P o i s
No commi men
Wi h commi men
No es: The le and igh panels plo s o al and ma ginal p o i s ac oss i m‑le el p oduc i i y,
espec i ely. The black line p o ides he benchma k o i ms wi h no commi men , while he dashed
ed line plo s p o i s o i ms unde commi men . Dashed e ical lines show he h eshold
p oduc i i y le els, z∗ and z∗∗ .
economy, she makes he op imal labo choice, as desc ibed abo e, and decides how
much o in es in p oduc i i y g ow h by choosing he Poisson a i al a e o imp o ing
16
Panel A: To al P o i s
Panel B: Ma ginal P o i s
IAB‑Discussion Pape 01|2024
he p oduc i i y, x, wi h he ollowing cos unc ion (in e ms o he homogeneous
good)
ϕ 2
˜c(x|z˜) = x zw
2
which is con ex in he success p obabili y x, and ϕ is he scale pa ame e o he cos .
This cos unc ion assumes ha he highe he cu en le el o p oduc i i y, he highe
he cos o in es men . The pa icula no maliza ion o he cu en le el o p oduc i i y
implies ha i m g ow h is consis en wi h Gib a ’s law in he absence o employmen
a ge s: he g ow h a e o su icien ly la ge i ms (high p oduc i e i ms) is independen
o hei size. I he in es men is success ul, he p oduc i i y imp o es om z o
(1 + λ)z, whe e λ is he pa ame e ha con ols he s ep size in p oduc i i y
imp o emen . Finally, we assume ha he labo commi men con ac s expi e a he
i m le el a he a e µ i.e., he employmen a ge becomes ze o and no longe
binding.
Gi en his s uc u e, he dynamic p oblem o he i m can be ep esen ed by he
ollowing alue unc ion:


 



2
˜π(˜z, l∗)w − ϕ x zw
2
∂V (˜z, l∗)
 
(3)
V (˜z, l∗) − = max −Cew, max +x [V (˜z(1 + λ), l∗) − V (˜z, l∗)]



∂ x≥0
+µ [V (˜z, 0) − V (˜z, l∗)]
whe e V (˜z, l∗) is he i m alue. The ou e maximiza ion p oblem de e mines he
endogenous exi decision o he i m. The alue o s aying is de e mined in he second
maximiza ion p oblem whe e he i m chooses how much o in es on p oduc i i y
g ow h.10 The i s line includes he ins an aneous p o i s, minus he cos o in es men
on p oduc i i y. The second line exp esses he change in i m alue when he i m is
success ul wi h i s in es men in imp o ing p oduc i i y a he a e x. The las line
ep esen s he change in alue when he labo commi men con ac expi es a he
a e µ.
The ex ensi e ma gin choice abo e gi es ise o he s anda d op imal s opping
p oblem. Fi ms ollow a cu o ule unde which hey choose o exi when hei
p oduc i i y alls below a ce ain h eshold. The h eshold p oduc i i y o exi is highe
o i ms wi h highe employmen a ge s due o he associa ed la ge ixed cos s. In
o he wo ds, condi ional on ini ial p oduc i i y, i ms wi h highe employmen a ge s
would be mo e likely o exi he economy.
10 Employmen choice was cha ac e ized abo e, so i is aken as gi en he e.
17
IAB‑Discussion Pape 01|2024
Fo i ms ha choose o s ay in he economy, op imal le el o in es men in
p oduc i i y is gi en by ( he a i al a e o imp o ing p oduc i i y):
V (˜z(1 + λ), l∗) − V (˜z, l∗)
x(˜z, l∗) = (4)
1
ϕz˜1−α
which depends on he inc ease in he alue o he i m in he case o a success ul
imp o emen in p oduc i i y. Since he alue unc ion inhe i s he p ope ies o he
p o i unc ion, he in es men a e on p oduc i i y mimics he pa e n o ma ginal
p o i s simila o he case illus a ed in Panel B o Figu e 2: i is highe o i ms ha
a e dis o ed by he binding employmen a ge s.
3.3 Taking S ock
We inish he model discussion by summa izing he main insigh s and p edic ions. Ou
model cla i ies h ee channels h ough which binding employmen commi men s a ec s
i m decision. The i s channel s ems om he i m’s s a ic labo decision unde
binding con ac s, which induces an upwa dly biased employmen choice (equa ion 1)
i.e., labo hoa ding owa ds he a ge employmen le el. This esul s in a ansi o y
expansion in employmen , wi h i ms e e ing o hei undis o ed size once he policy
expi es.11 The second channel a ises dynamically as i ms ope a ing unde hese
binding commi men s wi ness highe p oduc i i y g ow h induced by highe ma ginal
p o i s os e ing in es men in p oduc i i y imp o emen s. This dynamic e ec a ises as
i ms seek o “escape” om con ac ually imposed penal ies. Unlike he ansi o y
na u e o he i s e ec , he employmen gains esul ing om hese dynamic
imp o emen s in p oduc i i y a e pe sis en and con inue beyond he commi men
pe iod. The hi d channel ope a es h ough he ex ensi e ma gin choice o he i m o
exi . Fi ms wi h binding employmen a ge s a e mo e likely o exi o e he
commi men pe iod as binding a ge s in oduce a ixed‑cos ‑like s uc u e in he cash
low o he i m.
In he nex sec ion, we will empi ically es hese heo e ical implica ions o he model
in he da a. In pa icula , we look a he impac o binding employmen con ac s on (i)
employmen g ow h, (ii) p oduc i i y g ow h, and (iii) exi decision a he i m le el.
11 Fo simplici y, ou model abs ac s om labo adjus men cos s. Inco po a ing such cos s would no
a ec he ansi o y na u e o his channel bu would imply some ansi ion phase occu ing when
he commi men expi es.
18
IAB‑Discussion Pape 01|2024
4 Da a and Desc ip i e S a is ics
4.1 Con ac Da a
The analysis elies on a unique da ase om he Ge man Fede al A chi es
(Bundesa chi ) con aining all con ac s and documen a ion p oduced by he THA. These
da a a e con iden ial, and, hanks o an ins i u ional coope a ion wi h he IWH Ins i u e,
we a e among he i s o gain access o hem. Impo an ly, he agency digi ally
eco ded all con ac s o moni o ing and en o cemen pu poses in mo e han 500 da a
ables (ISUD Sys em). These ables p o ide comp ehensi e con ac ‑le el in o ma ion on
he p i a iza ion o asse s including all he employmen commi men s ha ha e been
ag eed upon as well as da es and audi s associa ed wi h each commi men . Appendix
Sec ion B p o ides a de ailed desc ip ion o he explo ed ISUD da a ables.
The da ase con ains 18,235 con ac s wi h labo commi men s. Fo each con ac , we
obse e he con ac da e e.g., he da e he con ac is signed wi h a no a y and he
commi ed le el o employmen along wi h he da e o he inal commi men . As
shown in Panel A o Figu e 3, 90 pe cen o employmen commi men s a e signed
be ween 1991 and 1994.12 In o al, all labo con ac s amoun o mo e han 900,000
commi ed wo ke s, ep esen ing abou 20 pe cen o he ini ial wo k o ce popula ion
o THA i ms.
These con ac s unde wen egula audi s conduc ed by con ac manage s employed
by he THA. These audi s se e as he basis o he ealized employmen le els
analyzed in his pape . The con ac manage s would app oach he con ac ing pa y
and conduc audi s ei he h ough physical isi s o he i m o ia documen a ion. On
a e age, each con ac was audi ed 6.3 imes, wi h a minimum o 1 audi and a
maximum o 84 audi s. Figu e 3, Panel B, illus a es ha app oxima ely 82 pe cen o all
con ac s we e audi ed a leas wice.
We ocus on hese con ac s o measu e employmen g ow h du ing he commi men
pe iod. While we always ha e da a on employmen le els a he inal commi men
da e, we app oxima e he ini ial employmen le els using he i s labo audi , which
ypically ook place be ween h ee o six mon hs a e he con ac was signed wi h he
no a y. Panel A o Table 1 shows con ac ‑le el employmen in o ma ion a he s a
da e o he con ac , he inal le el, and he a ge le el. We p o ide desc ip i e
12 Less han 2 pe cen o all labo con ac s a e w i en ou in 1996 o la e (15 con ac s a e obse ed
in 2002). Fo 168 con ac s we do no obse e he da e o he con ac .
19
IAB‑Discussion Pape 01|2024
s a is ics o he 14,726 con ac s wi h a leas wo audi s. Wi h, on a e age 66
employees, i ms had been ela i ely sizable a he onse o p i a iza ion. O e he
cou se o he commi men pe iod o , on a e age, h ee yea s, i ms dec eased hei
size. Panel B ela es he ini ial size o he inal a ge . The ac ion o i ms ini ially
below hei a ge is 22 pe cen , while abou 20 pe cen o he i ms ecei e a a ge
ha is equal o hei ini ial size. In 10 pe cen o cases he i m s ays a he commi ed
size in he i s and las audi s.
Figu e 3: Con ac s and Labo Audi s
A: Numbe o con ac s o e ime
0 200 400 600 800 1000
Numbe o commi ed employmen (in 1000)
0 2000 4000 6000
Numbe o con ac s
1990 1992 1994 1996 1998 2000
Yea
Con ac s Labo a ge
0 .05 .1 .15 .2
Pe cen
1 5 10 2015 25+
To al Numbe o Audi s pe Con ac
No es: Panel A plo s he o al numbe o signed con ac s wi h labo commi men s be ween 1990 and
2000 as well as he accumula ed numbe o commi men employmen . Panel B plo s he dis ibu ion
o labo audi s pe con ac .
Sou ce: ISUD.
20
B: Audi dis ibu ion
IAB‑Discussion Pape 01|2024

Table 1: Summa y S a is ics
N Mean SD Minimum Maximum
(1) (2) (3) (4) (5)
A: A e age i m size
Ini ial employmen 14,726 66.20 319.57 0.00 23,691
Final employmen 14,726 60.67 194.26 0.00 8540
Final employmen a ge 14,726 52.98 183.25 1.00 6906
B: Ini ial size ela i e o a ge
F ac ion ini ially below a ge 14,726 0.22 0.42 0.00 1.00
F ac ion ini ially a a ge 14,726 0.20 0.40 0.00 1.00
F ac ion ini ially & inally a a ge 14,726 0.10 0.30 0.00 1.00
C: Penal ies
Numbe o obse ed iola ion 1,272 2.24 1.29 1.00 12.00
To al numbe o iola ed labo 1,272 111.58 393.22 0.24 8,567.47
Penal y pe missed employee (in 1000 EUR) 1,272 10.77 10.67 0.10 58.52
D: P oduc i i y
Ini ial p oduc i i y 3,387 9.99 1.43 3.51 16.27
Final p oduc i i y 3,584 12.02 0.786 10.46 14.81
Ini ial TFP 3,118 6.81 1.23 2.70 9.79
Final TFP 2,219 7.32 1.08 3.53 10.31
E: Ma ke exi
Exi un il inal commi men yea 4,622 0.055 0.22 0.00 1.00
No es: The able shows summa y s a is ics o p i a iza ion con ac s. In Panel A, ini ial employmen
le el is calcula ed o con ac s wi h a leas wo obse a ions. This co esponds o 14,726 con ac s. In
Panel C, we obse e 1,272 con ac s wi h a leas one obse ed labo commi men iola ion. Panels D
and E a e based on he linkage be ween con ac s and ex e nal i m‑le el da a desc ibed in Appendices
C and D. Panel D p o ides model‑consis en p oduc i i y and TFP measu ed in logs. Panel E shows he
exi indica o a he end o he labo commi men pe iod.
Sou ce: ISUD, MUP, SOESTRA.
Fo a subse o 1,272 i ms, we obse e he o al amoun o penal ies claimed by he
THA due o iola ions o labo commi men s as well as he o al numbe o iola ions.
Based on hese numbe s, we calcula e he penal y pe missed employee aking in o
accoun he p o a a empo is condi ion. This means ha , o example, i a i m is
missing con inuously one employee o e he cou se o h ee yea s, he i m misses, in
o al, h ee commi men s and needs o pay h ee imes one employee. Condi ional on
ha ing a leas one labo iola ion, he a e age i m de ia es 2.2 imes. The cumula i e
numbe o missed employmen o e mul iple iola ions co esponds, on a e age, o 111
21 IAB‑Discussion Pape 01|2024
wo ke s.13 Finally, consis en wi h documen a ion on THA policy, ou calcula ions
sugges ha he a e age penal y pe missed employee amoun s o 10,768 EUR.14
Figu e 4 empi ically assesses he impo ance o employmen a ge s in a ec ing i m’s
labo choices. The ho izon al axis measu es he di e ence be ween he ealized
Figu e 4: Employmen Dis ibu ion a ound he Commi men Le el
Employmen > Commi men Employmen < Commi men
Excess mass (b) = 6.521
S anda d e o = .8195
Excess sha e (%) = 0.243
0 1000 2000 3000
F equency
-40 -20 0 20 40
Realized minus Commi ed Employmen
Coun e ac ual dis ibu ion Obse ed dis ibu ion
No es: The igu e shows he employmen dis ibu ion a ound he commi ed employmen (dema ca ed
by he e ical ed line a 0) o con ac s be ween 1990‑1995. The blue line in do s is a his og am o
ac ual employmen ela i e o he commi men a ge in he inal commi men yea . Each poin shows
he numbe o obse a ions in employmen coun bins (de ia ion be ween he a ge and he ealized
employmen ). The solid line benea h he empi ical dis ibu ion is a wel e‑deg ee polynomial i ed o
he empi ical dis ibu ion excluding he a ea o missing one employee and ha ing h ee employees
mo e han commi ed. The shaded egion in yellow is he es ima ed excess mass, which is 652 pe cen
o he a e age heigh o he coun e ac ual dis ibu ion benea h. S anda d e o is calcula ed using a
pa ame ic boo s ap p ocedu e. Es ima ion based on Che y e al. (2011).
Sou ce: ISUD.
employmen measu ed a he las audi o a commi men and he inal employmen
a ge . Fi ms below 0 a e smalle in e ms o hei ealized employmen ela i e o he
commi ed le el, whe eas i ms abo e 0 ha e a la ge employmen wi h espec o hei
commi ed le el. The igu e plo s he bin coun s a ound he no malized a ge shown
by he ed e ical line a ze o, wi h each bin ep esen ing a uni o employmen
de ia ion. A s iking ea u e o he da a is he la ge spike exac ly a he commi ed le el
o employmen , sugges ing he impo ance o hese cons ain s o i ms’ labo choices.
13 The maximum cumula i e missed employmen amoun s o 8,567 wo ke s. This is abo e he
maximum o he inal employmen a ge in Panel A, as he e can be mul iple iola ions. In addi ion,
he maximum numbe in Panel A co esponds o he inal commi men le el.
14 The e ec i e penal y can be lowe because o “condi ions beyond he pu chase ’s con ol”
(Dodds/Wäch e , 1993), enego ia ion, mino iola ions o a ge s (Baga ell all), judicial decisions, and
bank up cy.
22
IAB‑Discussion Pape 01|2024
Following Che y e al. (2011), we es ima e an excess mass a ound he h eshold o
652 pe cen ela i e o he a e age heigh o he coun e ac ual dis ibu ion.15
4.2 Ma ching Con ac s o Fi ms
The audi s conduc ed on he con ac ‑le el da a do no p o ide in o ma ion ega ding
i m‑le el sales and pos ‑p i a iza ion ma ke exi . To cons uc p oduc i i y measu es,
we u ilize da a om he Mannheim En e p ise Panel (MUP). By me ging he con ac
pa ne s’ names wi h he owne ship in o ma ion in he MUP, we can gene a e
p oduc i i y measu es s a ing om 1993. This me ging p ocess enables us o measu e
i m sales a he end o he commi men pe iod o nea ly all linked i ms. Fo a
de ailed desc ip ion o he da a me ge be ween he wo da ase s, please e e o
Supplemen a y Appendix C. O e all, we iden i y he espec i e legal uni behind 4,622
con ac s.
We compu e wo measu es o p oduc i i y g ow h o i ms unde employmen
commi men s. Fi s , we conside a model‑consis en p oduc i i y measu e gi en by
sales pe wo ke adjus ed by he labo sha e in he p oduc ion unc ion. To assess
ini ial p oduc i i y, we use in o ma ion om he opening balance shee s o THA i ms
wi h con ac da a ega ding employmen a he ime o p i a iza ion. To measu e
p oduc i i y a he end o he employmen commi men , we me ge he sales
in o ma ion om MUP wi h he inal employmen audi . Second, we use he Soes a
i m‑le el su ey o THA i ms o calcula e i m‑le el To al Fac o P oduc i i y (TFP) as
desc ibed in Appendix D.
Panel D o Table 1 e eals a subs an ial inc ease in p oduc i i y du ing he commi men
pe iod. This no ewo hy imp o emen aligns wi h he documen ed con e gence p ocess
obse ed in he yea s ollowing euni ica ion. Wi hin he i s decade a e euni ica ion,
app oxima ely hal o he measu ed labo p oduc i i y gap and o e one‑ hi d o he
GDP pe capi a gap be ween Eas and Wes ha e been closed (Bu da, 2006).16 S a ing
in 1990, ou calcula ion sugges s an inc ease in p oduc i i y o 2 log poin s. The
calcula ed imp o emen s in TFP be ween he ini ial con ac yea and he inal
commi men pe iod amoun s o 0.51 log poin s. Finally, Panel E o Table 1 p o ides
15 The ed line in Figu e 4 plo s he es ima ed coun e ac ual densi y based on a wel e‑deg ee
polynomial (p = 12) and an asymme ic window a ound he h eshold R = [3, −1]. R = [3, −1]
deno es he omi ed bunching ange including i ms ha ing up o h ee mo e employees han hei
commi ed a ge . The yellow shaded egion depic s he es ima ed excess mass a ound he h eshold.
Figu es A.5 o A.7 p o ide obus ness checks wi h espec o he deg ee o he polynomial, he
bunching window, and he bin de ini ion. Table A.11 shows he esul s by sub‑samples.
16 Based on agg ega e s a is ics, Bachmann e al. (2022) show ha GDP pe wo ke inc eased by abou
0.7 log poin s be ween 1991 and 2000.
23
IAB‑Discussion Pape 01|2024
in o ma ion on ma ke exi o he ma ched con ac s wi h he MUP. The sample size o
his analysis is la ge compa ed o he p oduc i i y assessmen , because o missing
da a in e ms o he sales a iable. A he end o he commi men pe iod, an exi sha e
o 5.5 pe cen is obse ed.
5 Empi ical Analysis o Labo
Commi men s and Fi m Dynamics
5.1 Iden i ica ion S a egy
Add essing he empi ical challenge o non‑ andom alloca ion o labo commi men s o
i ms is c ucial when analyzing i m‑le el esponses. Fo example, i high labo a ge s
a e assigned o low‑g ow h i ms, i may lead o an unde es ima ion o he impac o
employmen commi men s. To ackle his issue, we de elop a amewo k o
educed‑ o m iden i ica ion inspi ed by me hods used in s udies on judge leniency
(Bhulle e al., 2020; Dobbie/Song, 2015; Be ns ein e al., 2019) and pa en e alua o s
(Sampa /Williams, 2019). These s udies ypically es ima e he ixed ai s o p e e ences
o decision make s ega ding ou comes unde hei con ol, such as leniency o
oughness. By combining his es ima e o he ixed ai wi h he quasi‑ andom
alloca ion o decision make s, we ob ain an exogenous shi e ha helps mi iga e
po en ial biases in u u e cases.
The p oposed empi ical amewo k o he analysis is well‑sui ed o ou ins i u ional
se ing o se e al easons. The numbe o p i a iza ions in hose yea s mean ha THA
agen s ypically wo ked on mul iple cases. Impo an ly, he b eakneck speed o
p i a iza ions gene a ed, wi hin o ices, andomness in he assignmen o hese cases. A
consul an wi h he THA in hose yea s desc ibed he p ocess as “an excep ional
si ua ion whe e he e was a lo o imp o isa ion.” Finally, a he momen o
p i a iza ion, each THA agen possessed signi ican disc e ion in es ablishing he
condi ions o he i m o be p i a ized, hus lea ing oom o p i a ize ai s o ma e
in he p ocess.17
17 The o ganiza ional s ess and complexi y o p i a izing he Eas Ge man economy canno be
unde es ima ed. The THA was desc ibed as “an adolescen bu eauc acy, bo n o chaos and des ined
o be phased ou wi hou e e unc ioning no mally. I is a human c ea ion, whipped oge he
quickly and hen pu unde ex eme p essu e wi hou ime o p epa e” (Dodds/Wäch e , 1993). The
agency o icially e mina ed i s ope a ion a he end o 1994 and i s mission was aken o e by a
successo agency en i led Bundesans al ü e einigungsbeding e Sonde au gaben (Böick, 2018).
24
IAB‑Discussion Pape 01|2024
The igh panel o Figu e 6 p o ides he ull dis ibu ion o employmen g ow h
acco ding o he a io o ini ial size o e inal employmen a ge . A s iking nega i e
ela ionship eme ges be ween he dis ance o he inal a ge and subsequen
employmen g ow h. Fi ms ha ha e high a ge s ela i e o hei ini ial size g ow hei
wo k o ce signi ican ly mo e han i ms wi h a ge s close o hei ini ial size. Fi ms
wi h lax a ge s ela i e o hei ini ial employmen had leeway o adjus and
subsequen ly sh unk signi ican ly. O e all, he igu e sugges s he impo ance o i m
employmen a ge s as a de e minan o i m employmen policies.
Table 3 p esen s he es ima es o he labo g ow h equa ion. Columns (1) o (3)
p o ide OLS es ima es, while columns (4) o (6) p o ide IV es ima es. S anda d e o s
a e wo‑way clus e ed a he p i a ize and o ice le el. Columns (1) o (3) sugges ha
i m g ow h is posi i ely co ela ed wi h binding labo commi men s. Condi ional on
he se o baseline con ols, indus y dummies, and p i a ize cha ac e is ics, he
associa ion be ween binding con ac s and employmen g ow h is on a e age
49 pe cen poin s un il he inal commi men da e. The es ima ed OLS coe icien is
una ec ed by he inclusion o addi ional con ol a iables o hose in he baseline
speci ica ion.
IV es ima es in columns (4) o (6) sugges ha he causal e ec o binding labo a ge s
is signi ican ly la ge wi h espec o OLS es ima es. In hese speci ica ions, i ms a e
es ima ed o g ow hei wo k o ce by 68 pe cen poin s in he h ee yea s ollowing he
andom assignmen o a binding labo con ac . The e ec is no only economically
la ge bu also p ecisely es ima ed a he 1 pe cen signi icance le el. Consis en wi h
he p e ious e idence, he i s ‑s age s a is ics o weak ins umen s is la ge (Panel B).
The Kleinbe gen‑Paap F‑ es s ejec he hypo hesis o weak ins umen s wi h s a is ics
anging be ween 15 and 17. Economically, he i s ‑s age es ima es imply ha a
10 pe cen inc ease in labo p e e ences o p i a ize s esul in a 2 pe cen poin s
highe likelihood o a binding labo con ac . Consis en wi h he quasi‑ andom
assignmen mechanism, he inclusion o addi ional co a ia es does no a ec he
i s ‑s age coe icien s.25
25 Table A.7 p o ides a way o es whe he he di e ences be ween he OLS and he IV es ima es a e
d i en by ea men e ec he e ogenei y. We ollow Bhulle e al. (2020) and i s pe o m a p incipal
componen analysis using one componen based on p e‑de e mined employmen (see employmen
ca ego ies in Table 2) and e enue igu es measu e in 1990, as well as ini ial employmen a con ac
da e, and he sec o a ilia ion. We hen sepa a e he p edic ed componen in o qua iles and
sepa a ely es ima e he complie sha e o each qua ile g oup using he i s ‑s age eg ession
speci ica ion. Finally, we e‑weigh he ull es ima ion sample by using he sub‑sample complie
sha es as weigh s. Panel B o Table A.7 shows ha e‑weigh ing based on obse ed cha ac e is ics
inc eases he OLS es ima e sligh ly om 0.49 o 0.52. The di e ence be ween he e‑weigh ed OLS
and IV es ima e, howe e , emains s a k. This sugges s ha e ec he e ogenei y is unlikely o explain
he di e ences.
31
IAB‑Discussion Pape 01|2024

Figu e 6: Employmen Dynamics by Ini ial Size o Ta ge
A: Employmen G ow h
0 1 2 3
Densi y
-2 -1 0 1 2
Employmen g ow h dis ibu ion
Ini ial size abo e a ge Ini ial size below age
G ow h dis ibu ion
-.5 0 .5
Employmen g ow h
.2 .6 1 1.4 1.8 2.2
Ra io Ini ial Size o Final Ta ge
A e age g ow h a e 90% Con idence In e als
No es: Panel A shows he o e all employmen g ow h dis ibu ion as well as he employmen g ow h
dis ibu ion dis inguishing by i ms ini ially below o abo e (including i ms ini ially a hei a ge ) hei
commi men employmen le el. Panel B shows a e age g ow h a es by he dis ance o he ini ial size
o he inal a ge .
Sou ce: ISUD.
The esul s a e obus o a se ies o al e na i e speci ica ions. A challenge o ou
iden i ica ion s a egy is ha p i a ize decisions a e mul idimensional. The exclusion
es ic ion equi es ha p i a ize s a ec i ms’ ou comes only h ough binding labo
commi men s. THA p i a ize s, howe e , nego ia ed no only on labo commi men s,
bu also on associa ed penal ies, in es men commi men s, and sales p ice. Following
Bhulle e al. (2020), we add ess his issue by augmen ing ou baseline model wi h
con ols o hese dimensions o p i a iza ion con ac s. Consis en wi h he exclusion
32
B: Employmen G ow h by Ini ial Size o Ta ge
IAB‑Discussion Pape 01|2024
Table 3: Reg ession Resul s, Employmen G ow h
OLS Model IV‑Model
(1) (2) (3) (4) (5) (6)
Panel A: Second‑s age esul s
Binding con ac 0.4992*** 0.4973*** 0.4975*** 0.7016*** 0.6740*** 0.6873***
(0.031) (0.030) (0.030) (0.219) (0.231) (0.235)
Panel B: Fi s ‑s age esul s
P i a ize s ingency 0.0020*** 0.0018*** 0.0018***
(0.000) (0.000) (0.000)
Obse a ions 9,363 9,363 9,363 9,363 9,363 9,363
A e age employmen a
con ac da e
60.064 60.064 60.064 60.064 60.064 60.064
A e age g ow h a e
(non‑binding con ac s)
‑0.063 ‑0.063 ‑0.063 ‑0.063
Sha e wi h binding con ac s 0.207 0.207 0.207 0.207 0.207 0.207
F‑S a is ic 17.01 14.67 14.76
Sample condi ion
Baseline con ols Yes Yes Yes Yes Yes Yes
Indus y con ols No Yes Yes No Yes Yes
Indi idual con ols No No Yes No No Yes
No es: The able shows OLS and IV eg ession esul s o employmen g ow h on binding con ac s.
Panel A shows he educed o m eg essing he binding con ac indica o on he s ingency measu e.
Panel B shows he second‑s age esul s. All speci ica ions con ol o ully in e ac ed THA agency and
yea ixed e ec s and a e condi ional on ha ing a leas i e p i a iza ions pe p i a ize . F‑S a is ic
e e s o he Kleibe gen‑Paap F‑S a is ic. Baseline con ols a e ime be ween he i s and las audi s
measu ed in mon hs, ime be ween con ac da e and i s audi measu ed in mon hs, and log ini ial
employmen le el measu ed a he i s audi . Indus y con ols a e 2‑digi indus y dummies.
Indi idual con ols e e o he gende o he p i a ize and a dummy o a PhD deg ee. S anda d
e o s a e wo‑way clus e ed a p i a ize and THA o ice le el. Ins umen e e s o he lea e‑one‑ou
measu e o assigning binding con ac s. *p<0.1, **p<0.05, ***p<0.01.
Sou ce: ISUD.
es ic ion, Table A.4 shows ha adding ex ensi e and in ensi e in es men p e e ences
does no a ec ou baseline esul s quali a i ely and, u he , does no ha e any
explana o y powe in he i s and second s ages. In Panel A o Table A.5, we also
con ol o subsequen enego ia ion a emp s ini ia ed by buye s. Panel B a ies he
sample acco ding o he numbe o cases handled and he cons uc ion o he
ins umen . Panel C es ima es ou model using al e na i e measu es o i m g ow h.
Finally, Table A.6 add esses po en ial sample selec ion o GDR i ms in o labo
commi men s by including he es ima ed in e se mills a io om a Heckman model.
Again, es ima es a e una ec ed.26
26 The Heckman selec ion equa ion is based on a p obi eg ession wi h he ou come a iable being
equal o 1 i he ini ial GDR i m is obse ed among he con ac s wi h labo commi men s and ze o
o he wise. We use as explana o y a iables log ini ial i m size and log ini ial sales o e employmen
measu ed in 1990 as well as sec o ‑ and THA o ice ixed e ec s. Resul s o employmen g ow h,
p oduc i i y g ow h and i m exi in A.6.
33
IAB‑Discussion Pape 01|2024
5.3 Labo Commi men s and P oduc i i y G ow h
To disen angle he mechanism behind he g ow h in employmen we ex end he
empi ical analysis o p oduc i i y dynamics in he ma ched sample o 2,395
p i a iza ion con ac s wi h comple e in o ma ion. We cons uc he model‑consis en
measu e o p oduc i i y as sales/employmen α
, aking in o accoun he labo sha e in
he p oduc ion. In he baseline analysis, we se α = 0.8, consis en wi h he agg ega e
labo sha e du ing his pe iod. We also cons uc measu es o p oduc i i y based on
TFP by ma ching he con ac s o a su ey ha con ains in o ma ion on THA i m
in es men s. This allows us o ob ain he associa ed capi al s ock o he i m and
es ima e a Cobb‑Douglas p oduc ion unc ion.
Figu e 7 desc ibes he ela ionship be ween p oduc i i y g ow h and he a io o ini ial
employmen ela i e o he inal a ge . The igu e plo s local linea eg essions on bo h
sides o he e ical line sepa a ing ini ially binding and non‑binding con ac s. The
igu e p o ides wo majo insigh s. Fi s , g ow h in p oduc i i y is ela i ely cons an o
i ms abo e he h eshold o binding con ac s. The a e age g ow h a e in he da a
amoun s o 86.8 pe cen which indica es subs an ial imp o emen s in p oduc i i y
du ing he i s yea s a e euni ica ion. Second, p oduc i i y g ow h is signi ican ly
highe o i ms ini ially below hei commi ed employmen .
Table 4 p o ides OLS and IV es ima es o p oduc i i y g ow h ollowing he same
g ow h a e o mula as o employmen . The speci ica ion con ols again o ully
in e ac ed THA o ice and yea ixed e ec s, indus y dummies, log ini ial employmen ,
log ini ial p oduc i i y, he pu chasing p ice and he ime be ween he i s and las
audi s as well as be ween he con ac da e and he i s audi .
Columns (1) and (2) o Table 4 p o ide OLS e idence ha i ms wi h binding labo
con ac s expe ience highe p oduc i i y g ow h o a ound 8 o 9 pe cen poin s. Column
(2) adds con ols o decile dummies o he pu chasing p ice. Column (3) p o ides IV
e idence on p oduc i i y g ow h. To do so, we implemen a wo‑sample 2SLS es ima ion
by using he p edic ed alues om he i s ‑s age eg ession o he ull sample in he
second‑s age eg ession o he sub‑sample ha includes in o ma ion on p oduc i i y
g ow h. To calcula e he s anda d e o s, we pe o m 2,500 boo s ap eplica ions
p esen ed in pa en heses. O e he cou se o i e yea s measu ed be ween July 1990
and he end o he commi men pe iod, on a e age, i ms wi h binding labo con ac s
expe ience a o al p oduc i i y g ow h o oughly 73 pe cen poin s. Columns (4) o (6)
p esen ou esul s based on he TFP g ow h measu e. Again, we ind ha TFP inc eased
by 66 pe cen poin s mo e o i ms wi h binding con ac s o e he commi men pe iod.
As in he employmen g ow h eg essions, he OLS es ima es o p oduc i i y display a
downwa d bias wi h espec o he IV es ima es. We p o ide ex ensi e obus ness checks
34 IAB‑Discussion Pape 01|2024
Figu e 7: P oduc i i y G ow h and he Deg ee o Binding Con ac s
linea i p ojec ion
.5 .6 .7 .8 .9 1 1.1 1.2
G ow h in P oduc i i y
.4 .5 .6 .7 .8 .9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
Ra io Ini ial Employmen Size o Final Commi men
No es: The igu e plo s he g ow h in p oduc i i y be ween he ini ial yea o he con ac and he inal
commi men yea agains he a io o ini ial employmen ela i e o he inal commi men le el.
Con ac s below 1 ha e ini ially lowe employmen han commi ed. Con ac s abo e 1 ha e ini ially
highe employmen han commi ed. The plo ed alues in he local linea eg ession a e
mean‑s anda dized esiduals om a eg ession on ini ial labo p oduc i i y, employmen and
indus y‑ ixed e ec s. The wo g ey dashed lines co espond o he 90 pe cen CI. The blue line shows
a linea i o a eg ession o p oduc i i y g ow h on he a io o ini ial size o inal commi men
among con ac s o he le o o a 1. The ed line p ojec s he linea i in o he a ea whe e he
ini ial size is below he commi ed le el ( o he le o 1). The igu e excludes op and bo om
4 pe cen o he igh ness measu e. To al numbe o i ms is 2,395.
Sou ce: ISUD, MUP.
o ou p oduc i i y g ow h esul s in Appendix Tables A.9 and A.10. Amongs o he s, we
demons a e ha he es ima es emain obus when a ying he labo sha e pa ame e ,
and when including addi ional con ols such as pu chasing p ice, he p esence o
in es men a ge s, and he inclusion o exi ing i ms.
Finally, Table A.8 p o ides suppo ing e idence o he p oduc i i y channel by analyzing
pa en ing ac i i y. The ou come a iable is equal o 1 i he i m has a leas one pa en
du ing he commi men pe iod and 0 o he wise. OLS esul s show a posi i e and
signi ican associa ion be ween binding con ac s and pa en ing. Al hough imp ecisely
es ima ed, he 2S2SLS coe icien shows again a downwa d bias o he OLS poin es ima e.
5.4 Labo Commi men s and Ma ke Exi
We now u n o he analysis be ween binding labo commi men s and ma ke exi s. To
measu e i m exi we again use he me ged sample o con ac s o he MUP da a. We
also use a second measu e o exi based on he inal labo audi epo ing 0 wo ke s.
35 IAB‑Discussion Pape 01|2024
Table 4: Reg ession Resul s, P oduc i i y G ow h
P oduc i i y, α = 0.8 TFP
OLS OLS 2S2SLS OLS OLS 2S2SLS
(1) (2) (3) (4) (5) (6)
Binding con ac 0.0965*** 0.0835*** 0.7109** 0.1107*** 0.1239*** 0.6608***
(0.022) (0.023) (0.363) (0.039) (0.041) (0.213)
Obse a ions 2,395 2,395 1,612 1,825 1,825 1,825
A e age p oduc i i y a
con ac da e
10.599 10.599 10.633 6.813 6.813 6.813
A e age p oduc i i y
g ow h (non‑binding)
0.851 0.851 0.854 0.332 0.332 0.332
Sha e wi h binding
con ac s
0.171 0.171 0.155 0.162 0.162 0.162
Baseline con ols Yes Yes Yes Yes Yes Yes
Indus y con ols Yes Yes Yes Yes Yes Yes
Indi idual con ols Yes Yes Yes Yes Yes Yes
Pu chasing p ice No Yes Yes No Yes Yes
No es: The able shows OLS and 2S2SLS eg ession esul s o measu es o p oduc i i y g ow h on
binding con ac s. All speci ica ions con ol o ully in e ac ed THA agency and yea ixed e ec s.
Binding con ac s a e de ined as ini ial i m size below he commi ed a ge le el. Con ols a e as in
he baseline speci ica ion. Addi ional con ols a e log ini ial p oduc i i y and he pu chasing p ice.
S anda d e o s in columns (1)‑(3) a e wo‑way clus e ed a p i a ize and THA o ice le el. The
s anda d e o s in columns (4)‑(6) a e boo s apped using 2,500 eplica ions. *p<0.1, **p<0.05,
***p<0.01.
Sou ce: ISUD, MUP.
Figu e 8 desc ibes he ela ionship be ween he exi p obabili y o i ms and he a io o
ini ial employmen ela i e o he inal a ge . Simila o he p oduc i i y g ow h
pa e ns, he sha e o i ms exi ing he ma ke is ela i ely cons an o non‑binding
i ms abo e he e ical line o 1. The exi a e o hese i ms amoun s o 4.8 pe cen
on a e age. Wi h an a e age exi a e o 8.4 pe cen , i ms wi h binding con ac s show
a highe le el o ma ke exi ha is inc easing in he igh ness measu e.
Table 5 p o ides a eg ession e sion o Figu e 8, con olling again o he same
a iables as be o e. The i s h ee columns p o ide he esul s using he MUP exi
indica o , whe eas he las h ee columns a e based on ze o employmen in he ISUD
da a (condi ional on he same sample). Bo h eg ession speci ica ions gene a e simila
OLS esul s. Binding con ac s a e associa ed wi h an inc ease in ma ke exi o a ound
2.5 pe cen poin s. The IV es ima ion, al hough wi h lowe p ecision, con i ms he
highe p opensi y o exi o i ms wi h binding labo commi men s. These esul s a e
consis en wi h he model’s p edic ion ha lowe p o i s associa ed wi h he penal ies
lead i ms o exi a a highe a e.
36 IAB‑Discussion Pape 01|2024

Figu e 8: Ma ke Exi and he Deg ee o Binding Con ac s
linea i p ojec ion
0 .05 .1 .15 .2 .25 .3
Ma ke Exi P obabili y
.4 .6 .8 1 1.2 1.4 1.6 1.8 2
Ra io Ini ial Employmen Size o Final Commi men
No es: The igu e plo s ma ke exi a es agains he a io o ini ial employmen ela i e o he inal
commi men le el. Con ac s below 1 ha e ini ially lowe employmen han commi ed. Con ac s
abo e 1 ha e ini ially highe employmen han commi ed. The plo ed alues in he local linea
eg ession a e mean‑s anda dized esiduals om a eg ession on ini ial employmen and indus y‑ ixed
e ec s. The wo g ey dashed lines co espond o he 90 pe cen CI. The blue line shows a linea i o
a eg ession o ma ke exi on he a io o ini ial size o inal commi men among con ac s o he le
o o a 1. The ed line p ojec s he linea i in o he a ea whe e he ini ial size is below he
commi ed le el ( o he le o 1). The igu e excludes op and bo om 3 pe cen o he igh ness
measu e. To al numbe o i ms is 4,596.
Sou ce: ISUD, MUP.
Table 5: Reg ession Resul s, Exi P obabili y a Final Commi men
MUP exi indica o ISUD 0 employmen
OLS 2S2SLS OLS 2S2SLS
(1) (2) (3) (4) (5) (6)
Binding con ac 0.0262** 0.0248** 0.1302 0.0216* 0.0193* 0.0358*
(0.011) (0.012) (0.108) (0.011) (0.010) (0.021)
Obse a ions 4,563 4,563 2,804 4,563 4,563 2,804
Exi sha e (non‑binding) 0.049 0.049 0.047 0.035 0.035 0.030
Sha e wi h binding con ac s 0.171 0.171 0.171 0.171 0.171 0.171
Sample condi ion
Baseline con ols Yes Yes Yes Yes Yes Yes
Indus y con ols Yes Yes Yes Yes Yes Yes
Indi idual con ols Yes Yes Yes Yes Yes Yes
Pu chasing p ice No Yes Yes No Yes Yes
No es: The able shows OLS and 2S2SLS eg ession esul s o exi ing p obabili ies a he end o he
commi men pe iod. The ou come a iable akes he alue o 1 i he i m is exi ing by he end o he
commi men pe iod and 0 o he wise. All speci ica ions con ol o ully in e ac ed THA agency and yea
ixed e ec s. Binding con ac s a e de ined as ini ial i m size below he commi ed a ge le el. Con ols
a e as in he baseline speci ica ion. Addi ional con ol a iable is he pu chasing p ice. S anda d e o s
in columns (1), (2), (4) and (5) a e wo‑way clus e ed a p i a ize and THA o ice le el. The s anda d
e o s in columns (3) and (6) a e boo s apped using 2,500 eplica ions. *p<0.1, **p<0.05, ***p<0.01.
Sou ce: ISUD, MUP.
37
IAB‑Discussion Pape 01|2024
6 Quan i a i e Analysis
In his sec ion, we p esen he calib a ion o he model using i m‑le el da a and
p o ide se e al coun e ac ual analyses o quan i y he a ious channels by which he
binding employmen a ge s impac i m dynamics.
6.1 Calib a ion
We s a by se ing some o he pa ame e alues ex e nally. We choose he labo sha e
pa ame e in he p oduc ion unc ion, α, equal o 0.8 o ma ch he labo ea ning sha e.
Consis en wi h he a e age con ac leng h o h ee yea s in he da a, we se he
a i al a e o con ac expi a ion, µ, o 1/3. Annual wage g ow h a e is se o
10 pe cen o ma ch he a e age eal wage g ow h a e o e 1990 and 1996 in Eas
Ge many (Hun , 2001). The es o he pa ame e s a e calib a ed in e nally by
minimizing he dis ance be ween he momen s om he i m‑le el da a we used in he
empi ical pa o he pape and hei model implied coun e pa s.27 In pa icula , le
ME deno e he ec o o empi ical momen s and le M(Ω) deno e he ec o o
model‑simula ed momen s and Ω is he se o pa ame e s o be calib a ed in e nally.
We hen sea ch Ω o minimize he absolu e ela i e de ia ion be ween he model and
da a; ha is, we sol e
∑ |ME − Mm(Ω)|
m
min .
Ω |ME
|
m
m
We use he poin es ima es o he e ec o binding con ac s on employmen g ow h
and p oduc i i y g ow h p esen ed in Sec ion 5 o discipline he cos o no hi ing he
a ge , γ. We u he use eg ession esul s on he impac o binding con ac s on exi
a es o i ms o in o m he exi cos pa ame e , Ce. Finally, we include he g ow h a e
o o al employmen o i ms wi h binding con ac s o e he commi men pe iod o
pin down he in es men cos pa ame e , ϕ.
We use he ollowing p ocedu e o calib a e he model: Fo gi en alues o pa ame e s,
we i s sol e he alue unc ion in equa ion 3 and use he implied op imal decisions o
simula e a coho o i ms. We ini ialize he coho by using he sample o i ms used in
he empi ical pa o he pape and ake he employmen a ge as gi en in he da a.
C ucially, each i m is simula ed in line wi h he ime span om i s ini ial audi o i s
27 In ou se ing, we canno sepa a ely iden i y he s ep size and he cos scale pa ame e in
p oduc i i y imp o emen s, λ and ϕ, espec i ely. The e o e, we ix he alue o he s ep size a 0.25
and calib a e he cos scale pa ame e in e nally.
38
IAB‑Discussion Pape 01|2024
inal audi . Finally, we use he simula ed da a o cons uc he a ge ed momen s. We
epea his p ocess and sea ch o e he pa ame e space un il we minimize he dis ance
be ween model‑implied momen s and he da a.
6.2 Calib a ion Resul s and Goodness o Fi
Table 6 and 7 con ain he calib a ed pa ame e s and he a ge ed momen s,
espec i ely. As seen om Table 6, he model is able o eplica e he a ge ed momen s
well. In pa icula , we we e able o i highe employmen and p oduc i i y g ow h o
i ms wi h binding con ac s wi h a ela i ely pa simonious model. Ou calib a ion
sugges s ha o e e y missing employee ela i e o he commi ed labo a ge , i ms
pay a ine ha co esponds o 68 pe cen o he a e age wage, gi en by γ.
Table 6: Momen s Used in Calib a ion
# Desc ip ion Model Da a
M1 Employmen g ow h eg ession 0.489 0.498
M2 P oduc i i y g ow h eg ession 0.083 0.083
M3 Exi a e eg ession 0.030 0.027
M4 To al empl. g ow h a e o i ms wi h binding con ac s 0.614 0.672
Table 7: In e nally Calib a ed Pa ame e s
Desc ip ion Model Es ima e
Penal y o no hi ing a ge employmen γ 0.676
Scale o in es men cos pa ame e ϕ 0.030
Cos o exi Ce 58.39
Non‑ a ge ed Momen s
The calib a ed model also pe o ms well in ma ching some impo an pa e ns in he
da a ha we e no a ge ed. In Figu e 9, we depic he employmen g ow h by he a io
o ini ial employmen ela i e o a ge employmen , analogous o Panel B o Figu e 6.
The black and ed do s show he model‑implied employmen g ow h a es and he
da a, espec i ely. Al hough we only a ge he a e age excess g ow h a e o
binding‑con ac i ms (employmen eg ession coe icien ) in he calib a ion, ou model
success ully ma ches employmen g ow h a es ac oss he en i e ange o
employmen ‑ o‑ a ge a ios.
The calib a ed model also cap u es he pos ‑commi men employmen dynamics ai ly
well. Figu e 10 illus a es he e olu ion o o al employmen du ing and a e he
39 IAB‑Discussion Pape 01|2024
commi men pe iod bo h in he model and da a. The esul s sugges ha no only do
hese i ms wi h binding con ac s expe ience highe employmen g ow h du ing he
commi men pe iod, bu hese employmen gains a e pe sis en a leas six yea s
subsequen o he commi men pe iod. This pe sis en employmen e ec is consis en
wi h he dynamic p oduc i i y gains implied by binding labo a ge s in he model.
Figu e 9: Employmen G ow h
No es: The igu e depic s he employmen g ow h a he i m le el by he a io o ini ial employmen
ela i e o a ge employmen . The black and ed do s show he model‑implied employmen g ow h
a es and he da a, espec i ely. G ay do s show he employmen g ow h a es unde he
coun e ac ual economy whe e he e a e no employmen a ge s. The x‑axis is di ided in o 20 quan ile
bins and each do ep esen s a e age alue wi hin ha bin.
Figu e 10: To al Employmen in he Pos ‑Commi men Pe iod
No es: The igu e shows he e olu ion o o al employmen du ing and a e he commi men pe iod
bo h in he model (black lines) and da a ( ed lines). Dashed and solid lines show he o al
employmen o binding and no binding i ms, espec i ely. All se ies a e no malized o 1 a he
beginning o he policy pe iod.
40
IAB‑Discussion Pape 01|2024
Gio celli, Michela; Li, Bo (2021): Technology ans e and ea ly indus ial de elopmen :
e idence om he Sino‑So ie alliance. Tech. Rep., Na ional Bu eau o Economic
Resea ch.
Gli z, Alb ech ; Meye sson, E ik (2020): Indus ial espionage and p oduc i i y. In:
Ame ican Economic Re iew, Vol. 110, No. 4, p. 1055–1103.
G eens one, Michael; Ho nbeck, Richa d; Mo e i, En ico (2010): Iden i ying
agglome a ion spillo e s: E idence om winne s and lose s o la ge plan openings.
In: Jou nal o Poli ical Economy, Vol. 118, No. 3, p. 536–598.
Ha ison, Ma k; Zaksauskienė, Inga (2016): Coun e ‑in elligence in a command economy.
In: The Economic His o y Re iew, Vol. 69, No. 1, p. 131–158.
Heise, Sebas ian; Po zio, Tommaso (2021): The agg ega e and dis ibu ional e ec s o
spa ial ic ions. Tech. Rep., Na ional Bu eau o Economic Resea ch.
Hsieh, Chang‑Tai; Klenow, Pe e J (2009): Misalloca ion and manu ac u ing TFP in China
and India. In: Qua e ly Jou nal o Economics, Vol. 124, No. 4, p. 1403–1448.
Hun , Jenni e (2006): S aunching emig a ion om Eas Ge many: Age and he
de e minan s o mig a ion. In: Jou nal o he Eu opean Economic Associa ion,
Vol. 4, No. 5, p. 1014–1037.
Hun , Jenni e (2001): Pos ‑uni ica ion wage g ow h in Eas Ge many. In: Re iew o
Economics and S a is ics, Vol. 83, No. 1, p. 190–195.
Johnson, Paul; Papageo giou, Ch is (2020): Wha emains o c oss‑coun y con e gence?
In: Jou nal o Economic Li e a u e, Vol. 58, No. 1, p. 129–75.
Juhász, Réka; Lane, Na han J; Rod ik, Dani (2023): The New Economics o Indus ial
Policy. Tech. Rep., Na ional Bu eau o Economic Resea ch.
Kaloup sidi, My o (2018): De ec ion and impac o indus ial subsidies: The case o
Chinese shipbuilding. In: The Re iew o Economic S udies, Vol. 85, No. 2, p.
1111–1158.
Kline, Pa ick; Mo e i, En ico (2014): Local economic de elopmen , agglome a ion
economies, and he big push: 100 yea s o e idence om he Tennessee Valley
Au ho i y. In: The Qua e ly Jou nal o Economics, Vol. 129, No. 1, p. 275–331.
K uege , Alan B; Pischke, Joe n‑S e en (1995): A Compa a i e Analysis o Eas and Wes
Ge man Labo Ma ke s: Be o e and A e Uni ica ion. In: in: F eeman und Ka z:
Di e ences and Changes in Wage S uc u es„ p. 405–446.
Lane, Na han (2022): Manu ac u ing e olu ions: Indus ial policy and indus ializa ion
in Sou h Ko ea. In: A ailable a SSRN 3890311.
Lipschi z, Leslie; McDonald, Donogh (1990): Ge man uni ica ion: economic issues. Tech.
Rep., In e na ional Mone a y Fund Washing on.
47 IAB‑Discussion Pape 01|2024

Liu, E nes (2019): Indus ial policies in p oduc ion ne wo ks. In: The Qua e ly Jou nal
o Economics, Vol. 134, No. 4, p. 1883–1948.
López‑de Silanes, Flo encio (1997): De e minan s o p i a iza ion p ices. In: The
Qua e ly Jou nal o Economics, Vol. 112, No. 4, p. 965–1025.
Ma in, Leslie A; Na a aj, Shan hi; Ha ison, Ann E (2017): In wi h he big, ou wi h he
small: Remo ing small‑scale ese a ions in India. In: Ame ican Economic Re iew,
Vol. 107, No. 2, p. 354–86.
Megginson, William L; Ne e , Je y M (2001): F om s a e o ma ke : A su ey o
empi ical s udies on p i a iza ion. In: Jou nal o economic li e a u e, Vol. 39, No. 2,
p. 321–389.
Me gele, Lukas; Hennicke, Mo i z; Lubczyk, Mo i z (2020): The big sell: p i a izing Eas
Ge many’s economy. In: .
Pe e s, Michael (2022): Ma ke size and spa ial g ow h—e idence om Ge many’s
pos ‑wa popula ion expulsions. In: Econome ica, Vol. 90, No. 5, p. 2357–2396.
Redding, S ephen J; S u m, Daniel M (2008): The cos s o emo eness: E idence om
Ge man di ision and euni ica ion. In: Ame ican Economic Re iew, Vol. 98, No. 5, p.
1766–1797.
Res uccia, Diego; Roge son, Richa d (2008): Policy Dis o ions and Agg ega e
P oduc i i y wi h He e ogeneous Es ablishmen s. In: Re iew o Economic Dynamics,
Vol. 11, No. 4, p. 707–720.
Romano, Joseph P; Wol , Michael (2005a): Exac and app oxima e s epdown me hods
o mul iple hypo hesis es ing. In: Jou nal o he Ame ican S a is ical Associa ion,
Vol. 100, No. 469, p. 94–108.
Romano, Joseph P; Wol , Michael (2005b): S epwise mul iple es ing as o malized da a
snooping. In: Econome ica, Vol. 73, No. 4, p. 1237–1282.
Sampa , Bha en; Williams, Heidi L (2019): How do pa en s a ec ollow‑on inno a ion?
E idence om he human genome. In: Ame ican Economic Re iew, Vol. 109, No. 1,
p. 203–36.
Shlei e , And ei (1998): S a e e sus p i a e owne ship. In: Jou nal o Economic
Pe spec i es, Vol. 12, No. 4, p. 133–150.
Siebe , Ho s (1991): Ge man uni ica ion: he economics o ansi ion. In: Economic
Policy, Vol. 6, No. 13, p. 287–340.
Sinn, Hans‑We ne (2002): Ge many’s economic uni ica ion: An assessmen a e en
yea s. In: Re iew o in e na ional Economics, Vol. 10, No. 1, p. 113–128.
Snowe , Dennis J; Me kl, Ch is ian (2006): The ca ing hand ha c ipples: The Eas
Ge man labo ma ke a e euni ica ion. In: Ame ican Economic Re iew, Vol. 96,
No. 2, p. 375–382.
48 IAB‑Discussion Pape 01|2024
Sy e son, Chad (2011): Wha de e mines p oduc i i y? In: Jou nal o Economic
li e a u e, Vol. 49, No. 2, p. 326–65.
Uhlig, Ha ald (2008): The slow decline o Eas Ge many. In: Jou nal o Compa a i e
Economics, Vol. 36, No. 4, p. 517–541.
Wahse, Jü gen; Dahms, Ve a; Schä e , Reinha d; Kühl, Jü gen (1996):
Beschä igungspe spek i en on p i a isie en Un e nehmen. In: Mi eilungen aus
de A bei sma k ‑und Be u s o schung, Vol. 29, No. 1, p. 106–116.
49 IAB‑Discussion Pape 01|2024
Supplemen a y Appendix
A Fu he Empi ical Resul s
Figu e A.1: P i a iza ions pe P i a ize
Numbe o p i a ize s: 1659
Numbe o con ac s: 11194
0 .1 .2 .3
Pe cen
1 5 10 15 20 25+
Numbe o obse a ions pe p i a ize
No es: The igu e plo s he numbe o p i a iza ion handled pe indi idual p i a ize (winso ized a 25).
The o al numbe o p i a iza ions is 11,194. These cases a e handled by 1,659 indi iduals.
5.04 pe cen o all cases a e o ganized by p i a ize s only obse ed once in he sample. This
co esponds o 652 indi iduals.
Sou ce: ISUD.
50
IAB‑Discussion Pape 01|2024
Table A.1: Tes o Random Assignmen o In es o s o P i a ize s
Indep. a iable: S ingency Dep. a iables
Coe icien p‑ alue Adj. p‑ alue Mean S anda d
de ia ion
(1) (2) (3) (4) (5)
Employmen
Log in es o size ‑0.0033 0.1228 0.9740 2.4600 1.8220
In es o size > 100 employees ‑0.0008 0.1135 0.9181 0.1400 0.3480
C edi a ing
C edi wo hiness in es o 0.0768 0.3334 0.9990 284.38 101.58
High a ing ‑0.0004 0.1101 0.9800 0.0640 0.2460
Loca ion
Wes Ge man in es o 0.0001 0.8356 0.9990 0.6780 0.4680
Sec o a ilia ion
Ag icul u e, o es y, ishing 0.0002 0.5666 0.9930 0.0120 0.1120
Mining and qua ying ‑0.0001 0.4517 0.9980 0.0040 0.0700
Manu ac u ing ‑0.0006 0.1441 0.9860 0.1740 0.3800
Ene gy ‑0.0003 0.0505 0.0859 0.0040 0.0680
Wa e 0.0001 0.7685 0.9990 0.0380 0.1940
Cons uc ion ‑0.0005 0.3395 0.9860 0.1260 0.3320
Re ail ade 0.0012 0.0113 0.5375 0.2000 0.4000
T anspo a ion ‑0.0003 0.2162 0.9940 0.0540 0.2280
Hospi ali y 0.0001 0.3576 0.9990 0.0400 0.1940
ICT ‑0.0001 0.6517 0.9990 0.0420 0.2000
Baning and Insu ance 0.0000 0.9975 1.0000 0.0320 0.1760
Real Es a e 0.0001 0.7775 0.9990 0.0580 0.2340
Technical se ices ‑0.0003 0.5271 0.9980 0.1020 0.3040
Economic se ices 0.0002 0.2320 0.9940 0.0340 0.1800
O he 0.0003 0.6932 0.9940 0.0760 0.2640
No es: The sample is based on 4,993 con ac s ma ched o in es o cha ac e is ics. Each line ep esen s
a single eg ession o he explana o y a iable on he s ingency measu e ha akes alues be ween 0
(minimum) and 100 (maximum) con olling o THA o ice and yea o p i a iza ion ixed e ec s.
S anda d e o s a e wo‑way clus e ed a p i a ize and THA o ice le el. p‑ alues in column (2)
co espond o he eg ession model and a e wo‑way clus e ed a he p i a ize and THA o ice le el.
p‑ alues in column (3) adjus o mul iple es ing using Romano‑Wol p ocedu e (Romano/Wol , 2005a;
Romano/Wol , 2005b) wi h 1,000 boo s ap eplica ions. **p<0.1, **p<0.05, ***p<0.01.
Sou ce: ISUD.
51
IAB‑Discussion Pape 01|2024
Table A.2: Fi s ‑S age Reg ession Resul s by Sub‑Samples
Baseline Employmen in 1990 Re enue in 1990 Sec o a ilia ion
< p(75) < p(50) < p(75) < p(50) T ade‑
able
Non‑
adeable
(1) (2) (3) (4) (5) (6) (7)
P i a ize s ingency 0.0015*** 0.0011** 0.0015*** 0.0012** 0.0012* 0.0016*** 0.0009*
(0.000) (0.001) (0.000) (0.000) (0.001) (0.000) (0.000)
Obse a ions 10,616 6,077 6,545 5,985 3,982 5,230 3,003
A e age employmen
a con ac da e
63.22 57.75 73.638 72.39 70.166 69.032 70.554
A e age g ow h a e .062 .028 .084 .038 .038 .048 .006
Sha e wi h binding
con ac s
.208 .184 .232 .194 .194 .22 .146
Sample condi ion
Baseline con ols Yes Yes Yes Yes Yes Yes Yes
Indi idual con ols Yes Yes Yes Yes Yes Yes Yes
Indus y con ols Yes Yes Yes Yes Yes Yes Yes
No es: The able shows IV eg ession esul s. All speci ica ions con ol o ully in e ac ed THA agency
and yea ixed e ec s and a e condi ional on ha ing a leas 2 p i a iza ions pe p i a ize . All s a a
a iables (e.g., employmen in 1990) e e o he ini ial i m om whe e he con ac was gene a ed.
The e a e 335 con ac s a ilia ed wi h he ag icul u e sec o no p esen ed in he able. F‑S a is ic
e e s o he Kleibe gen‑Paap F‑S a is ic. Baseline con ols a e he ime be ween he i s and las
audi s measu ed in day, ime be ween con ac da e and i s audi measu ed in days, and log ini ial
employmen le el (+1) measu ed a he i s audi . Indi idual con ols a e he gende o he p i a ize
and academic deg ee (PhD). Indus y con ols a e 2‑digi indus y dummies. S anda d e o s a e
wo‑way clus e ed a p i a ize and THA o ice le el. Ins umen e e s o he lea e‑one‑ou measu e o
assigning binding con ac s. **p<0.1, **p<0.05, ***p<0.01.
Sou ce: ISUD.
52
IAB‑Discussion Pape 01|2024

Table A.3: Tes o Join Null o Mono onici y and Exclusion
10 kno s 15 kno s
ω = 1 ω = 0.8 ω = 0.5 ω = 0.3 ω = 1 ω = 0.8 ω = 0.5 ω = 0.3
(1) (2) (3) (4) (5) (6) (7)
Tes s a is ic 535 535 535 535 484 484 484 484
d. . (509) (509) (509) (509) (504) (504) (504) (504)
P‑ alue [0.204] [0.255] [0.408] [0.680] [0.733] [0.916] [1.000] [1.000]
No es: The able p esen s esul s om he es p oposed in F andsen/Le g en/Leslie (2023) o he join
null hypo hesis ha he mono onici y and exclusion es ic ions hold. We es his null using THA o ice
imes yea ‑o ‑p i a iza ion e ec s condi ional on ha ing handled a leas 5 p i a iza ions. Columns (1)
o (4) p o ide he esul s imposing 10 kno s in he quad a ic spline unc ion. Columns (5) o (8)
p o ide he esul s imposing 15 kno s in he quad a ic spline unc ion. Each column is associa ed wi h
di e en weigh ing schemes be ween he i and slope componen s o he es . A ailu e o ejec he
null implies ha we canno ejec he hypo hesis ha he mono onici y and exclusion es ic ions
join ly hold. The es was implemen ed in S a a ia he package es j e.
Sou ce: ISUD.
Table A.4: Reg ession Resul s Accoun ing o Ex ensi e/In ensi e Ma gin P e e ences
IV‑Model Resul s Fi s s age
(1) (2) (3) (4)
Binding con ac 0.7246*** 0.7317*** 0.7070***
(0.223) (0.225) (0.249)
In es men p e e ences
Ex ensi e ma gin ‑0.0005 ‑0.0005 ‑0.0004 ‑0.0005
(0.001) (0.001) (0.001) (0.000)
In ensi e ma gin ‑0.0005 ‑0.0004 ‑0.0004 ‑0.0004
(0.001) (0.001) (0.001) (0.000)
P i a ize s ingency (ins umen ) 0.0018***
(0.000)
Obse a ions 9,363 9,363 9,363 9,363
A e age employmen a con ac da e 60.064 60.064 60.064 60.064
A e age g ow h a e .064 .064 .064 .064
Sha e wi h Binding con ac s 0.207 0.207 0.207 0.207
F‑S a is ic 17.67 17.03 14.47
Sample condi ion
Baseline con ols Yes Yes Yes Yes
Indi idual con ols No Yes Yes Yes
Indus y con ols No No Yes Yes
No es: The able shows IV eg ession esul s. All speci ica ions con ol o ully in e ac ed THA agency
and yea ixed e ec s and a e condi ional on ha ing a leas 5 p i a iza ions pe p i a ize . Ex ensi e
ma gin in es men p e e ence e e o con ac s wi h any in es men commi men s. In ensi e ma gin
p e e ence e e o con ac s wi h he in es men a ge o e ini ial employmen in uppe decile o he
dis ibu ion. F‑S a is ic e e s o he Kleibe gen‑Paap F‑S a is ic. Baseline con ols a e he ime be ween
he i s and las audi s measu ed in days, he ime be ween con ac da e and i s audi measu ed in
mon hs, and ini ial employmen le el measu ed a he i s audi . Indi idual con ols a e he gende o
he p i a ize and academic deg ee (PhD). Indus y con ols a e 2‑digi indus y dummies. S anda d
e o s a e wo‑way clus e ed a p i a ize and THA o ice le el. Ins umen e e s o he lea e‑one‑ou
measu e o assigning igh con ac s. **p<0.1, **p<0.05, ***p<0.01.
Sou ce: ISUD.
53
IAB‑Discussion Pape 01|2024
Figu e A.2: Pe sis ence o P i a ize Cha ac e is ics
-20
0
20
40
60
80
100
Lea e-one-ou p i a ize cha ac e is ic
-20 -10 0 10 20 30 40 50 60 70 80 90 100
Lag lea e-one-ou p i a ize cha ac e is ic
coe .: .914 (s.e.: .0071)
No es: The igu e plo s he lea e‑one‑ou a e o a igh con ac (ini ial i m size < inal commi ed
size) in he p e ious case agains he lea e‑one‑ou a e o a binding con ac in he cu en case. All
plo ed alues a e mean‑s anda dized esiduals om eg essions on ully in e ac ed THA o ice and yea
o p i a iza ion ixed e ec s. The blue line co esponds o a linea eg ession. The igu e is cons uc ed
by condi ioning o ha ing handled a leas i e p i a iza ion con ac s. To al numbe o obse a ions is
8,759.
Sou ce: ISUD.
Figu e A.3: Labo P oduc i i y ac oss Fi m Size
9.5
10
10.5
11
11.5
Ln labo p oduc i i y
0
.05
.1
.15
.2
.25
.3
Densi y
2 4 6 8
Log employmen in 1990
No es: The igu e plo s labo p oduc i i y ac oss he i m size dis ibu ion among 7,620 ini ial GDR
i ms wi h sales and employmen in o ma ion in 1990. The igu e exclude he op and he bo om
1 pe cen o he p oduc i i y measu e.
Sou ce: ISUD.
54
IAB‑Discussion Pape 01|2024
Table A.5: Robus ness Tes s, Employmen G ow h
Dep. a iable:
Fi m g ow h
Random
Assignmen
Coe icien Fi s ‑s age F‑S a is ic Join F‑ es
(p‑ alue)
(1) (2) (3) (4)
A: Ins umen cons uc ion
Only pas decisions o ins umen 1.1690*** 0.0008*** 9.505 0.226
(0.354) (0.000)
Abo e 10 cases pe p i a ize 0.9589** 0.0023*** 11.38 0.355
(0.346) (0.001)
Ve y igh con ac s 1.0160** 0.0015*** 8.511 0.237
(0.424) (0.001)
Full igh ness dis ibu ion ‑0.0097*** 0.1285*** 21.58 0.848
(0.003) (0.028)
Con ac wi h ze o employmen in
i s & las
0.6380** 0.0016*** 14.83 0.392
(0.254) (0.000)
B: Con ol a iables & sample selec ion
Con ol o enego ia ion a emp s 0.6588*** 0.0018*** 16.24 0.408
(0.222) (0.000)
Con ol o penal y clause 0.7621*** 0.0016*** 13.65 0.408
(0.264) (0.000)
Con ol o pu chasing p ice
& in es men a ge
0.5524** 0.0017*** 13.65 0.408
(0.222) (0.000)
Yea s be ween con ac signed
& i s audi < 2
0.6558** 0.0020*** 16.86 0.617
(0.273) (0.001)
Mon h be ween i s & las audi > 12 0.7387** 0.0018*** 14.26 0.695
(0.269) (0.000)
MUP subsample 0.5435** 0.0018*** 10.45 0.486
(0.256) (0.001)
C: Manipula ion o he ou come a iable
Log employmen di e ences 0.8953** 0.0018*** 12.33 0.408
(0.331) (0.000)
Annualized i m g ow h,
(L /L −1)1/#yea − 1
( immed a he uppe pe cen ile)
0.2822** 0.0018*** 14.76 0.408
(0.114) (0.000)
G ow h a e < 2 & > −2 0.7443** 0.0015*** 12.33 0.404
(0.314) (0.000)
No es: The able shows IV eg ession esul s. All speci ica ions con ol o ully in e ac ed THA o ice
and yea ixed e ec s and a e condi ional on ha ing a leas i e p i a iza ions pe p i a ize . Fo
sample size easons, he MUP subsample is condi ional on ha ing a leas h ee obse a ions pe
p i a ize . Column (1) shows he poin es ima e o he main a iable o in e es (excep he
speci ica ion wi h a leas 10 obse a ions pe p i a ize ). Column (2) shows he co esponding
i s ‑s age coe icien . F‑S a is ic in column (3) e e s o he Kleibe gen‑Paap F‑S a is ic ( i s ‑s age). All
speci ica ions condi ion on he ull se o con ol a iables including baseline con ols (log) ime
be ween he i s and las audi s (+1) measu ed in days, log ime be ween con ac da e and i s audi
(+1) measu ed in days, and log ini ial employmen le el (+1) measu ed a he i s audi ), indi idual
con ols (gende o he p i a ize and academic deg ee (PhD)), and 2‑digi indus y con ols. Column
(4) shows he F‑S a is ic o a join F‑ es o andom assignmen . The dependen a iable is always he
ins umen eg essed on log ini ial employmen a iables (accoun ing, pu chasing, HR, p oduc ion,
sales, adminis a ion, R&D), and log ini ial e enue measu ed in 1990 (condi ional on indus y‑ ixed
e ec s and ully in e ac ed THA o ice and ime ixed e ec s). S anda d e o s a e wo‑way clus e ed a
p i a ize and THA o ice le el. ***p<0.1, **p<0.05, ***p<0.01.
Sou ce: ISUD.
55
IAB‑Discussion Pape 01|2024
Table A.6: OLS Reg ession Resul s, Adjus men o Sample Selec ion
Employmen P oduc i i y Exi
(1) (2) (3) (4) (5) (6)
Binding con ac 0.4313*** 0.4302*** 0.0938*** 0.0884*** 0.0229* 0.0223*
(0.025) (0.025) (0.022) (0.023) (0.011) (0.011)
Mills a io ‑0.0444 ‑0.1695** ‑0.0194**
(0.030) (0.076) (0.008)
Obse a ions 8,333 8,333 2,399 2,399 3,877 3,876
A e age employmen a
con ac da e
70.926 70.926 47.336 47.336 47.336 47.336
Mean ou come
(non‑binding con ac s)
‑0.062 ‑0.062 0.852 0.852 0.051 0.051
Sha e wi h binding
con ac s
0.192 0.192 0.188 0.188 0.188 0.188
Sample condi ion
Baseline con ols Yes Yes Yes Yes Yes Yes
Indus y con ols No No No No No No
Indi idual con ols No No No No No No
No es: The able shows OLS eg ession esul s o employmen g ow h, p oduc i i y g ow h and i m
exi on binding con ac s wi h and wi hou he in e se mills a io. The in e se mills a io is calcula ed
based on a p obi speci ica ion wi h he ou come a iable being equal o 1 i he GDR ini ial i ms is
obse ed wi h p i a iza ions con ac s ha include labo commi men . Explana o y a iable in he
selec ion equa ion a e log employmen measu ed in 1990, log sales o e employmen measu ed in
1990, THA o ice FE and indus y FE. The selec ion equa ion con ols o missing alues in employmen
and sales o e employmen by in oducing dummy a iables. All speci ica ions in he second s age
con ol o ully in e ac ed THA agency and yea ixed e ec s. Baseline con ols a e ime be ween he
i s and las audi s measu ed in mon hs, ime be ween con ac da e and i s audi measu ed in
mon hs, and log ini ial employmen le el measu ed a he i s audi . S anda d e o s a e wo‑way
clus e ed a p i a ize and THA o ice le el. *p<0.1, **p<0.05, ***p<0.01.
Sou ce: ISUD, MUP, SOESTRA.
56
IAB‑Discussion Pape 01|2024
Figu e A.7: Bunching wi h pe cen de ia ion bin
A: 1 pe cen bins B: 2 pe cen bins
Employmen > Commi men Employmen < Commi men
Excess mass (b) = 23.64
S anda d e o = 7.275
0 1000 2000 3000 4000
F equency
-40 -20 0 20 40
Commi ed o e Realized Employmen
Coun e ac ual dis ibu ion Obse ed dis ibu ion
C: 5 pe cen bins
Employmen > Commi men Employmen < Commi men
Excess mass (b) = 11.18
S anda d e o = 2.483
0 1000 2000 3000 4000
F equency
-40 -20 0 20 40
Commi ed o e Realized Employmen
Coun e ac ual dis ibu ion Obse ed dis ibu ion
Employmen < Commi men Employmen > Comi men
Excess mass (b) = 2.95
S anda d e o = .6922
0 1000 2000 3000 4000
F equency
-50 -25 0 25 50
Commi ed o e Realized Employmen
Coun e ac ual dis ibu ion Obse ed dis ibu ion
No es: The igu es show he employmen dis ibu ion a ound he commi ed employmen (dema ca ed
by he e ical ed line a 0) o con ac s be ween 1990‑1995. The blue line in do s is a his og am o
ac ual employmen ela i e o he commi men a ge in he inal commi men yea . Each poin shows
he numbe o obse a ions in employmen coun bin (de ia ion be ween he a ge and he ealized
employmen ). The solid line benea h he empi ical dis ibu ion is a wel e‑deg ee polynomial i ed o
he empi ical dis ibu ion excluding he a ea o missing one employee and ha e 4 employees mo e
han commi ed. The shaded egion in yellow is he es ima ed excess mass. S anda d e o is
calcula ed using a pa ame ic boo s ap p ocedu e. Es ima ion based on Che y e al. (2011). Panel A
shows he esul s by cons uc ing 1 pe cen age bin de ia ions. Panel B shows he esul s by
cons uc ing 2 pe cen age bin de ia ions. Panel C shows he esul s by cons uc ing 5 pe cen age bin
de ia ions.
Sou ce: ISUD.
63
IAB‑Discussion Pape 01|2024

Table A.11: Bunching by Sub‑Samples
Excess mass (b) S anda d e o
(1) (2)
A: Indus y a ilia ion
Ag icul u e, ene gy, mining 9.076 3.220
Chemis y, plas ics 4.952 0.9231
Ex ac ion o cu ‑s one, i on, cas ing, s eel o ming 7.842 3.351
S eel cons uc ion, mechanical & elec ical enginee ing,
au omobile
6.699 1.023
Pape , p in , ex ile, ood 7.617 1.070
Cons uc ion and buildings ades, wholesale, e ail 7.257 1.227
T anspo a ion, communica ion, insu ance 5.799 1.325
B: Con ac ma u i y
16 o 31 mon hs 6.574 1.198
Below 16 mon hs 8.697 3.010
Abo e 31 mon hs 7.212 1.118
C: Numbe o audi s
Mul iple audi s 6.521 0.773
D: Penal y condi ion
Exclude con ac s wi hou penal y clause 6.627 0.889
E: Ini ial size
Below a ge 6.473 0.989
Abo e a ge 5.151 0.540
No es: The able shows bunching es ima es o he employmen dis ibu ion a ound he commi ed
employmen o con ac s be ween 1990‑1995 by di e en g oups. The coun e ac ual dis ibu ion is a
based on a wel e‑deg ee polynomial i ed o he empi ical dis ibu ion excluding he a ea o missing
one employee and ha ing h ee employees mo e han commi ed. S anda d e o s a e calcula ed using
a pa ame ic boo s ap p ocedu e wi h 100 eplica ions. Es ima ion based on Che y e al. (2011). Panel
A shows he esul s by indus y. Panel B shows he esul s by con ac ma u i y cu ing a 25 h (16
mon hs be ween con ac da e and inal commi men ) and 75 h (31 mon hs be ween con ac da e and
inal commi men ) pe cen ile. Panels C and D selec only con ac s wi h mul iple audi s and wi h a
penal y clause, espec i ely. Panel E dis inguishes by ini ial con ac size (measu e a he i s audi )
ela i e o he inal a ge .
Sou ce: ISUD.
64
IAB‑Discussion Pape 01|2024
B Da a Addendum – ISUD Da a
En i onmen
This sec ion p o ides an o e iew and a desc ip ion o da a used in he empi ical
analysis. The da a we e p o ided o he au ho s on he basis o an ag eemen be ween
he IWH (Halle) and he Ge man Fede al A chi es (Bundesa chi ). This ag eemen
in ol ed he ans e o mo e han 500 sepa a e da a ables in digi ized o ma (cs ) on
ac i i ies o T euhand.
The imeline in Figu e B.1 isualizes he le el and iming o obse a ions. The main
iden i ie s in he ISUD en i onmen a e a he i m le el and a he con ac le el. The
o me is cons i u ed by in o ma ion om i ms submi ing a balance shee (DM
E ö nungsbilanz) and ansi ioning in o he THA po olio. The THA assigns ini ial IDs o
each i m, and, in he case o es uc u ings and i m sepa a ions, new IDs a e c ea ed.
Once asse s a e sold ou o he i ms, we obse e con ac IDs. These con ac s a e
o ganized and used by he con ac managemen eams (VM) o ollow up on paymen s
and obliga ions o buye s.28
Two ables a e used o measu e i m‑le el in o ma ion: basis_kennzi e n and
basis_kennzi e n_91. The able basis_kennzi e n_91 comp ises mos o he in o ma ion
and, he e o e, is he main able. In case o missing alues, we sea ch o in o ma ion
in basis_kennzi e n o complemen and o cons uc a comp ehensi e c oss‑sec ion o
i m in o ma ion o he yea 1990.29 The in o ma ion ela es o employmen (including
a b eakdown in o p oduc ion wo ke s, HR, and adminis a ion), e enues (including a
b eakdown o e enues in Eas and Wes Eu ope), and he assignmen o i ms o THA
o ices (headqua e s o local subsidia y). The da a con ains a o al o 13,552 legal i m
en i ies, ou o which 93.3 pe cen a e obse ed o he i s ime in 1990.30 We
complemen he da a wi h addi ional indus y in o ma ion om he SOESTRA su ey
(see Me gele/Hennicke/Lubczyk, 2020). The inal da a se is used in he analysis o
s udy andom assignmen o i ms o p i a ize s in Table 2, o calcula e labo
p oduc i i y g ow h be ween 1990 and he inal commi men yea in Table 4, ma ke
exi e ec s in Table 5, and o cons uc Figu e A.3.
28 Sec ion C desc ibes he me ge be ween con ac s and ex e nal i m‑le el da a, he Mannheim
En e p ise Panel (MUP), o s udy dynamics beyond he commi men pe iod.
29 The in o ma ion can be combined o cons uc a yea ly panel wi h in o ma ion a he i m le el
be ween 1989 and 1994. This da ase canno be used o s udy he e olu ion o i ms o e ime
because he i m disappea s om he da ase once he i m ansi ions ou o he THA po olio
ei he because o a p i a iza ion o liquida ion.
30 THA c ea ed legal en i ies o e ime, and, as a esul , 5.1 pe cen o i ms a e obse ed o he i s
ime in 1991, and 1.2 pe cen in 1992, and 0.48 pe cen in 1993.
65
IAB‑Discussion Pape 01|2024
E ö nungsbilanz THA Pe iod P i a iza ions Ma ke Pe iod
(01.07.1990)
Fi ms Fi m ID Con ac ID
Liquida ion
Balance Shee s THA Planning VM Con ol Managemen MUP
Baseline numbe s Gua an ee Paymen s G ow h
(Basiskennzi e n) Employmen
(Bü gscha en) (Ra en)
Financials Legacy Deb Ou lays & In lows Exi
(Finanzie ungsda en) (Al k edi ) (Budge e olgung)
Liquida ion Liabili ies
(Abwicklungsda en) (Ve bindlichkei en)
Commi men s Audi s
(Zusiche ungen) (Übe p ü ung)
Figu e B.1: Timeline om euni ica ion o he ma ke pe iod
Sou ce: Own p esen a ion.
66 IAB‑Discussion Pape 01|2024
A second se o da a ables p o ides in o ma ion on owne ship changes o i ms:
besi z_91 and besi z. Simila ly, besi z_91 comp ises mos in o ma ion, and besi z is
used o ill missing alues. Combining he wo ables gene a es a da ase wi h
in o ma ion on 13,051 i ms abou pa ial sales, p i a iza ions and liquida ion decisions.
These da a allows us o no only ack changes in owne ship, bu also o calcula e he
sha e o i ms p i a ized o liquida ed. We e e o his es ima ed sha e in Sec ion 2.
One o he main challenges o he ISUD da a en i onmen is o link in o ma ion a he
i m le el wi h con ac ‑le el in o ma ion. This link is impo an o wo easons. Fi s , i
allows us o s udy andom assignmen , p oduc i i y g ow h, and ma ke exi . Second, i
p o ides us in o ma ion on which THA di ision handles he p i a iza ion o he i m. We
i s desc ibe he da a ables used o cons uc he link be ween i ms and con ac s.
Table B.1 p o ides an o e iew o he da a ables and a sho desc ip ion.
The da a able ASVA01T o ms he main sou ce o in o ma ion o con ac s. I p o ides
us wi h in o ma ion on he con ac ID and he con ac da e. I does no , howe e ,
p o ide in o ma ion on he link be ween he con ac s and he i m. Fo his eason, we
sea ch o his in o ma ion ac oss he ISUD sys em. The ables ASVA02T, VATVT,
ASVA22T, ASVA50T, and FE3_VT a e iden i ied o be candida es ha possess he link.
Due o he deg ee o non‑missing in o ma ion, he wo mos impo an ables a e
ASVA02T and VATVT. The sea ch p ocess gene a es 48,086 unique con ac s wi h a i m
link.
Ano he ad an ageous ea u e o ASVA01T is ha i con ains no only he con ac ID
bu also he s ing names o p i a ize s who handle he con ac s and
communica e/nego ia e wi h po en ial in es o s. We clean he a iable “PNAME” which
is labeled as “Name d. zus ändigen P i a isie e s”. In he o e all ile, we gene a e 3,521
unique names o 58,544 con ac s a e name cleaning. The main eason o losing
con ac s is missing alues in his name a iable. Ou o he 256,842 con ac s in he
da a able, 147,060 do no ha e in o ma ion on he name o he p i a ize . The eason
why mos o he con ac s do no possess a name o a p i a ize is because he
con ac s a e no ela ed o i ms bu ep esen es a e, machine y o land deals.
The e o e, hese con ac s a e no ela ed o i ms and consequen ly do no ha e a
p i a ize a ached o i . Linking con ac s o con ain p i a ize in o ma ion, labo
commi men con ac s, and i m links gene a es a sample o 11,194 con ac s as shown
in Sec ion 5.
A e his p epa a ion o baseline ables, we ob ain in o ma ion on labo commi men s
and labo audi s. We s a wi h he o iginal iles ha a e called VAPST o commi men
in o ma ion and VAPIT o in o ma ion on audi s (see Panel B o Table B.1). These wo
ables can be seen as he o iginal ables as sugges ed om he deli e ed pd
67 IAB‑Discussion Pape 01|2024
Table B.1: Con ac ‑Le el Da a Tables
Table
names
Desc ip ion
A: Baseline ables
ASVA01T
The ables con ains mas e da a and s a us in o ma ion o con ac s signed wi h he
THA. I combines many a iables om di e en ables. The able con ains he con ac ID
(sysn ), he da e o he con ac signed wi h he no a y, and he name o he p i a ize .
To al numbe o unique con ac s: 256,842.
ASVA02T
The able p o ides in o ma ion on pa ial con ac s. I con ains he link be ween he con‑
ac s and he i ms, he ixed p ice payed by he con ac pa ne , and he assignmen
o THA o ices. To al numbe o unique con ac s: 213,052. Unique con ac s wi h a non‑
missing con ac ‑ i m link: 22,837.
VATVT
The able p o ides in o ma ion on pa ial con ac s. I con ains he link be ween he con‑
ac s and he i ms. To al numbe o unique con ac s: 37,967. Unique con ac s wi h a
non‑missing con ac ‑ i m link: 30,745.
ASVA22T
This able p o ides in o ma ion on mappings. I con ains he link be ween he con ac s
and he i ms. To al numbe o unique con ac s: 40,036. Unique con ac s wi h a non‑
missing con ac ‑ i m link: 9,784.
ASVA50T
This able p o ides heade da a o conce ed ac ion. I con ains he link be ween he
con ac s and he i ms. To al numbe o unique con ac s: 82. Unique con ac s wi h a
non‑missing con ac ‑ i m link: 82.
FE3_VT
This able p o ides in o ma ion on p ocesses/ope a ions o main ables ela ed o inan‑
cials. I con ains he link be ween he con ac s and he i ms. To al numbe o unique
con ac s: 1,723. Unique con ac s wi h a non‑missing con ac ‑ i m link: 1,710.
B: Labo Commi men s & Audi s
VAPST This able p o ides in o ma ion on labo commi men s o he con ac pa ne . To al num‑
be o unique con ac s: 17,753. To al numbe o obse a ions: 52,438.
VAPIT This able p o ides in o ma ion on labo audi s. To al numbe o unique con ac s: 16,583.
To al numbe o obse a ions: 116,619.
VAPITH This able p o ides in o ma ion on labo audi s and is labeled as his o y in he documen‑
a ion. To al numbe o unique con ac s: 19,052. To al numbe o obse a ions: 102,933.
ASVA12T
This able, among o he s, p o ides in o ma ion on labo commi men s. To al numbe o
unique o e all con ac s: 275,054. To al numbe o unique con ac s wi h posi i e numbe
o commi ed labo : 22,535. To al numbe o obse a ions: 322,829.
ASVA13T
This able, among o he s, p o ides in o ma ion on labo audi s. To al numbe o unique
o e all con ac s: 47,111. To al numbe o unique con ac s wi h posi i e numbe o au‑
di ed labo : 15,702. To al numbe o obse a ions: 153,155.
C: In es men Commi men s & Audi s
VAZST This able p o ides in o ma ion on in es men commi men s o he con ac pa ne . To al
numbe o unique con ac s: 18,120. To al numbe o obse a ions: 20,366.
VAZIT This able p o ides in o ma ion on in es men audi s. To al numbe o unique con ac s:
16,806. To al numbe o obse a ions: 32,096.
VAZITH
This able p o ides in o ma ion on in es men audi s and is labeled as his o y in he
documen a ion. To al numbe o unique con ac s: 26,195. To al numbe o obse a ions:
60,159.
ASVA15T
This able, among o he s, p o ides in o ma ion on in es men commi men s. To al numbe
o unique o e all con ac s: 274,375. To al numbe o unique con ac s wi h posi i e num‑
be o commi ed in es men : 24,220. To al numbe o obse a ions: 280,370.
ASVA16T
This able, among o he s, p o ides in o ma ion on in es men audi s. To al numbe o
unique o e all con ac s: 47,111. To al numbe o unique con ac s wi h posi i e numbe
o audi ed in es men : 15,619. To al numbe o obse a ions: 64,725.
Sou ce: Own p esen a ion.
68
IAB‑Discussion Pape 01|2024

documen a ion by he Ge man Fede al A chi es. The pd ile o labo commi men s is
shown in Figu e B.2. I shows he empla e how he da a was collec ed in he i s
place by THA employees. The op igh co ne co esponds o he ables VAPST and
VAPIT, espec i ely. In hese wo da a ables we obse e 17,753 unique con ac s wi h
labo commi men s and 16,583 con ac s wi h a leas one audi . As p esen ed in Panel
B, he o al numbe o obse a ions in bo h ables is highe because he e can be
mul iple commi men s o di e en yea s o he commi men pe iod as well as se e al
audi s pe commi men .
Figu e B.2: Pape File: Labo Commi men
No es: The igu es show he o iginal empla e used by he THA o documen labo commi men s.
Sou ce: Handbuch T euhandans al .
We pe o m he ollowing s eps o clean he da a. Fi s , we d op obse a ions wi hou
da e in o ma ion in bo h ables and selec he i s con ac wi hin he con ac ID in
case he e a e se e al pa ial con ac s pe ID. Ou o he 116,619 con ac ‑audi
obse a ions, hese selec ion s eps educe he sample by 36 and 674, espec i ely. Ou
o he 52,438 con ac ‑commi men obse a ions, hese selec ion s eps educe he
sample by 1,414 and 367, espec i ely. Wi hin he VAPIT ile we also d op obse a ions
whe e he numbe o employees a he audi is ze o, bu he a iable ha s a es
whe he employee in o ma ion is epo ed is se o ze o. This educes he sample
u he by 2,536 obse a ions. In o de o ob ain an ini ial i m size measu e a he
con ac le el, we selec he i s audi . The las audi ed labo in o ma ion p o ides a
69 IAB‑Discussion Pape 01|2024
measu e o he size a he inal commi men ime. We u he pe o m basic da a
cleaning s eps: (i) we d op con ac s i he da e o he las commi men is be o e he
da e o he con ac wi h he no a y (7 obse a ions), (ii) i he ime be ween wo
consecu i e commi men s is nega i e, and (iii) i he inal employmen commi men is
ze o (224 obse a ions). This gene a es a sample wi h 15,538 labo commi men
con ac s wi h a leas one ma ched employmen audi .
The ISUD en i onmen u he con ains a able called ASVA12T wi h labo commi men
con ac s. The o iginal able has 322,829 obse a ions. The majo i y o hese
obse a ions a e labeled as ha ing no labo commi men s. We compa e his da a able
wi h he o iginal VAPST able. Condi ional on obse ing one con ac ID in bo h ables
(VAPST and ASVA12T) shows ha he in o ma ion is iden ical. Howe e , ASVA12T has
5,125 addi ional con ac s wi h labo commi men s ha a e no included in VAPST.
These addi ional con ac s a e, on a e age, la e w i en ou and a e en e ed in o he
ISUD da a sys em mainly in 2003 and 2004. A e ollowing he same da a cleaning
s eps, we end up wi h 3,385 addi ional con ac s. In e ms o labo audi s, howe e ,
hese con ac s a e no obse ed in VAPIT. The e exis s ano he da a able ha is a
na u al suspec and is called ASVA13T. Bu again, his able does no con ain audi
in o ma ion o he addi ional con ac s wi h obse ed labo commi men s.31 A e
sea ching o possible con ac s wi h addi ional audi in o ma ion, we ound ha he
his o y e sion o VAPIT, called VAPITH, is sui able o ill pa s o he missing audi s
om ASVA12T. Among he 3,385 addi ional con ac s a e basic da a cleaning s eps, we
a e able o me ge he audi in o ma ion o 2,702 con ac s. Toge he , hese da a ables
gene a e ou inal sample o 18,235 con ac s wi h labo commi men s.
Fo he empi ical speci ica ions accoun ing o ex ensi e/in ensi e ma gin p i a ize
p e e ences p esen ed in Table A.4, we make u he use o in es men commi men
con ac s. The logic and s eps in he da a cleaning p ocess apply simila ly o
in es men commi men con ac s. Figu e B.3 shows he empla e used o he
documen a ion o in es men commi men s. The baseline da a able o in es men
wi h in o ma ion on in es men commi men s is called VAZST, whe eas he able o
in es men audi s is called VAZIT. Panel C o Table B.1 p o ides a lis and sho
desc ip ion o he in es men commi men ela ed da a ables.
A e basic da a cleaning s eps and combining commi men in o ma ion in VAZST wi h
audi in o ma ion in VAZIT, we ob ain a da ase wi h 15,086 in es men commi men s.
The da a able ASVA15T has 7,127 addi ional con ac s ha a e no obse ed in he
baseline iles. Simila o he addi ional employmen con ac s, ASVA16T does no
con ain audi s o hese addi ional con ac s. Again, exploi ing VAZITH, he his o y ile o
31 Ou o he 5,125 addi ional con ac s wi h labo commi men s ASVA12T, 17 con ac s a e ound in
VAPIT and 22 con ac s a e ound in ASVA13T.
70
IAB‑Discussion Pape 01|2024
Figu e B.3: Pape File: In es men Commi men
No es: The igu es show he o iginal empla e used by he THA o documen in es men commi men s.
Sou ce: Handbuch T euhandans al .
VAZIT, we a e able o add 4,978 con ac s. Toge he , hese da a ables gene a e ou
inal sample o 20,062 con ac s wi h in es men commi men s.
One ema kable di e ence be ween in es men and labo commi men con ac s is he
numbe o audi s. While he sha e o con ac s wi h only one audi is abou 17 pe cen
among he labo commi men con ac s, his sha e is 65.2 pe cen . Due o he low
na u e o in es men commi men , he e a e ewe audi s du ing he commi men
pe iod. Combining labo wi h in es men con ac s esul s in a sample o 23,662 unique
con ac ‑le el obse a ion. Among hem, 14,635 con ac s ha e bo h, labo and
in es men commi men s, 5,427 only ha e in es men commi men s, and 3,600
con ac s only ha e labo commi men s. In o de o calcula e ex ensi e ma gin
p e e ences i.e., w i ing con ac s wi h any labo commi men condi ion we me ge his
combined da ase wi h he 58,544 con ac s wi h cleaned p i a ize names.
71 IAB‑Discussion Pape 01|2024
C Da a Addendum – Me ging
Con ac s o Mannheim En e p ise
Panel Da a
This sec ion desc ibes he me ge be ween ou baseline con ac ‑le el da a and he
Mannheim En e p ise Panel da a, which co e i ms in Eas Ge many s a ing om 1993
o 2019 ( he mos ecen wa e). The Mannheim En e p ise Panel (MUP), is he mos
comp ehensi e mic o da abase o companies in Ge many ou side o adminis a i e
da a. O icial adminis a i e da a is usually no accessible o he public. The da a
con ains de ailed in o ma ion on he i m‑le el ha is o en ha d o come by in
adminis a i e eco ds such as, o ins ance, he da e o c ea ion and closu e o a
company, owne ship s uc u es, and c edi a ing sco es. Besides ha , he da ase
comp ises employmen , sales, and indus y a ilia ion in o ma ion. The MUP is based on
he i m da a pool o C edi e o m e.V., which is he la ges c edi a ing agency in
Ge many. While i has b oad o e all co e age i does no o e 100 pe cen co e age
( o u he de ails, see Be sch e al. (2014)).
A he le el o he con ac s, we do no obse e i m names ha would allow a s ing
ma ching based on hese names. Ins ead, we explo e he owne ship in o ma ion in
bo h da ase s. In he MUP da a, we obse e o each i m owne . In he con ac ‑le el
da a, we ha e access o he con ac pa ne , who ususally becomes he new owne o
he company a e he con ac is signed wi h he no a y.
Among he 18,235 con ac s in he baseline da a, we s a o wi h 9,538 ha can be
linked ia name ma ching be ween he owne s in he MUP and con ac pa ne s in he
con ac da a. These obse a ions co espond o 11,199 con ac pa ne s. These
indi iduals usually ha e mul iple links o i ms a di e en poin s in ime and ac oss
space. In o de o selec he co ec i m o he con ac , we pe o m he ollowing
p e‑selec ion:
• D op i i m is loca ed in Wes Ge many
• D op i o iginal i m unde T euhand is loca ed in di e en Fede al S a e han MUP
i m
• D op i i m/con ac loca ion, da e o inco po a ion, con ac da e is missing
• D op i da e o inco po a ion/owne ship s a is a e 2000
• D op i con ac da e is i e yea s a e da e o inco po a ion
72 IAB‑Discussion Pape 01|2024
D Da a Addendum – T euhand Fi m
Su ey Da a
This sec ion desc ibes how we cons uc i m‑le el capi al s ock and TFP es ima es and
he me ge be ween ou baseline con ac ‑le el da a and he THA i m su ey da a. The
bi‑annual su ey was conduc ed by he he SOESTRA ins i u e wi h i s i s wa e in Ap il
1991. The su ey da a has been used and analyzed, among o he s, by Wahse e al.
(1996) and Me gele/Hennicke/Lubczyk (2020).
The ocus o he ques ionnai e was on employmen and mos o he su ey wa es also
con ain ques ions on i m e enue. Impo an o ou pu pose o cons uc he
i m‑le el capi al s ock is he ac ha some wa es also con ain in o ma ion on
in es men s. Apa om hese main a iables, he su ey con ains baseline in o ma ion
on he sec o a ilia ion, he loca ion o he i m, and end da es o THA owne ship and
labo commi men s (i any). Ou o hese wa es, we i s cons uc an (unbalanced)
mon hly i m panel be ween 1991 and 2000. This ini ial panel con ains 11,105 T euhand
i ms.
D.1 Cons uc ing Fi m‑Le el Capi al S ock and TFP
Measu es
The i s aim is o con e he mon hly panel in o a yea ly panel. Ou o 36,735 e enue
obse a ions o e he yea s be ween 1991 and 2000 and belonging o 9,596, 69 pe cen
o he in o ma ion belongs o an end‑o ‑yea ques ion. Thus, mo e o he e enue
in o ma ion is ela ed o a ull calenda yea . Fu he , 15 pe cen o he e enue
ques ions ask o e enue numbe s du ing he i s hal o he yea , and he emaining
belongs ei he o he i s qua e o he yea (9.4 pe cen ) o o he hi d qua e o he
yea (6.6 pe cen ). Likewise, he su ey co e s 17,896 in es men in o ma ion belonging
o 6,743 i ms. The majo i y o 95.3 pe cen o he in es men numbe s a e ela ed o
he ull calenda yea , and he emaining 4.6 pe cen ela e o he i s six mon hs o
he yea . The e o e, we ha monize he da a o he yea ly le el by assuming linea i y
e.g., i we only obse e e enue/in es men in o ma ion o he i s six mon h o he
yea , we mul iply by 2 o cons uc he numbe o he yea . In mos cases, howe e ,
in o ma ion a e ypically a ailable o he ull yea and o a ac ion o he yea . We
inally impu e o 652 i ms e enue in o ma ion and o 834 i ms in es men
in o ma ion o he end o he yea . Rega ding employmen , we cons uc he a e age
79 IAB‑Discussion Pape 01|2024

employmen le el ou o he mon hly in o ma ion. We complemen he su ey da a on
yea ly employmen and e enue wi h basis_kennzi e n as desc ibed in Appendix B.
The ini ial capi al s ock is cons uc ed using balance shee in o ma ion submi ed by
he i ms o he yea 1990. The da a able is called DM_BIL_N. The ini ial capi al s ock
consis s o angible asse s, including mainly p ope ies, ( echnical) equipmen , and
machine y. These angible asse s ep esen 97 pe cen o he ini ial capi al s ock. The
emaining ac ion comes om b eeding s ock, concessions, and soil imp o emen .
Ini ial capi al s ock in o ma ion is a ailable o 7,182 i ms. We hen clean he da ase
and d op i ms en i ely i he i m does no ha e a single employmen o sales
in o ma ion, which d ops he ini ial sample o 11,105 i ms o 10,390 i ms. In he
occu ence ha employmen and e enue in o ma ion wi hin he i m con ain gaps, we
linea ly impu e hese gaps o up o wo yea s.
In o de o calcula e he yea ly capi al s ock a he i m le el, we s a wi h he ini ial
capi al s ock measu ed in 1990, add in es men s, and assume a 10 pe cen dep ecia ion
a e. All Deu sch Ma k (DM) alues a e de la ed by he CPI measu edin 2016 p ices. The
capi al s ock can only be es ima ed i in es men in o ma ion is a ailable. Table D.1
shows in column (4) ha he ques ion on in es men p ima ily exis s o he yea s
be ween 1992 and 1995. Co e age is pa icula ly low owa ds he end o he sample
pe iod and in 1991. Fo example, he e a e only 560 i ms wi h ull in es men
in o ma ion be ween 1991 and 1994, and only 160 always ha e in es men numbe s.
Likewise, bu o a lowe ex en , column (3) shows he numbe o i ms wi h e enue
in o ma ion. In he i s wo yea s, a ound 98 pe cen o all i ms do ha e in o ma ion
on e enue, whe eas his sha e dec eases o 65 pe cen in 2000.
Table D.1: Ac ual and Impu ed In es men In o ma ion
Yea
(1)
N
(2)
N wi h in es men
(3)
N wi h impu ed in es men
(4)
1991 6,764 682 5,767
1992 6,764 3,572 3,130
1993 6,707 2,428 3,694
1994 6,583 1,364 4,198
1995 6,003 1,145 2,962
1996 5,187 500 2,383
1997 4,369 535 1,963
1998 3,633 555 1,447
1999 2,711 515 1,293
No es: The able shows he numbe o i ms in he inal su ey da a as well as he numbe o i ms
wi h ac ual and impu ed e enue and in es men in o ma ion.
Sou ce: SOESTRA.
To cons uc he capi al s ock, we i s employ a machine‑lea ning assis ed impu a ion
app oach by p edic ing in es men numbe s and use he p edic ed alues in case
80 IAB‑Discussion Pape 01|2024
ac ual numbe s a e missing. We employ a s anda d leas absolu e sh inkage and
selec ion ope a o (lasso) wi h an op imal uning pa ame e using a 10‑ old
c oss‑ alida ion. The co a ia es used in he baseline lasso eg ession include e enue
and employmen , bo h measu ed in size bins and 259 4‑digi sec o dummies. We
pe o m he p edic ion exe cise sepa a ely o e e y yea . We p o ide he esul s o he
in es men impu a ion also using ln(employmen ) and ln( e enue) as well as hese
a iables in oduced wi h a second deg ee polynomial. Due o he ac ha T euhand
i ms go es uc u ed ( o di e en deg ees) un il p i a iza ion, we also use a
p opo ional impu a ion app oach. Fo his app oach, we app oxima e he ini ial capi al
s ock by mimicking he ac ion o employmen a p i a iza ion ela i e o he ini ial i m
size. Fo example, i a i m ge s p i a ized wi h 50 employees and he ini ial i m size in
1990 was 500 employees, we assume he ini ial capi al s ock o be 10 pe cen o he
ac ual capi al s ock measu ed in 1990.
Table D.2 p o ides baseline in o ma ion o each lasso speci ica ion measu ing
employmen and e enue in bins. Speci ically, we in oduce 11 employmen size bins
[1‑4; 5‑19; 50‑99; 100‑149; 150‑249; 250‑499; 500‑749; 740‑1449; 1450‑2999; 3000+] and 9
(ln) e enue size bins [<12.51356; 12.51356‑13.26366; 13.26366‑14.36855;
14.36855‑15.50374; 15.50374‑16.67438; 16.67438‑17.80855; 17.80855‑18.57818;
18.57818‑20.10738; 20.10738+]. The numbe o non‑ze o co a ia es dec eases as he
sample size dec eases, indica ed by a highe op imal c oss‑ alida ed penal y
pa ame e .
Table D.2: Lasso Resul s: ln(in es men )
N
(1)
Op imal
lambda
(2)
Numbe o non‑ze o
coe icien s
(3)
C oss‑ alida ed minimum
p edic ion e o
(4)
1991 4,908 0.015 163 2.131
1992 3,665 0.021 142 2.174
1993 2,112 0.032 99 2.041
1994 1,864 0.035 95 2.294
1995 750 0.057 65 2.322
1996 825 0.038 78 2.077
1997 851 0.039 94 2.046
1998 833 0.046 69 2.099
1999 739 0.033 86 1.848
No es: The able shows summa y esul s om yea ly lasso eg essions wi h ln(in es men ) as he
ou come a iable.
Sou ce: SOESTRA.
Figu e D.1 shows ac ual s p edic ed in es men numbe s pooled o e he whole ime
pe iod. On a e age, ac ual and p edic ed numbe s line up a he 45 deg ee line. Based
on hese p edic ions, we impu e in es men in o ma ion in case ac ual in es men
81 IAB‑Discussion Pape 01|2024
in o ma ion is missing and he selec ed co a ia es a e no missing. Column (4) o Table
D.1 shows he numbe o impu ed obse a ions o e ime.
Figu e D.1: Co ela ion Ac ual and P edic ed Values
10 12 14 16 18
In es men
10 12 14 16 18
P edic ed in es men
No es: The igu e plo s ac ual s. p edic ed in es men numbe s pooling all yea s be ween 1991 and
1999 wi h he c oss‑ alida ed lambda.
Sou ce: SOESTRA.
In a nex s ep, we cons uc he capi al s ock a he i m le el s a ing wi h he ini ial
capi al s ock in 1990 and add hese (ac ual and impu ed) in es men numbe s and
sub ac a 10 pe cen dep ecia ion a e. Figu e D.2 p o ide i m‑le el a e ages o e he
pe iod be ween 1990 and 1999. Al hough hese numbe s migh no be ep esen a i e
o he Eas Ge many economy due o selec i i y and panel a i ion, he panels A and
C o he igu e show an inc easing end in he cons uc ed capi al s ock measu e and
i m e enue. A e age in es men amoun s dec ease o e ime ,indic ing a
disp opo ional high in es men need. A e age i m‑le el employmen dec eases o e
ime. The d op in i m le el employmen is consis en wi h o al employmen in he
economy, wi h he la ges dec eased happening be ween 1990 and 1991.
In a nex s ep, we aim o cons uc a measu e o o al ac o p oduc i i y (TFP). Due o
he ac ha we ha e no in o ma ion on in e media e inpu s such as ma e ial, we un a
simple Cobb‑Douglas eg ession speci ica ion o each yea , wi h inpu ac o s being
i m‑le el employmen and he cons uc ed measu e o capi al. Ou pu is measu ed by
e enue. All a iables a e de la ed by he CPI. Speci ically, we es ima e
yi = α + βlli + βkki + ϵi
82
In es men
IAB‑Discussion Pape 01|2024
Figu e D.2: Main a iables used om Soeas a i m su ey
A: Capi al s ock
20000 25000 30000 35000 40000
Capi al s ock (in 1000 DM)
1990 1992 1994 1996 1998 2000
Yea
B: In es men
3000 4000 5000 6000
In es men (in 1000 DM)
1990 1992 1994 1996 1998 2000
Yea
C: Re enue D: Employmen
20000 30000 40000 50000 60000
Re enue (in 1000 DM)
1990 1992 1994 1996 1998 2000
Yea
100 200 300 400
Employmen
1990 1992 1994 1996 1998 2000
Yea
No es: The igu es plo a e age i m‑le el capi al s ock, in es men , e enue, and employmen numbe s
be ween 1991 and 1999.
Sou ce: SOESTRA.
whe e yi is he loga i hm o he i m’s ou pu , in ou case, e enue. li and ki a e he
loga i hm o he i m inpu s, in ou case, he numbe o employees and he capi al
s ock. We cons uc TFP as ωi = exp(yi − β
ˆ
lli − β
ˆ
kki). Table D.3 p o ides he eg ession
esul s sepa a ely o each yea be ween 1991 and 1999. In Panel A, we p o ide he
esul s using he baseline impu a ion app oach o i m‑le el in es men . Excep o he
i s and he las yea o he sample, we es ima e βl o be a ound 0.65 and βk o be
a ound 0.35. Towa ds he end o he sample, bo h coe icien s inc ease wi h a
signi ican dec ease o he size o he sample. The es ima es’ elas ici ies in he yea
1991 a e a he o equal size, and bo h a e below 0.5. This migh be he esul s o
dis o ed i m sizes unde socialism.
Panels B and C p o ide he es ima ion esul s o he di e en lasso speci ica ions.
Panel D p o ides he esul s wi h he baseline impu a ion p ocedu e using he
p opo ionali y app oxima ion o he ini ial capi al s ock. While Panels B and C show
a he simila esul s, Panel D shows ha βk is highe by a magni ude o a ound 0.1,
83
IAB‑Discussion Pape 01|2024
Table D.3: Reg ession Resul s: ln( e enue)
1991 1992 1993 1994 1995 1996 1997 1998 1999
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A: baseline, co a ia es: employmen & e enue dummies
Ln(Empl.) 0.4899*** 0.6679*** 0.6937*** 0.6482*** 0.6683*** 0.5895*** 0.6070*** 0.6887*** 0.7290***
(0.025)(0.020) (0.015) (0.013) (0.012) (0.014) (0.019) (0.019) (0.023)
Ln(Capi al) 0.4582*** 0.3497*** 0.3454*** 0.3859*** 0.3654*** 0.4019*** 0.4407*** 0.4093*** 0.4002***
(0.023)(0.018) (0.015) (0.014) (0.014) (0.014) (0.018) (0.019) (0.021)
N 6,448 6,449 5,823 5,251 3,847 2,677 2,296 1,771 1,502
R2 0.560 0.538 0.577 0.583 0.623 0.597 0.634 0.672 0.699
Panel B: co a ia es: ln(employmen ) & ln( e enue)
Ln(Empl.) 0.4413*** 0.6505*** 0.6832*** 0.6378*** 0.6622*** 0.5794*** 0.5977*** 0.6796*** 0.7251***
(0.026)(0.020) (0.016) (0.013) (0.012) (0.014) (0.019) (0.020) (0.023)
Ln(Capi al) 0.4927*** 0.3613*** 0.3502*** 0.3897*** 0.3634*** 0.3988*** 0.4347*** 0.4056*** 0.3927***
(0.022)(0.018) (0.015) (0.014) (0.013) (0.014) (0.018) (0.018) (0.021)
N 6,448 6,449 5,823 5,251 3,847 2,677 2,296 1,771 1,503
R2 0.566 0.542 0.580 0.587 0.625 0.599 0.635 0.673 0.699
Panel C: co a ia es: ln(employmen ) & ln( e enue) wi h second polynomial o de
Ln(Empl.) 0.4293*** 0.6509*** 0.6843*** 0.6383*** 0.6628*** 0.5805*** 0.5996*** 0.6763*** 0.7214***
(0.026)(0.020) (0.016) (0.013) (0.012) (0.014) (0.019) (0.019) (0.023)
Ln(Capi al) 0.5007*** 0.3603*** 0.3499*** 0.3909*** 0.3633*** 0.3947*** 0.4300*** 0.4092*** 0.3984***
(0.023)(0.018) (0.015) (0.014) (0.013) (0.014) (0.018) (0.018) (0.021)
N 6,439 6,440 5,816 5,244 3,840 2,669 2,290 1,766 1,499
R2 0.563 0.538 0.577 0.584 0.621 0.592 0.629 0.671 0.699
Panel D: baseline wi h p opo ional ini ial capi al s ock
Ln(Empl.) 0.3012*** 0.5456*** 0.6002*** 0.5460*** 0.5764*** 0.4876*** 0.4872*** 0.5961*** 0.6467***
(0.033)(0.022) (0.020) (0.017) (0.015) (0.017) (0.024) (0.024) (0.029)
Ln(Capi al) 0.6537*** 0.4454*** 0.4109*** 0.4728*** 0.4569*** 0.4836*** 0.5351*** 0.4759*** 0.4508***
(0.029)(0.020) (0.021) (0.018) (0.018) (0.018) (0.023) (0.024) (0.027)
N 4,279 4,280 4,004 3,683 2,748 1,919 1,649 1,283 1,063
R2 0.600 0.528 0.569 0.572 0.622 0.579 0.618 0.653 0.671
Mean e enue 15.33 15.474 15.434 15.55 15.66 15.744 15.684 15.602 15.556
Mean employmen 4.636 3.966 3.676 3.484 3.394 3.338 3.436 3.526 3.714
Mean capi al 15.674 15.748 15.81 15.878 15.95 15.948 15.922 15.906 15.85
No es: The able shows p oduc ion unc ion es ima ion esul s o ln( e enue) o each yea be ween
1991 and 1999 wi h inpu s ln(employmen ) and ln(capi al). Di e en panels indica e di e en lasso
speci ica ions o impu e in es men o cons uc ing i m‑le el capi al s ock. Panel A (baseline) uses as
co a ia es g oup size bins in employmen and e enue. Panel B uses as co a ia es ln( e enue) and
ln(employmen ). Panel C uses as co a ia es ln( e enue) and ln(employmen ) wi h a polynomial deg ee
o o de 2. Panel D uses as co a ia es he baseline e enue and employmen in oduced wi h size
dummies. All lasso speci ica ion include 259 4‑digi sec o dummies.
Sou ce: SOESTRA.
whe eas βl is lowe by abou he same magni ude. The eason migh be ha he
impu ed in es men numbe s a e ela i ely la ge, ela i e o he app oxima ed ini ial
capi al s ock, which inc eases he elas ici y o capi al in he p oduc ion unc ion
es ima ion.
D.2 Me ging Con ac s o T euhand Fi m Su ey Da a
The sec ion desc ibes he linkage be ween he con ac s and he su ey da a. This
combined da ase allows us o es ima e he e ec s o binding labo commi men
con ac s on TFP g ow h. The main challenge o linking he wo da ase s come om
84 IAB‑Discussion Pape 01|2024

he ac ha he su ey da a co e s ini ial i m uni s, whe eas he con ac s migh
belong o only pa o he i m asse s. This becomes appa en because we obse e
mul iple con ac s wi hin ini ial T euhand i ms.
The ini ial i m su ey sample co e s 11,105 T euhand i ms wi h in o ma ion on
employmen , e enue, and in es men s measu ed a di e en poin s in ime a he
mon hly le el. The ISUD da a en i onmen con ains 47,322 con ac s me ged o 10,023
T euhand i m IDs. In o de o selec he con ac s ha belong o he legal uni o he
T euhand i m, we me ge con ac s wi h labo commi men s a he le el o he
T euhand i m ID and mon h o he yea . Fo example, in he case o wo labo
commi men con ac s belonging o he same ini ial T euhand i m, we can compa e
employmen in o ma ion om he su ey and he audi s and selec he bes ma ch.
Simila o Appendix Sec ion C, we calcula e he ela i e employmen di e ences as:
(emplsu ey − emplISUD)
employmen di = ,
(emplsu ey + emplISUD)
whe e emplsu ey and emplISUD e e o he espec i e employmen igu es in bo h
da ase s and keep he con ac wi h he smalles absolu e de ia ion. In addi ion, we
d op ma ched pai s i he absolu e di e ence in bo h employmen numbe s is abo e
1000 employees (30 obse a ions) and also d op 71 obse a ions because wo o mo e
con ac s gene a e he same de ia ion in employmen , making i impossible o selec
he co ec one. This gene a es a sample o 5,221 T euhand i ms wi h selec ed labo
commi men con ac s.
To judge he success o he linkage, we de ine a ma ch o be close o accep able i he
employmen di e ence is smalle o equal o he ollowing h eshold alue:
1
abs(employmen di ) ≤ √ .
(min[emplsu ey, emplISUD] + 1)
Ou o he 5,221 linked con ac s, 73.07 pe cen ul ill his condi ion.
We combine his da ase wi h he TFP measu e a he i m le el calcula ed and
desc ibed in Sec ion D.1. A he con ac le el, we me ge in o ma ion ela ed o he
labo commi men ( i s and las labo audi in o ma ion including he iming, he inal
commi men le el, he da e o he con ac signed wi h he no a y) and ela ed o he
con ac in gene al (p i a ize in o ma ion, THA o ice in o ma ion, sales p ice,
in es men a ge ). This gene a es a sample o 2,185 i ms wi h in o ma ion on he
change in TFP be ween he ini ial con ac yea and he inal yea o he labo
commi men .
85 IAB‑Discussion Pape 01|2024
We ollow he empi ical speci ica ion om he baseline model – including THA o ice
imes yea ixed e ec s, ini ial i m size, ime be ween he i s and las audi s, and
indus y and p i a ize ‑le el con ac s – and show esul s o binding con ac s on TFP
g ow h wi h and wi hou including pu chasing p ice and in es men a ge s as con ol
a iables. We de ia e om he baseline model by condi ioning he sample on obse ing
h ee o mo e p i a iza ions pe p i a ize (ins ead o i e) o sample size easons. This
de ines he inal sample o 1,962 i m‑con ac obse a ions.
Panel A o Table A.10 shows he esul s o binding labo con ac s on i m‑le el TFP
g ow h ollowing he capi al s ock impu a ion o Panel A in Table D.3. OLS esul s show
ha binding labo commi men s a e associa ed wi h and inc ease in TFP g ow h o
abou 15 pe cen poin s. This poin es ima e is a he s able ac oss di e en empi ical
speci ica ions and close o he ϵ ans o med labo p oduc i i y esul s o 0.145
p esen ed in Tables 4 and A.9. The poin es ima es dec ease by abou 6 pe cen poin s
o 10 pe cen poin s when including i m exi s in Panel B o Table A.10. Column (4)
p o ides he 2S2SLS esul s wi h a highly signi ican (boo s apped) poin es ima e o
a ound 0.93 (Panel A). Compa ed o Tables 4 and A.9 esul s a e highly in line wi h each
o he . When including i m exi s, 2S2SLS es ima es dec ease o 0.58 (signi ican a he
1 pe cen le el). Again, compa ed o Panel C o Table A.9 wi h documen ed poin
es ima es o a ound 0.55, hese esul s a e e y consis en .
Table D.4: OLS Robus ness Resul s: TFP G ow h
Co a ia es:
ln( e ) & ln(empl)
Co a ia es:
ln( e ) & ln(empl), poly 2
Co a ia es:
Baseline/p opo ional
ini ial capi al
No
impu a ion
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Binding con ac s 0.097*** 0.112*** 0.114*** 0.092*** 0.107*** 0.109*** 0.063 0.066 0.068* 0.1019
(0.034) (0.035) (0.036) (0.032) (0.034) (0.034) (0.037) (0.038) (0.039) (0.251)
Obse a ions 1,962 1,962 1,962 1,961 1,961 1,961 1,962 1,962 1,962 91
TFP g ow h .474 .474 .474 .486 .486 .486 .556 .556 .556 .386
Baseline con ols Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Indi idual con ols Yes Yes Yes Yes Yes Yes Yes Yes Yes No
Indus y con ols Yes Yes Yes Yes Yes Yes Yes Yes Yes No
Pu chasing p ice No Yes Yes No Yes Yes No Yes Yes No
In es men a ge No No Yes No No Yes No No Yes No
No es: The able shows OLS eg ession esul s o TFP g ow h on binding con ac s o di e en lasso
speci ica ions o impu e in es men o cons uc ing i m‑le el capi al s ock. Columns (1)‑(3) use as
co a ia es ln( e enue) and ln(employmen ). Columns (4)‑(5) use as co a ia es ln( e enue) and
ln(employmen ) wi h a polynomial deg ee o o de 2. Columns (7)‑(8) use as co a ia es he baseline
e enue and employmen in oduced wi h size dummies. All lasso speci ica ion include 259 4‑digi
sec o dummies. Column (10) p o ides he esul s wi hou in es men impu a ion. All eg ession
speci ica ions con ol o ully in e ac ed THA agency and yea ixed e ec s. Binding con ac s a e
de ined as ini ial i m size below he commi ed a ge le el. Baseline con ols a e as in he baseline
speci ica ion. The pu chasing p ice is lexibly in oduced using decile dummies. In es men a ge s is a
dummy is he con ac con ains in es men commi men s. S anda d e o s a e wo‑way clus e ed a
p i a ize and THA o ice le el. *p<0.1, **p<0.05, ***p<0.01.
Sou ce: ISUD, SOESTRA.
86
IAB‑Discussion Pape 01|2024
Table D.4 shows OLS es ima ion esul s wi h di e en speci ica ions o he impu a ion
o he capi al s ock a iables (condi ional on su i al). The i s six columns use
e enue and employmen in di e en combina ions, whe eas columns (7) o (9)
app oxima e he ini ial capi al s ock measu ed in 1990 p opo ional o he employmen
sha e in he yea o he con ac . All speci ica ions p o ide posi i e poin es ima es
be ween 0.07 and 0.12 pe cen poin s. The inal column (10) p o ides he esul s
wi hou he impu a ion o he capi al s ock. Following he desc ip ion in Sec ion D.1,
his esul s in a sample size o 91 obse a ions. Al hough insigni ican due o he
sample size, he poin es ima e is wi h 0.102, a he close o he speci ica ions wi h
impu ed capi al s ock.
87 IAB‑Discussion Pape 01|2024
Lis o Figu es
Figu e 1: THA Headqua e s and Subsidia ies.............................................. 12
Figu e 2: P o i s ac oss Fi m P oduc i i y................................................... 16
: Con ac s and Labo Audi s ....................................................... 20Figu e 3
Figu e 4: Employmen Dis ibu ion a ound he Commi men Le el .................... 22
Figu e 5: Fi s ‑s age analysis.................................................................. 26
Figu e 6: Employmen Dynamics by Ini ial Size o Ta ge ................................ 32
Figu e 7: P oduc i i y G ow h and he Deg ee o Binding Con ac s ................... 35
Figu e 8: Ma ke Exi and he Deg ee o Binding Con ac s .............................. 37
Figu e 9: Employmen G ow h................................................................ 40
Figu e 10: To al Employmen in he Pos ‑Commi men Pe iod ........................... 40
Figu e 11: To al Employmen unde Coun e ac ual Economies .......................... 42
Figu e A.1: P i a iza ions pe P i a ize ........................................................ 50
Figu e A.2: Pe sis ence o P i a ize Cha ac e is ics ......................................... 54
Figu e A.3: Labo P oduc i i y ac oss Fi m Size.............................................. 54
Figu e A.4: Co ela ion P oduc i i y Measu es ................................................ 60
Figu e A.5: Bunching wi h di e en polynomials............................................. 61
Figu e A.6: Bunching wi h symme ic R....................................................... 62
Figu e A.7: Bunching wi h pe cen de ia ion bin ............................................ 63
Figu e B.1: Timeline om euni ica ion o he ma ke pe iod ............................. 66
Figu e B.2: Pape File: Labo Commi men ................................................... 69
Figu e B.3: Pape File: In es men Commi men ............................................. 71
Figu e C.1: Compa ison o Employmen Figu es be ween Con ac s and MUP .......... 76
Figu e C.2: Close Ma ches be ween Con ac s and MUP .................................... 77
Figu e C.3: Employmen Dis ibu ion a ound he Commi men Le el using Fi m‑Le el
Da a ................................................................................... 78
Figu e D.1: Co ela ion Ac ual and P edic ed Values ........................................ 82
Figu e D.2: Main a iables used om Soeas a i m su ey ................................ 83
88 IAB‑Discussion Pape 01|2024