New Evidence on News-Driven Business Cycles

Hae el, Thomas; Lucke, Be nd Wo king Pape New E idence on News-D i en Business Cycles Economics Discussion Pape s, No. 2007-27 P o ided in Coope a ion wi h: Kiel Ins i u e o he Wo ld Economy – Leibniz Cen e o Resea ch on Global Economic Challenges Sugges ed Ci a ion: Hae el, Thomas; Lucke, Be nd (2007) : New E idence on News-D i en Business Cycles, Economics Discussion Pape s, No. 2007-27, Kiel Ins i u e o he Wo ld Economy (I W), Kiel This Ve sion is a ailable a : h ps://hdl.handle.ne /10419/17950 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 p://c ea i ecommons.o g/licenses/by-nc/2.0/de/deed.en discussion Pape s Discussion Pape 2007-27 July 9, 2007 New E idence on News-D i en Business Cycles Thomas Hae el and Be nd Lucke Uni e si y o Hambu g Abs ac : We s udy he Beaud y and Po ie (2006)-hypo hesis o delayed- echnology di usion and news- d i en business cycles. Fo Ge man da a on TFP and s ock p ices we ind quali a i ely simila empi ical e idence. Quan i a i ely, howe e , an impulse esponse analysis sugges s ha a sub- s an ial pa o he o al TFP esponse is immedia e a he han delayed. We ela e his o disem- bodied echnological change and noisy da a on TFP. Ne e heless, we con i m he echnology in e p e a ion o s uc u al shocks by showing ha hey a e G ange -causal o da a on pa en s g an ed by he Ge man pa en agency. JEL: E32 Keywo ds: news, business cycles, TFP, s uc u al VAR Co espondence: Uni e si y o Hambu g, on-Melle-Pa k 5, D-20146 Hambu g, email: Thomas.Hae [email protected] g.de Uni e si y o Hambu g, on-Melle-Pa k 5, D-20146 Hambu g, email: luck[email p o ec ed].de h p://www.economics-ejou nal.o g/economics/discussionpape s © Au ho (s) 2007. This wo k is licensed unde a C ea i e Commons License - A ibu ion-NonComme cial 2.0 Ge many 1 In oduc ion In a ecen pape , Beaud y and Po ie (2006) ha e emphasized ha s ock p ices may ha e ele an in o ma ional con en o unde s anding mac oeconomic luc ua ions. New in o ma ion, hei a gumen goes, may al e expec a ions abou u u e undamen als. Fo wa d looking a iables such as s ock p ices will eac o changes in expec a ions much ea lie han he o eseen changes in undamen als a ec o he mac oeconomic ime se ies. In pa icula , news abou echnological inno a ions may a ec s ock p ices ins an aneously, bu due o an implemen a ion lag, i may ake some ime un il hey ac ually al e o al ac o p oduc i i y (TFP). Thus, ou unde s anding o expec a ions-d i en mac oeconomic luc ua ions could be enhanced i news abou expec ed changes in undamen als could be p ope ly iden i ied om, among o he s, s ock ma ke da a. Fo his pu pose, Beaud y and Po ie (BP) sugges o use s uc u al ec o au o eg essions (SVAR). Imposing di e en iden i ying es ic ions on he es ima ed lag polynomial o a mo ing a e age (MA) ep esen a ion (c . Blancha d and Quah (1989)) yields a se o s uc u al shocks o eady compa ison. I simila shocks a e ound unde di e en iden i ying assump ions, hen he ype o iden i ying assump ion e eals impo an in o ma ion abou he way a speci ic shock hi s he economy. This, in u n, allows in e ence on he alidi y o compe ing models and hei assump ions. Fo ins ance, in a bi a ia e ec o au o eg ession o US TFP and s ock p ices, BP ind wo almos co-linea shocks unde wo pola iden i ying assump ions. The i s iden i ying assump ion imposes ha he e exis s a shock which does no al e TFP in he sho un, he al e na i e iden i ying assump ion imposes ha he e exis s a shock which does no al e TFP in he e y long- un. The wo co-linea shocks a e consis en wi h he o me bu inconsis en wi h he la e . Hence, BP conjec u e ha hey ep esen a echnological inno a ion which a ec s TFP wi h conside able delay. Howe e , his echnological inno a ion a ec s s ock p ices immedia ely and may he e o e cause expec a ions-d i en luc ua ions in consump ion and in es men . The idea o a p ominen ole o echnology- ela ed news in mac oeconomic luc ua ions has ecen ly gi en ise o qui e a ew o he pape s, e. g. Lo enzoni (2006) and Jaimo ich and Rebelo (2006). I is he e o e e y in e es ing o in es iga e i he empi ical inding o BP in a o o news-d i en business cycles is a obus business cycle ac which can be documen ed o o he coun ies and samples as well. Mo eo e , one would like o know i he e is any kind o di ec e idence which suppo s he in e p e a ion o he iden i ied shocks as being echnology shocks. In he i s line o esea ch, Beaud y and Po ie (2005) ook he lead by epea ing hei analysis wi h Japanese da a. He e hey came up wi h essen ially he same inding as o he US: Two almos co-linea “ echnology” shocks unde al e na i e iden i ying assump ions. In his pape , we look a Ge many as a hi d coun y and p esen a simila , bu sligh ly weake piece o e idence: The e is e idence o a g adually inc easing esponse o TFP o ce ain shocks in excess o a clea ly posi i e e ec on impac . The e o e, i he iden i ied shocks a e indeed echnological, he o e all e idence is qui e suppo i e o a s ylized business cycle ac o delayed TFP esponse o echnology shocks. Going u he , we es whe he he iden i ied shocks a e igh ly conside ed as echnology shocks by con on ing hem wi h da a on pa en s g an ed by o applied o a he Ge man pa en agency. Fo a ious measu es o TFP and pa en s, we ha e e y obus e idence ha 2 he iden i ied shocks G ange -cause pa en s. Con e sely, he iden i ied non-pe sis en shock in he SVAR-app oach is no G ange -causal in any o he speci ica ions we es . This seems o be ai ly s ong e idence o he hypo hesis ha he iden i ied shock, which a ec s TFP on impac and wi h a delay, is indeed a echnology shock. The sequel o he pape is o ganized as ollows: In sec ion 2, we illus a e ha he BP app oach can be used o in e key model p ope ies. We p esen a modi ied Long and Plosse (1983) model which allows o mul i-pe iod usage o capi al goods. We compu e s ock p ices as he discoun ed sum o expec ed e u ns o capi al. We hen show ha a s anda d speci ica ion o TFP shocks has e y di e en implica ions o he shocks iden i ied in he Blancha d-Quah app oach han a delayed-implemen a ion speci ica ion. In sec ion 3, we b ie ly illus a e he econome ic app oach and apply i o Ge man da a. We use h ee di e en measu es o TFP in o de o check he obus ness o ou esul s. Sec ion 4 analyzes G ange - causali y be ween he iden i ied shocks and di e en measu es o pa en s. Sec ion 5 concludes. 2 The Model We will illus a e he po en ial o he BP app oach by conside ing wo e sions o he same model: One wi h a s anda d, ins an aneous eac ion o TFP o a echnological inno a ion, he o he wi h a delayed esponse. This is simila o BP (2005). Howe e , hei model assumes 100% dep ecia ion on physical capi al, which makes i di icul o model s ock p ices, since, essen ially, i ms shu down each pe iod. Hence, BP (2005) do no conside s ock p ices bu a he ocus on bonds whose p ice is in e sely ela ed o he e u n on he (one-pe iod) capi al goods. By con as , we use a model whe e he p oduc i e use o capi al goods ex ends o e many pe iods and s ock p ices a e compu ed as he discoun ed sum o expec ed e u ns o capi al. The model is aken om Long and Plosse (1983). We agg ega e hei model o jus a single sec o , bu ex end i s p oduc ion echnology o a mul i-pe iod se ing. Speci ically, in es men goods I can be used o p+1 pe iods un il hey a e comple ely wo n ou . The p oduc ion elas ici ies o in es men o pe iod - τ is gi en by a τ and we can allow o any kind o dep ecia ion schedule by secu ing 1 aa p ττ τ + >∀<. Labo inpu is L wi h p oduc ion elas ici y b>0, so ha cons an e u ns imply 01 pab τ τ = + = ∑ . 1 + Λ is TFP o pe iod +1 and p oduc ion is gi en by 11 0 p a b YLI τ τ τ + + = =Λ − ∏ (1) o (using small le e s o deno e logs) 1 0 p ybl ai ττ τ 1 λ + − = =+ + ∑+ 1 . (2) The ep esen a i e agen has a s anda d in e empo al u ili y unc ion wi h subjec i e discoun ac o 0 β << and isk a e sion cap u ed by 1 σ >. 2 e η θ = is a s a iona y p e e ence shock and he inno a ion 2 η is, o simplici y, whi e noise wi h uni a iance. 3 00 ln L UE C σ βθ σ ∞ = ⎡ ⎤ ⎛⎞ =− ⎢ ⎥ ⎜⎟ ⎝⎠ ⎣ ⎦ ∑ Maximizing u ili y unde he budge cons ain CIY + = (3) yields i s o de condi ions 1 1 00 11 1: pp II aE aE CC ττ τ τ ττ ττ βγβ ++ ++ ++ == ++ ++ 1 1 I C τ τ ⎡ ⎤⎡ =+=+ ⎤ ⎢ ⎥⎢⎥ ⎣ ⎦⎣ ∑∑ ⎦ (4) and 1 1 1 I LbEC σ θβ + + ⎡ ⎤ =+ ⎢ ⎥ ⎣ ⎦. (5) Imposing s abili y, sol ing (4) o wa d and using (3) yields ( ) 1, CYI Y γ γ =− = . (6) This is he policy unc ion, since is a s a e a iable. Fo labo , we compu e Y 2 1 ln ln1 b lL β η σγ ⎛ == − ⎜ − ⎝⎠ ⎞ ⎟ Y . (7) Inse ing (6) in o (1) we ge 11 0 p a ab YL τ τ τ γ + + = =Λ − ∏ , (8) whe e 0 :p aa τ τ = =∑. Se ing () :ln ln 1 bb a β κγ σ γ =+ − and aking logs we ge : () 1121 0 : p b yay aLy ττ τ 12 b κ ληκ λ σ +−+ ++ = =+ + − =+ + − ∑ η σ (9) () () 21 21 : 11 1 b b yA aaL a λη κκ σλη σ − − ⎛⎞ − ⎜⎟ ⎛ ⎝⎠ =+ =+ − ⎜ −− − ⎝⎠ L ⎞ ⎟ , (10) whe e a e polynomials in he lag ope a o L and () () ,aL AL ( )() 00,0aA1 = =. Ne p o i s a e ou pu minus labo and in es men cos s: () 1 11 1 :1 Y YLIb L γ + ++ + 1 Y + ∂ Π= − − =−− ∂ (11) 4 Le 1 : SP E τ τ τ β ∞ + = ⎡ =Π ⎢ ⎣⎦ ∑⎤ ⎥ ] 1+ be he s ock p ice, i. e. he discoun ed sum o p o i s expec ed in pe iod . By de ini ion, we ha e [] [] [ 111 11 : SP E E E E E SP ττ ττ ττ ββ ββ β β ∞∞ ++ +++ == ⎡⎤ ⎡ ⎤ = Π=Π+ Π=Π+ ⎢⎥ ⎢ ⎥ ⎣⎦ ⎣ ⎦ ∑∑ . Log-linea izing his equa ion we ge : () () () () () () ( ) () () 11 1 0 11 0 12 1 0 1 1 1 :1 :1 sp cns E E sp cns E b cns E A L cns L E A L τ τ τ τ ττ τ τ τ τ βπ β ββπ β ββ λ η σ δη β β λ ++ ∞ ++ = ∞ ++ + = ∞ ++ = ≈+− + ⎛⎞ =+− ⎜⎟ −⎝⎠ ⎛ ⎛ =+− − ⎜⎟ ⎜⎟ ⎝⎠ ⎝⎠ =− +− ∑ ∑ ∑ 2 , ⎞ ⎞ (12) whe e . () 00 δ ≠ Le us now conside wo di e en speci ica ions o he s ochas ic p ocesses d i ing he model. The s anda d speci ica ion would be a andom walk o log TFP: 11 λ λη − = + (13) He e, 1 η is assumed o be whi e noise wi h uni a iance. The al e na i e speci ica ion would speci y a delayed esponse o log TFP o pe manen echnological inno a ions. Assume ha log TFP is he sum o a andom walk componen ζ and a s a iona y p ocess . The e a e o hogonal, uni - a iance-whi e noise inno a ions 1 η and 3 η o ζ and , espec i ely. The 3 η inno a ion a ec s log TFP in he same pe iod in which i becomes known, while we assume ha he 1 η inno a ion a ec s TFP wi h a delay o one pe iod. Thus, he s ochas ics a e desc ibed by 1 11 13 ,1 λζ ζζ η ρηρ − − − = + =+ = +< (14) To p ese e he same numbe o shocks as unde he s anda d speci ica ion, we assume ha he e is no p e e ence shock in he delayed- esponse speci ica ion, 20 η = ∀. Unde he s anda d speci ica ion we de i e om (12) 5 ( ) ( ) ()( ) ()( ) () ()( ) 12 2 12 11 1, sp cns L A L sp A L L L L AL L L δη λ λ δη ηδ η ≈− + Δ≈ − − − =−−    whe e () A L  is a lag polynomial wi h ( ) ( ) 00 1 A A≠≠  . Hence, he mo ing a e age ep esen a ion is gi en by () ()( ) () 1 1 22 10 : 1 CL AL L L sp 1 λ ηη δ η η Δ⎛⎞ ⎛ ⎞ ⎛⎞ ⎛⎞ = ⎜⎟ ⎜ ⎟ ⎜⎟ ⎜ −− Δ ⎝ ⎠ ⎝⎠ ⎝⎠ ⎝⎠ = ⎟ . (15) Unde he delayed- esponse speci ica ion we know ( ) ( ) () () 2 23 1 sp cns A L B L BL cns A L L ζ ζη ρ ≈+ + =+ + −    wi h . The mo ing a e age ep esen a ion o he i s di e ences is hen gi en by () () 00 1B≠≠  B  ( ) () () () () 1 2 33 1 1: 1 1 L LLCL sp BL AL L L 1 λ ηη ρ η η ρ ⎛−⎞ ⎜⎟ Δ− ⎛ ⎞ ⎛⎞ ⎛⎞ ⎜⎟ = ⎜ ⎟ ⎜⎟ ⎜ ⎜⎟ Δ ⎝ ⎠ ⎝⎠ ⎝⎠ ⎜⎟ − ⎜⎟ − ⎝⎠   = ⎟ . (16) F om (15) and (16) we in e ha he i s ow o bo h ( ) 11C and ( ) 21C is ( , while he i s ow o and a e e y di e en : ) 1, 0 () 10C () 20C ( ) 1, 0 and ( ) 0,1 , espec i ely. This implies ha he iden i ica ion o he s uc u al shocks gi es ise o di e en pa e ns, oo, and hese pa e ns can be used o in e wha he ue unde lying model is. We will discuss his in de ail in he nex sec ion. 3 The Econome ic App oach Conside empi ical ime se ies o log TFP and log s ock p ices, deno ed λ and as be o e. We assume hese a e in eg a ed o o de one and coin eg a ed wi h each o he , i. e. is I(0). Using Wold’s decomposi ion heo em, sp ( , sp λ ΔΔ ) ' ( ) , sp λ ΔΔ ' ⎞ ⎟ ⎠ can be w i en in educed o m wi h () 1 2 u CL sp u λ Δ ⎛⎞ ⎛ = ⎜⎟ ⎜ Δ ⎝⎠ ⎝ 1 (): i i i CL I CL ∞ = =+ ∑ (17) and in s uc u al o m 6 () 1 2 DL sp λ ε ε Δ ⎛⎞ ⎛ = ⎜⎟ ⎜ Δ ⎝⎠ ⎝ ⎞ ⎟ ⎠ wi h . (18) 0 (): i i i DL DL ∞ = =∑ Iden i ying he s uc u al shocks ε equi es knowledge o D0. This ma ix can be eco e ed om he es ima ed ( ) CL ma ices o he educed o m (17), i one es ic ion is imposed on he pa ame e s o ( ) D L. We ollow Beaud y and Po ie (2005, 2006) by using wo al e na i e assump ions, which we call he sho - un and he long- un es ic ion. The o me pos ula es ha he (1,2) elemen o ( ) 0D is ze o, i. e. he s ock ma ke shock 2 ε has no e ec on TFP on impac . The la e pos ula es ha he (1,2) elemen o is ze o, i. e. he s ock ma ke shock () 1D 2 ε has no long- un e ec on TFP. Le us hink o (18) as he ep esen a ion ob ained unde he sho - un es ic ion and le (19) be he ep esen a ion ob ained unde he long- un es ic ion: 1, 2, () DL sp ε λ ε Δ⎛⎞ ⎛⎞ =⋅ ⎜ ⎜⎟ Δ ⎝⎠ ⎝⎠   ⎟ wi h 0 (): i i i DL D L ∞ = = ⋅ ∑  (19) I he empi ical da a we e gene a ed by he model o sec ion 2 wi h s anda d speci ica ion, i. e. by equa ion (15), he impac ma ix () () () 1 10 000 CA δ ⎛⎞ =⎜⎟ − ⎝⎠  would al eady ul ill he sho - un iden i ying assump ion, hence, as s uc u al shocks we would iden i y 11 2 ,2 ε ηεη ==. On he o he hand, he long- un ma ix is () () 1 10 110 CA ⎛⎞ =⎜⎟ ⎝⎠ , hus, we would immedia ely ha e 11 2 ,2 ε ηεη = =  . The impo an poin is ha unde bo h iden i ying assump ions we would ind he same esul o 1 ε and 1 ε . I , con e sely, he empi ical da a we e gene a ed by he delayed echnology speci ica ion, he impac ma ix would be () () () 2 01 000 CAB ⎛⎞ =⎜⎟ ⎝⎠ . In his case, he iden i ying assump ions imply ha 13 2 , 1 ε ηεη = =, whe eas unde he long- un es ic ion we ha e () () 2 10 110 CA ⎛⎞ =⎜⎟ ⎝⎠  7 and hence 11 2 , 3 ε ηεη ==  . Unde his model, we would hus no ind he same esul o 1 ε and 1 ε , bu a he we would ind ha 21 ε ε = . The e o e, i he empi ical analysis sugges s ha 21 ε ε ≈ we may in e ha a model wi h delayed echnology esponse is mo e app op ia e han a s anda d speci ica ion. We now u n o an in es iga ion o his issue o Ge man da a. 3.1 Da a desc ip ion As in Beaud y and Po ie (2006), h ee di e en TFP a iables a e calcula ed: he s anda d Solow esidual, he Solow esidual adjus ed o a iable capi al u iliza ion and a TFP measu e ollowing he me hodology o G o h e al. (2004) and Oul on (2001). We ha e qua e ly da a om 1970(1) o 2005(2). The simple TFP measu e (wi hou capi al u iliza ion) is compu ed om da a on GDP, hou s wo ked and annual capi al s ock da a in e pola ed wi h cons an wi hin yea qua e ly g ow h a es. Unde he assump ion o cons an e u ns o scale, he obse ed qua e ly labo sha e is used as he p oduc ion elas ici y o labo . The log o his measu e is deno ed TFP_D1. Modi ying he capi al s ock da a by mul iplying wi h he capaci y u iliza ion a e in manu ac u ing gi es a second measu e o TFP whose log is deno ed TFP_D2. Compu a ion o he hi d TFP measu e (TFP_D3) akes se e al c i icisms o he s anda d Solow esidual in o accoun . Quali y aspec s a e conside ed when measu ing labo inpu . Unde he assump ion o pe ec compe i ion, he quali y o wo k is e lec ed by wages. The e o e, quali y adjus ed labo inpu L can be cons uc ed as 1 1 ˆ ˆ2 n i i i ss i L h − = + ⎛⎞ = ⋅ ⎜⎟ ⎝⎠ ∑, (20) whe e ha ed a iables a e g ow h a es, n cons i u es he numbe o employee ca ego ies, si desc ibes ou pu con ibu ion and hi wo king hou s o g oup i in pe iod . In he case o Ge many, ele an da a in e ms o g oss ea nings exis o sala ied employees o he se ice and manu ac u ing sec o s, and in e ms o g oss wages o wage ea ne s o manu ac u ing and ag icul u e. F om mic oda a, ou ca ego ies o labo inpu can be dis inguished. The concep o capi al inpu does no e e o he capi al s ock bu uses a measu e o capi al se ices. Di e en ypes o asse s a e weighed by hei en al p ices o ep esen he alue o se ices which can be ealized a pe ec compe i ion. Ren al p ices MP o asse ype j in pe iod a e, in p inciple, compu ed as ,, , , , ˆ j j j j j j M PT p p δ ⎡ ⎤ ⎛⎞ =⋅ +− ⋅ ⎜⎟ ⎢ ⎥ ⎝⎠ ⎣ ⎦, (21) whe e j, a e oppo uni y cos s, δj is dep ecia ion and Tj, exp esses axa ion and in es men allowances1. 1 As in Oul on (2001) dwellings a e excluded om buildings because dwellings may no con o m wi h s ic p o i maximizing beha io . Dep ecia ion a es o machine y and buildings a e assumed o be 13% and 2.5%. Ma ke p ices o bo h asse s esul om he a io o nominal and eal alues o he espec i e g oss ixed capi al o ma ion. Ra es o e u n a e compu ed by he a io o g oss ope a ing su plus and capi al s ock alue. The ax ac o is dis ega ded due o lack o adequa e da a. 8 hence many o he applica ions do appa en ly no ep esen in en ions. Also, PAG is e y s ongly skewed due o a ew obse a ions in he la e 1990s, when pa en applica ions we e inc easing emendously. The e o e, any kind o in e ence is e y di icul o PAG and we ha e ocused on PAT. We do no epo he esul s in de ail, bu hey a e a ailable upon eques . Su ice i o say ha no G ange -causali y can be ound be ween PAT and any o he iden i ied shocks. This is wha one would expec i in en o s y o keep hei in en ions sec e as long a possible, bu he in en ions a e disclosed a e submi ing he pa en applica ion. (Recall ha he ime lag be ween submission and disclosu e is a mos 18 mon hs, i i is sho e , i may well all wi hin he same yea .) One migh c i icize ou p ocedu e o simply agg ega ing qua e ly shocks o yield annual equency. In o de o check o a possible empo al agg ega ion bias, we es ima e he bi a ia e VARs also wi h annual da a. Mo eo e , he equency con e sion may se e as a es o he obus ness o ou esul s. We hus ecalcula e he ime se ies TFP_D1b, TFP_D2b, TFP_D3b and DAX1b wi h annual da a. Looking a he esul s o he Johansen ace es , we now do no ind e idence o coin eg a ion, in ma ked con as o ou esul s wi h qua e ly da a. This is an immedia e con adic ion, since coin eg a ion o he qua e ly se ies would imply ha he annual se ies a e also coin eg a ed. In iew o he ela i ely low p- alues o he Johansen es s a is ics o annual da a one may hus suspec ha he Johansen es s o annual da a su e om low powe . We he e o e e e se he es p ocedu e by es ing o he null hypo hesis o coin eg a ion using he Shin (1994) es . This es is also no able o ejec he null hypo hesis and his inding is obus o a ious choices o he unca ion lag. Consequen ly, he e is s ill no con incing e idence agains he iew ha TFP and s ock ma ke p ices a e coin eg a ed. We he e o e edo he analysis in comple e analogy o ou handling o qua e ly da a. All ele an esul s o VECM es ima ion and impulse esponse analysis a e p esen ed in he appendix, as hey a e quali a i ely e y simila o he qua e ly esul s. The esul ing annual s uc u al esiduals can hen be used in G ange causali y es s which a e no subjec o empo al agg ega ion bias. The esul s a e gi en in Tables 3 and 4. Basically, he conclusions om he ea lie es s a e unal e ed. In mos cases, he iden i ied echnology shocks a e G ange -causal o g an ed pa en s and in he ew cases whe e his is no he case, he p- alues a e e y close o 5%. The con e se is no ue, i. e. nowhe e a e g an ed pa en s G ange - causal o he echnology shocks. Mo eo e , he e is no causali y in ei he di ec ion be ween g an ed pa en s and he non- echnology shocks. Hence he a ailable indica es ha he SVAR analysis co ec ly iden i ies echnology and non- echnology inno a ions o he Ge man economy. 15 Table 3: G ange Causali y Tes s (iden i ied shocks om annual VECM) Shocks suspec ed o be echnological PGG PGT ε 20.060 0.895 0.037 0.313 TFP_D1b 1 ε  0.051 0.814 0.036 0.632 ε 20.050 0.731 0.029 0.223 TFP_D2b 1 ε  0.015 0.826 0.016 0.263 ε 20.028 0.662 0.033 0.142 TFP_D3b 1 ε  0.020 0.372 0.067 0.052 Uppe le co ne : P- alue o null: Row a iable does no G ange -cause column a iable. Lowe igh co ne : P- alue o null: Column a iable does no G ange -cause ow a iable. Table 4: G ange Causali y Tes s (iden i ied shocks om annual VECM) Shocks suspec ed o ha e no pe manen e ec on TFP PGG PGT ε 10.397 0.836 0.351 0.446 TFP_D1b 2 ε  0.417 0.999 0.3800 0.945 ε 10.136 1.000 0.190 0.701 TFP_D2b 2 ε  0.636 0.762 0.463 0.488 ε 10.221 0.500 0.421 0.236 TFP_D3b 2 ε  0.357 0.880 0.281 0.687 Uppe le co ne : P- alue o null: Row a iable does no G ange -cause column a iable. Lowe igh co ne : P- alue o null: Column a iable does no G ange -cause ow a iable. 5 Conclusions The pu pose o his s udy was o p o ide u he e idence on he BP-hypo hesis o delayed- echnology di usion and news-d i en business cycles. Fo Ge man da a on TFP and s ock p ices we ind quali a i ely he same esul as BP do: A high co ela ion be ween a shock wi h pe manen e ec s on TFP in he long un and – unde a di e en iden i ica ion scheme - a shock which has an immedia e e ec on s ock p ices bu does no a ec TFP on impac . The co ela ion is less p onounced, hough, as in BP’s analysis o US and Japanese da a. Also, he impulse esponse analysis sugges s ha o Ge many a subs an ial pa o he o al TFP esponse is immedia e a he han delayed. Using a quali y-adjus ed measu e o TFP, he e is almos no delayed di usion any mo e. This sugges s ha he delayed di usion is 16 con ined o embodied echnological change. Disembodied echnological inno a ions seem o ha e immedia e e ec s on TFP. Bu since he sha e o disembodied echnological p og ess in o al echnological p og ess may be small, he ela i ely la ge size o he quali y-adjus ed TFP measu e sugges s, ha each measu e o Ge man TFP may ac ually con ain a ai ly la ge pa o unexplained non- echnology in luences, i. e. ou igno ance abou he ue na u e o wha we measu e as TFP may be ai ly la ge. I may he e o e be he case ha noise in TFP da a is esponsible o he immedia e eac ion o s anda d TFP measu es in he impulse esponse analysis. Gi en possibly noisy da a on TFP, we hen checked how well he iden i ica ion o echnology shocks in he SVAR app oach wo ked. The answe seems o be: Su p isingly well. Shocks suspec ed o be echnology shocks a e G ange -causal o he numbe o pa en s g an ed by he Ge man pa en agency, while shocks wi hou pe manen e ec on echnology a e no . This esul is e y obus ac oss di e en speci ica ions, measu es and iden i ica ion schemes. I may he e o e be he case ha he SVAR app oach is able o sepa a e he ue, pe manen echnology shocks om ansi o y noise which also a ec s measu ed TFP. Unde his in e p e a ion, ou esul s a e qui e suppo i e o BP’s news-d i en business cycle hypo hesis. 17 6 Appendix Resul s o qua e ly da a: Table A1 ADF es Va iable Le el / 1s di . Lags Tes s a is ics C i ical alue (5%) p- alue Le el 0 -2.55 -3.44 0.3041 TFP_D1b 1s di . 0 -14.00 -3.44 0.0000 Le el 1 -2.64 -3.44 0.2643 TFP_D2b 1s di . 0 -15.86 -3.44 0.0000 Le el 1 -1.70 -3.44 0.7483 TFP_D3b 1s di . 0 -15.11 -3.44 0.0000 Le el 0 -2.60 -3.44 0.2810 DAX1b 1s di . 0 -10.34 -3.44 0.0000 Table A2 Johansen ace es Va iables Lag leng h (1s di .) T end in e o co ec ion e m (EC) / O hogonal T end (OT) Hypo hesis Tes s a is ics C i ical alue (5%) p- alue = 0 16.24 15.49 0.0385 OT = 1 1.08 3.84 0.2991 TFP_D1b & DAX1b 0 EC = 0 19.69 25.87 0.2420 = 0 19.29 15.49 0.0128 OT = 1 2.08 3.84 0.1491 TFP_D2b & DAX1b 1 EC = 0 24.10 25.87 0.0817 = 0 15.57 15.49 0.0487 OT = 1 1.58 3.84 0.2089 TFP_D3b & DAX1b 0 EC = 0 19.78 25.87 0.2372 18 Es ima ion esul s o coe icien s in VECM {TFP_D1b, DAX1b}1 () 11 12 _1 _1 0.021* 0.046* 1.000 0.220* 11 0.203* 0.399* TFP D b TFP D b u cns D AX b DAX b u − − Δ− ⎛⎞ ⎛⎞ ⎛⎞ ⎛⎞ =− + ⎜⎟ ⎜⎟ ⎜⎟ ⎜⎟ Δ− ⎝⎠ ⎝⎠ ⎝⎠ ⎝⎠ ⎛⎞ + ⎜⎟ ⎝⎠ Tes ype Tes s a is ics p- alue Au oco . 65.91 0.3432 He e osc. 9.99 0.9322 Non- No mal. 69.54 0.0000 Implied SVAR coe icien s1 sho - un es ic ion: long- un es ic ion: 0 0.0080* 0.0000 0.0003 0.0979* D⎛⎞ =⎜⎟ − ⎝⎠ i 00.0049* 0.0063* 0.0766* 0.0609* D− ⎛⎞ =⎜⎟ ⎝⎠ 1 *=signi ican a he 5% le el. 19 Es ima ion esul s o coe icien s in VECM {TFP_D2b, DAX1b}1 () 1 1 _2 _2 0.023* 1.000 0.187* 11 0.194* TFP D b TFP D b DAX b DAX b − − Δ− ⎛⎞ ⎛ ⎛⎞ =− ⎜⎟ ⎜ ⎜⎟ Δ⎝⎠ ⎝⎠ ⎝ ⎞ ⎟ ⎠ 11 12 _2 0.287* 0.000 0.046* 1 2.241* 0.141 0.346* TFP D b u cns D AX b u − − Δ −⎛⎞ ⎛⎞ ⎛⎞ ++ ⎜⎟ ⎜⎟ ⎜⎟ Δ− ⎝⎠ ⎝⎠ ⎝⎠ ⎛⎞ + ⎜⎟ ⎝⎠ Tes ype Tes s a is ics p- alue Au oco . 55.33 0.5753 He e osc. 16.34 0.5690 Non- No mal. 58.09 0.0000 Implied SVAR coe icien s1 sho - un es ic ion: long- un es ic ion: 0 0.0081* 0.0000 0.0075 0.0947* D⎛⎞ =⎜⎟ ⎝⎠ i 00.0051* 0.0063* 0.0783* 0.0537* D− ⎛⎞ =⎜⎟ ⎝⎠ Impulse- esponse unc ions2 a) Sho - un es ic ion ( esponse o ε2): b) Long- un es ic ion ( esponse o 1 ε ): 1 *=signi ican a he 5% le el. 2 Con idence in e als a e ob ained by 2500 eplica ions wi h he boo s apping p ocedu e o Hall (1992). They ep esen he 95% quan iles. 20 Es ima ion esul s o coe icien s in VECM {TFP_D3b, DAX1b}1 () 11 12 _3 _3 0.033* 0.060* 1.000 0.186* 11 0.258* 0.436* TFP D b TFP D b u cns D AX b DAX b u − − Δ− ⎛⎞ ⎛⎞ ⎛⎞ ⎛⎞ =− + ⎜⎟ ⎜⎟ ⎜⎟ ⎜⎟ Δ− ⎝⎠ ⎝⎠ ⎝⎠ ⎝⎠ ⎛⎞ + ⎜⎟ ⎝⎠ Tes ype Tes s a is ics p- alue Au oco . 74.70 0.1292 He e osc. 14.12 0.7215 Non- No mal. 58.89 0.0000 Implied SVAR coe icien s1 sho - un es ic ion: long- un es ic ion: 0 0.0129* 0.0000 0.0021 0.0974* D⎛⎞ =⎜⎟ ⎝⎠ i 00.0094* 0.0088* 0.0681* 0.0697* D− ⎛⎞ =⎜⎟ ⎝⎠ 1 *=signi ican a he 5% le el. 21 Resul s o annual da a: Table A3 ADF es Va iable Le el / 1s di . Lags Tes s a is ics C i ical alue (5%) p- alue Le el 1 -2.16 -3.55 0.4970 TFP_D1b 1s di . 0 -3.12 -2.95 0.0344 Le el 1 -2.38 -3.55 0.3847 TFP_D2b 1s di . 0 -2.75 -2.95 0.0763 Le el 0 -2.04 -3.55 0.5578 TFP_D3b 1s di . 0 -5.01 -2.95 0.0003 Le el 1 -3.01 -3.55 0.1437 DAX1b 1s di . 0 -3.94 -2.95 0.0048 Table A4 Johansen ace es Va iables Lag leng h (1s di .) T end in e o co ec ion e m (EC) / O hogonal T end (OT) Hypo hesis Tes s a is ics C i ical alue (5%) p- alue OT = 0 12.00 15.49 0.1570 TFP_D1b & DAX1b 1 EC = 0 18.32 25.87 0.3227 OT = 0 12.29 15.49 0.1435 TFP_D2b & DAX1b 1 EC = 0 17.52 25.87 0.3769 OT = 0 14.39 15.49 0.0729 TFP_D3b & DAX1b 0 EC = 0 17.93 25.87 0.3485 22 Table A5 Shin es (LM es s a is ics o TFP_D1b & DAX1b) Lagged di e ences 2 4 6 2 0.140 0.116 0.054 4 0.112 0.103 0.061 Lag unca ions 8 0.127 0.124 0.160 C i ical alue a he 5% le el: 0.314 No linea end assump ion unde null hypo hesis Table A6 Shin es (LM es s a is ics o TFP_D2b & DAX1b) Lagged di e ences 2 4 6 2 0.138 0.109 0.049 4 0.112 0.096 0.072 Lag unca ions 8 0.127 0.125 0.113 C i ical alue a he 5% le el: 0.314 No linea end assump ion unde null hypo hesis Table A7 Shin es (LM es s a is ics o TFP_D3b & DAX1b) Lagged di e ences 2 4 6 2 0.127 0.062 0.042 4 0.103 0.057 0.043 Lag unca ions 8 0.131 0.117 0.165 C i ical alue a he 5% le el: 0.314 No linea end assump ion unde null hypo hesis 23 Es ima ion esul s o coe icien s in VECM {TFP_D1b, DAX1b}1 () 1 1 _1 _1 0.035 1.000 0.226* 11 0.700* TFP D b TFP D b DAX b DAX b − − Δ− ⎛⎞ ⎛ ⎛⎞ =− ⎜⎟ ⎜ ⎜⎟ Δ⎝⎠ ⎝⎠ ⎝ ⎞ ⎟ ⎠ 11 12 _1 0.418* 0.011 0.079* 1 2.601 0.346* 1.336* TFP D b u cns D AX b u − − Δ ⎛⎞ ⎛⎞ ⎛⎞ ++ ⎜⎟ ⎜⎟ ⎜⎟ Δ −− ⎝⎠ ⎝⎠ ⎝⎠ ⎛⎞ + ⎜⎟ ⎝⎠ Tes ype Tes s a is ics p- alue Au oco . 8.40 0.5894 He e osc. 6.75 0.6636 Non- No mal. 2.37 0.6680 Implied SVAR coe icien s1 Iden i ied shocks sho - un es ic ion: 0 0.0084* 0.0000 0.0296 0.1492* D⎛⎞ =⎜⎟ ⎝⎠ long- un es ic ion: i 00.0067* 0.0051* 0.1134* 0.1013* D− ⎛⎞ =⎜⎟ ⎝⎠ Impulse- esponse unc ions2 a) Sho - un es ic ion ( esponse o ε2): b) Long- un es ic ion ( esponse o 1 ε ): 1 *=signi ican a he 5% le el. 2 Con idence in e als a e ob ained by 2500 eplica ions wi h he boo s apping p ocedu e o Hall (1992). They ep esen he 95% quan iles. 24