School of Economics and Management

Transcription

1 School of Economics and Managemen TECHNICAL UNIVERSITY OF LISBON Deparmen of Economics Carlos Pesana Barros & Nicolas Peypoch Anónio Afonso and Ricardo M. Sousa A Comparaive Analysis of Produciviy Change in Ialian and Fiscal Policy, Poruguese Housing Airpors and Sock Prices WP 58/28/DE/UECE WP 6/27/DE WORKING PAPERS ISSN Nº

2 Fiscal Policy, Housing and Sock Prices * Anónio Afonso # and Ricardo M. Sousa $ December 28 Absrac This paper invesigaes he link beween fiscal policy shocks and movemens in asse markes using a Fully Simulaneous Sysem approach in a Bayesian framework. Building on he works of Blanchard and Peroi (22), Leeper and Zha (23), and Sims and Zha (1999, 26), he empirical evidence for he U.S., he U.K., Germany, and Ialy shows ha i is imporan o explicily consider he governmen deb dynamics when assessing he macroeconomic effecs of fiscal policy and is impac on asse markes. In addiion, he resuls from a VAR couner-facual exercise sugges ha: (i) fiscal policy shocks play a minor role in he asse markes of he U.S. and Germany; (ii) hey subsanially increase he variabiliy of housing and sock prices in he U.K..; and (iii) governmen revenue shocks have apparenly conribued o an increase of volailiy in Ialy. Keywords: Bayesian Srucural VAR, fiscal policy, housing prices, sock prices. JEL classificaion: C32, E62, G1, H62. * We are graeful o Jürgen von Hagen and o an anonymous referee for helpful commens and suggesions and o Silvia Albrizio and Mahijs Lof for research assisance. The opinions expressed herein are hose of he auhors and do no necessarily reflec hose of he ECB or he Eurosysem. # European Cenral Bank, Direcorae General Economics, Fiscal Policies Division, Kaisersraße 29, D Frankfur am Main, Germany. ISEG/TULisbon - Technical Universiy of Lisbon, Deparmen of Economics; UECE Research Uni on Complexiy and Economics; R. Miguel Lupi 2, Lisbon, Porugal. UECE is suppored by FCT (Fundação para a Ciência e a Tecnologia, Porugal), financed by ERDF and Poruguese funds. s: anonio.afonso@ecb.europa.eu, aafonso@iseg.ul.p. $ Universiy of Minho, Deparmen of Economics and Economic Policies Research Uni (NIPE), Campus of Gualar, Braga, Porugal; London School of Economics, Deparmen of Economics and Financial Markes Group, Houghon Sree, London WC2 2AE, Unied Kingdom. rjsousa@eeg.uminho.p; r.j.sousa@lse.ac.uk. Ricardo Sousa would like o hank he Fiscal Policies Division of he ECB for is hospialiy. 1

3 Conens Non-echnical summary Inroducion Idenificaion of fiscal shocks A Fully Simulaneous Sysem approach Resuls and discussion Daa VAR resuls A VAR couner-facual exercise Conclusion References... 2 Appendix A. Assessing poserior uncerainy in a Fully Simulaneous SVAR Appendix B. Daa sources

4 Non-echnical summary This paper analyzes he effecs of fiscal policy on economic aciviy, wih an emphasis on asse markes. In paricular, we ask how sock prices and housing prices are affeced by fiscal policy shocks, and look a he persisence of he effecs. We idenify fiscal policy shocks using informaion abou he elasiciy of fiscal variables o he economic aciviy. In addiion, we esimae a Fully Simulaneous Sysem approach in a Bayesian framework, herefore, accouning for he poserior uncerainy of he impulse-response funcions. Anoher added value of he paper is ha we explicily include he feedback from governmen deb in our esimaions. In addiion, we use quarerly fiscal daa o analyze empirical evidence from he U.S., he U.K., Germany, and Ialy respecively for he periods 197:3-27:4; 1971:2-27:4; 1979:2-26:4; 1986:2-24:4. The se of quarerly fiscal daa is aken from naional accouns (in he case of he U.S. and he U.K.) or based on fiscal cash daa (for Germany and Ialy). The main resuls of our work can be summarized as follows. Governmen spending shocks: (i) have a posiive and persisen effec on GDP in he case of he U.S. and he U.K., while for Germany and Ialy, he (posiive) impac is emporary and becomes negaive afer 4 o 8 quarers; (ii) have a posiive and persisen effec on housing prices, alhough housing markes end o respond wih a lag of around 4 quarers; (iii) have a negaive effec on sock prices, alhough he ime of reacion is faser han for housing prices; (iv) have posiive effecs on he price level in he case of he U.K. and Ialy, and negaive effecs for he U.S. and Germany; and (v) reduce unemploymen only in he U.S. On he oher hand, governmen revenue shocks: (i) have an iniial negaive effec on GDP ha laer becomes posiive; (ii) have a negaive impac on housing prices for he U.S. and Ialy, and a posiive impac for he U.K. and Germany; (iii) have a small and posiive effec on sock prices; (iv) have, in general, a negaive and persisen effec on he price level; and (iv) have a posiive and persisen impac on he unemploymen rae. When we explicily ake ino accoun he link beween governmen deb and deficis, including he feedback from governmen deb, long-erm ineres raes and GDP are more responsive and he effecs on hese variables also become more persisen. 3

5 Finally, in a VAR couner-facual exercise, we show ha fiscal policy shocks play a minor role in he paerns ha one observes for sock prices and housing prices in he U.S. and Germany. Neverheless, while boh spending and revenue shocks seem o have an imporan effec on asse markes for he U.K., for Ialy only governmen revenue shocks have conribued o an increase of volailiy in housing and sock prices, in paricular, in he nineies. 4

6 1. Inroducion This paper evaluaes he effecs of fiscal policy on economic aciviy, wih a paricular emphasis on he linkages beween fiscal policy and asse markes. We ask how sock and housing prices are affeced by fiscal policy shocks, and, o he exen ha we find a link beween hem, we look a he magniude and he persisence of he effecs. We idenify fiscal policy shocks using informaion abou he elasiciy of fiscal policy variables o economic aciviy, herefore, aking ino accoun he auomaic response of governmen spending and revenue o oupu, inflaion, and he ineres rae as in Blanchard and Peroi (22). Moreover, we accoun for he poserior uncerainy of he impulse-response funcions by esimaing a Fully Simulaneous Sysem of equaions in a Bayesian framework based on he works of Leeper and Zha (23), and Sims and Zha (1999, 26). Anoher added value of he paper is ha we explicily include he link beween governmen deb and deficis in our framework, and, consequenly, including he governmen deb feedback dynamics in our esimaions. In his respec, he presen work follows Favero and Giavazzi (28) so we consider he response of fiscal variables o he level of he deb. Finally, using quarerly fiscal daa, we analyze empirical evidence from he U.S., he U.K., Germany, and Ialy respecively for he periods 197:3-27:4; 1971:2-27:4; 1979:2-26:4; 1986:2-24:4. The se of quarerly fiscal daa, is aken from naional accouns (in he case of he U.S. and he U.K.) or based on fiscal cash daa (for Germany and Ialy). The main resuls of our work can be summarized as follows. Governmen spending shocks: (i) have a posiive and persisen effec on GDP in he case of he U.S. and he U.K., while for Germany and Ialy, he (posiive) impac is emporary and becomes negaive afer 4 o 8 quarers; (ii) have a posiive and persisen effec on housing prices, alhough housing markes end o respond wih a lag of around 4 quarers; (iii) have a negaive effec on sock prices, alhough he ime of reacion is faser han for housing prices; (iv) have mixed effecs on he price level, ha is, he response is posiive in he case of he U.K. and Ialy, and negaive for he U.S. and Germany; and (v) reduce unemploymen only in he U.S. On he oher hand, governmen revenue shocks: (i) have an iniial negaive effec on GDP ha laer becomes posiive; (ii) have a negaive impac on housing prices for he US and Ialy, 5

7 and a posiive impac for he U.K. and Germany; (iii) have a small and posiive effec on sock prices; (iv) have, in general, a negaive and persisen effec on he price level; and (iv) have a posiive and persisen impac on he unemploymen rae. When we include he feedback from governmen deb in he esimaions longerm ineres raes and GDP become more responsive, and he effecs on hese variables also become more persisen. Finally, we perform a VAR couner-facual exercise, and show ha fiscal policy shocks play a minor role in he paerns ha one observes for sock prices and housing prices in he U.S. and Germany. Neverheless, while boh spending and revenue shocks seem o have an imporan effec on asse markes for he U.K., in he case of Ialy only governmen revenue shocks have conribued o an increase of volailiy in housing and sock prices, in paricular, in he nineies. The res of he paper is organized as follows. Secion wo briefly reviews idenificaion schemes of fiscal policy shocks in he relaed lieraure. Secion hree explains he empirical sraegy used o idenify he effecs of fiscal policy shocks, and o ake ino accoun he auomaic response of fiscal policy o economic aciviy and he uncerainy regarding he poserior disribuion of he impulse-response funcions. Secion four provides he empirical analysis and discusses he resuls. Secion five concludes wih he main findings and policy implicaions. 2. Idenificaion of fiscal shocks While a large number of sudies have been devoed o he analysis of he macroeconomic effecs of moneary policy, 1 he empirical evidence on he role of fiscal policy as a ool for economic sabilizaion is somewha lagging and here is no consensus abou he idenificaion of fiscal policy shocks. Roemberg and Woodford (1992) idenify exogenous movemens in U.S. governmen purchases wih innovaions o defence purchases. In conras, Ramey and Shapiro (1998) use he narraive approach o isolae poliical evens ha led o hree large miliary build-ups unrelaed o developmens in he U.S. economy. They find ha whils nondurable consumpion displays a small decline, durables consumpion falls persisenly afer a brief rise. In he same vein, Edelberg e al. (1999) show ha specific episodes of miliary build-ups (idenified in Ramey and Shapiro, 1998) have a 1 See, for example, Chrisiano e al. (25), Sims and Zha (1999, 26), and Leeper and Zha (23). 6

8 significan and posiive shor-run effec on U.S. oupu and consumpion, and allowing for anicipaion effecs of fiscal policy does no change he sign of he response. Faás and Mihov (21) and Favero (22) consider a Cholesky ordering in he idenificaion of fiscal shocks. They rely on he effecs of changes in governmen spending, and base heir decision on wo argumens: (i) alernaive heories imply differen economic dynamics following a change in governmen spending while having qualiaively similar predicions for he effecs of changes in ax raes; and (ii) i does no require ha one models he conemporaneous ineracion beween axes and economic aciviy. They sugges ha increases in governmen expendiures are expansionary, bu lead o an increase in privae invesmen ha more han compensaes for he fall in privae consumpion, a feaure ha goes agains he predicions of he Real Business Cycle (RBC) model. Blanchard and Peroi (22) exploi he decision lags in policymaking and use informaion abou he elasiciy of fiscal variables o economic aciviy, o idenify he auomaic response of fiscal policy, and find ha expansionary fiscal shocks increase oupu, and have a posiive effec on privae consumpion and a negaive one on privae invesmen. Mounford and Uhlig (25) use sign resricions on he impulse responses, and idenify an expendiure shock by a posiive response of expendiure for up o four quarers afer he shock. The auhors also find a negaive effec in residenial and nonresidenial invesmen. Despie he differen idenificaion schemes of fiscal policy shocks aimed a analysing he macroeconomic effecs of fiscal policy, less aenion has been given o he poenial role played by fiscal policy on asse markes or he discussion has been cenred on is effecs on long-erm ineres raes. 2 In fac and o he bes of our knowledge, only Afonso and Sousa (28) have ried o ackle his quesion. The auhors esimae a Bayesian Srucural Vecor Auoregression model based on a recursive idenificaion scheme and: (i) look a he impac of fiscal policy on he composiion of oupu; (ii) assess is effecs on asse markes (via housing sock prices, and ineres raes) and on he exernal secor (via exchange rae); and (iii) analyze he poenial ineracions beween fiscal and moneary policy. The scope of he paper is, 2 For a revision of he effecs of fiscal policy on long-erm ineres raes, see, for example, Gale and Orszag (23), Laubach (23), and Brook (23). 7

9 herefore, a more generalis one, as i discusses he macroeconomic effecs of fiscal policy. In his paper, we idenify fiscal policy shocks using a Fully Simulaneous sysem of equaions approach in a Bayesian framework based on he works of Blanchard and Peroi (22), Leeper and Zha (23), and Sims and Zha (1999, 26). Therefore, we ake ino consideraion he auomaic response of fiscal policy o economic aciviy. Moreover, we do no assume ha he governmen reacs only o variables ha are predeermined relaive o policy shocks, and assume ha here are no predeermined variables wih respec o fiscal policy shock. 3. A Fully Simulaneous Sysem approach Consider he following srucural VAR (SVAR) Γ( L) X {{ + id 1 = Γ X + Γ1 X id 1 = c + ε (1) n n n 1 d 1 i G T = + d + (1 + π )(1 + µ ) 1 PY (2) 1 ε v = Γ, (3) where ε, s < ~ Ν(, Λ), Γ(L) is a marix valued polynomial in posiive powers of X s he lag operaor L, n is he number of variables in he sysem, ε are he fundamenal economic shocks ha span he space of innovaions o X. As in Favero and Giavazzi (28), we explicily include he feedback from governmen deb as shown by specificaion (2), where i, G, T, π, Y, P,, µ and d represen, respecively, he ineres rae (or he average cos of deb refinancing), governmen primary expendiures and governmen revenues, inflaion, GDP, price level, real growh rae of GDP, and he deb-o-gdp raio a he beginning of he period We follow Favero and Giavazzi (28), ha is, we add he governmen deb o he VAR and append a non-linear budge ideniy o accumulae deb. This is in conras wih Chung and Leeper (27), who linearize he ineremporal budge consrain and impose i as a se of cross-equaion resricions on he esimaed VAR coefficiens. 4 A feedback from he level of deb raio o governmen revenue and governmen spending could be imporan in he fiscal reacion funcion whenever fiscal auhoriies aach some weigh o deb sabilizaion and heir behaviour is Ricardian. Addiionally, ineres raes depend on fuure expeced moneary policy and on he risk premium, and boh may be affeced by he deb dynamics. Finally, he impac of he level of deb on inflaion (Canzoneri e al., 21) canno be ruled ou ex-ane. Moreover, deb may also have an impac on oupu flucuaions (Barro, 1974; Kormendi, 1983). 8

10 v The vecor v conains he innovaions of X, where ~ Ν(, Σ) and Σ : = Γ 1 1 ' Λ( Γ ). Moreover, Γ pins down he conemporaneous relaions among he variables in he sysem. We use he normalizaion Λ=I. The srucural VAR approach ha we follow is buil on he esimaion of fully simulaneous sysems as in Leeper and Zha (23) and Sims and Zha (26), and on he idenificaion procedure of Blanchard and Peroi (22). We use Bayesian inference o assess he poserior uncerainy abou he impulse-response funcions in he Fully Simulaneous sysem of equaions and consider a Mone Carlo imporance sampling weigh algorihm. Appendix A provides a deailed descripion of he compuaion of he error bands. We consider he following se of variables [ SP G, T, Y, P, i, U, HP ] ' X =,, where SP represens he sock price index, G, he governmen expendiures, T, he governmen revenue, Y, he GDP, P, he GDP deflaor, U, he unemploymen rae, i, he average cos of deb financing (or long-erm ineres rae), and HP, he housing ' ' price index. In paricular, we pariion he daa such ha X = [ X G, T X ] ' X 1 = Y P [ SP ] X = i. ; 2 U HP 1,, 2, where: The economy is divided ino 3 secors: a financial, a public and a producion secor. The financial secor - summarized by sock prices index, SP reacs conemporaneously o all new informaion, in recogniion of he fac ha hey are deermined in markes characerized by a coninuous aucion srucure. The public secor ha allows for simulaneous effecs, comprises he equaions for governmen spending and governmen revenue, and links hem wih he log real GDP, Y, he GDP deflaor, P, and he average cos of financing deb, i. The producion secor consiss of log real GDP, Y, he GDP deflaor, P, unemploymen rae, U, and he housing price index, HP. The orhogonalizaion wihin his secor is irrelevan o idenify fiscal policy shocks correcly. All hese variables are no predeermined relaive o he fiscal policy shocks bu i is assumed ha he policy shock can influence hem conemporaneously. Addiionally, we adop an idenificaion of he fiscal policy shocks based on Blanchard and Peroi (22) and Peroi (24). This idenificaion scheme consiss of 9

11 wo seps: (i) insiuional informaion abou ax and ransfer sysems and he iming of ax collecions is used o idenify he auomaic response of axes and governmen spending o economic aciviy, ha is, o compue he elasiciy of governmen revenue and spending o macroeconomic variables; and (ii) he fiscal policy shock is esimaed. The idenifying resricions on he marix of conemporaneous effecs, Γ, can be defined as: Γ = where he parameers ξ ξ 14 G, Y G, Y ξ ξ 15 G, π G, π ξg, i 22 ξ G, i SP 18 G T Y (4), P i U 88 HP ξ ij can be idenified using exernal informaion. For insance, ξ G, Y, ξ G, π, and ξ G, i are he elasiciies of governmen spending respecively o GDP, GDP deflaor, and long-erm ineres rae. The descripion of he elasiciies used in he idenificaion procedure is repored in Table 1. Table 1 Elasiciies of Governmen Spending and Revenue. Elasiciies of Governmen Spending Elasiciies of Governmen Revenue ξ G,Y ξ G, π ξ G, i ξ T, Y ξ T, π ξ T, i U.S U.K Germany Ialy Noe: The esimaes of he elasiciies for he U.S. are based on Blanchard and Peroi (22), Peroi (27) and Favero and Giavazzi (28). The esimaes of he elasiciies for he U.K. are considered o be he same as in he U.S. The esimaes for Germany and Ialy are based respecively on Heppke-Falk e al. (26) and Giordano e al. (26). 4. Resuls and discussion 4.1 Daa We use quarerly daa for four counries: U.S., U.K., Germany and Ialy. All he variables are in naural logarihms unless saed oherwise. A deailed descripion of he daa is provided in Appendix B. 1

12 For he idenificaion of he fiscal policy shocks, we use he following variables; he producion secor includes he log real GDP, Y, he GDP deflaor, P, and he unemploymen rae, U, he average cos of financing he deb, i, and he housing price index, HP. Whils Leeper and Zha (23) summarize he financial secor by a commodiy prices index, we use a sock price index, SP, insead, as he focus of our analysis is on he reacion of differen asse markes (housing and financial markes) o fiscal policy shocks. Finally, as measure of he fiscal policy insrumens we use eiher he governmen expendiures or he governmen revenues. In he se of exogenous variables, we include a consan (or quarerly seasonal dummies), and he governmen deb-o-gdp raio as described in he previous secion. For Germany, we also include wo dummies: (i) one for 1991:1, corresponding o he German reunificaion; and (ii) anoher one for 2:3, o rack he spike in governmen revenue associaed wih he sale of UMTS (Universal Mobile Telecommunicaions Sysem) licenses. Regarding he quarerly fiscal daa, we consider he Federal Governmen spending and revenue in he case of he U.S.A., and he Public Secor spending and revenue in he case of he U.K. Boh for he U.S.A. and he U.K., quarerly fiscal daa is available direcly from naional accouns. In wha concerns Germany and Ialy, we compue he quarerly series of governmen spending and revenue using he fiscal cash daa, which is monhly published by he fiscal auhoriies of boh counries. In his case, daa for governmen spending and revenue are available in a cash basis, and refer o he Cenral Governmen. The daa are available over he following samples: 197:3-27:4, in he case of he U.S.A.; 1971:2-27:4, in he case of he U.K.; 1979:2-26:4, in he case of Germany; and 1986:2-24:4, in he case of Ialy. 4.2 VAR resuls We sar by esimaing a Bayesian Srucural VAR (B-SVAR) wihou including he deb feedback. Tha is, in pracice, we look a specificaion (1) no considering, as is commonly done in he exising lieraure, he ideniy ha links governmen revenues, governmen spending, governmen deb, GDP, real GDP growh, inflaion and he ineres rae, as defined in (2). We also provide he resuls of he esimaion of he srucural VAR including he feedback from governmen deb as described by specificaions (1), (2), and (3). 11

13 Figures 1, 3, 5, and 7 plo he impulse-response funcions o a fiscal policy shock. The solid line corresponds o he median response when he VAR is esimaed wihou he deb feedback, and he dashed lines are, respecively, he median response and he 68 percen poserior confidence inervals from he VAR esimaed by including he feedback from governmen deb. The confidence bands are consruced by using a Mone-Carlo imporance sampling normalized weighs algorihm, and based on 1 draws. We also plo in Figures 2, 4, 6, and 8 he forecas-error variance decomposiions o a fiscal policy shock, including he deb dynamics. The hinner line corresponds o he median esimae, and he dashed lines indicae he 68 percen poserior confidence inervals esimaed by using a Mone-Carlo imporance sampling normalized weigh algorihm, and based on 1 draws. U.S. Figure 1a displays he impulse-response funcions of all variables in X o a shock in governmen spending in he U.S. When he model is esimaed wihou including he feedback from governmen deb, he resuls show ha governmen spending declines seadily following he shock, and i roughly vanishes afer 12 quarers. Moreover, he increase in governmen spending is followed by a shor fall in governmen revenue ha erodes afer 6 quarers. The effecs on GDP are posiive and relaively large in magniude, peaking a afer 6 quarers. The evidence also suggess ha governmen spending shocks have a negaive and persisen impac on he price level. On he oher hand, here is a negaive effec on long-erm ineres raes, shown as he cos of deb. In wha concerns he reacion of asse markes he empirical evidence suggess ha whils here is a posiive bu almos negligible effec on sock markes, he reacion of housing prices is large and persisen, peaking a afer 8 o 1 quarers. The effecs on unemploymen are negaive and also persisen. When one includes he deb feedback, he effecs of a governmen spending shock on GDP become smaller. On he oher hand, and conrary o he previous findings, here is iniially a posiive impac on he average cos of refinancing he deb, which laer becomes negaive. Looking a he reacion of asse markes, he shock has a small and negaive (alhough) persisen impac on sock prices, whils he effec on housing prices remains posiive. Unemploymen also becomes less volaile. 12

14 Figure 1b shows he impulse-response funcions of all variables o a shock in governmen revenue. When he deb feedback is no aken ino accoun in he, he resuls sugges ha governmen revenue declines seadily following he shock which erodes afer 1 quarers. Addiionally, he shock is iniially followed by a fall in governmen spending which hen recovers and becomes posiive. Conrary o a shock in governmen spending, he effecs on GDP are slighly negaive and very persisen, peaking a afer 1 quarers. The evidence also suggess ha governmen revenue shocks have a posiive and persisen effec on he price level. On he oher hand, here is a posiive and persisen effec on long-erm ineres raes. In wha concerns he reacion of asse markes, he empirical evidence suggess ha he effecs of revenue shocks end o be raher small: despie a very small posiive impac on housing and sock prices ha persiss for around 6 o 8 quarers, he effecs hen mean rever, erodes and become even slighly negaive. The effecs on he unemploymen rae also poin o a persisen increase ha peaks a afer around 12 quarers. When one includes he deb feedback, he resuls sugges ha governmen revenue also increases afer he shock, reflecing he fall in deb-o-gdp raio. The effecs on GDP are iniially negaive, bu mean-rever a afer around 6 quarers and become posiive. Moreover, he evidence suggess ha governmen revenue shocks have a posiive (bu no persisen) effec on he price level, whils he effec on long-erm ineres raes flips sign (vis-à-vis he absence of he budge consrain) and now becomes persisenly negaive in accordance o he deb sabilizing effecs. In wha concerns he reacion of asse markes, he empirical evidence suggess ha he effecs of revenue shocks end o be amplified: sock prices are posiively and persisenly impaced by he shock, whils housing prices move in he opposie direcion. The effecs on he unemploymen rae poin o an increase ha peaks a afer around 4 quarers, herefore, shorer han in he previous case. Figure 2a plos he forecas error-variance decomposiion of all variables o a shock in governmen spending. The empirical findings show ha governmen spending shocks explain only a small percenage of he forecas-error variance decomposiion of he majoriy of he variables included in he VAR. Ineresingly, whils he forecaserror variance decomposiion of sock prices remains roughly consan a around a 2% level over ime (reflecing he quick response of sock markes o he shock), he forecas-error variance decomposiion of housing prices slighly increases up o 5% (in accordance o a slow adjusmen of housing markes o he shock). In addiion, 13

15 governmen spending shocks explain a very imporan share of he forecas-error variance decomposiion of governmen spending: iniially, hey represen more han 9% of he forecas-error and even afer 2 quarers hey correspond o around 4%, herefore, implying a high degree of persisence. The forecas-error variance decomposiions ploed in Figure 2b are also similar o he ones for he governmen spending shock, and show ha governmen revenue shocks play a minor role. U.K. Figure 3a displays he impulse-response funcions of all variables o a shock in governmen spending in he U.K. Conrary o he U.S., he resuls show ha alhough governmen spending declines following he shock, his occurs a a very slow pace so he effec does no vanish even afer 2 quarers. This is also refleced on he governmen revenue which increases persisenly afer he shock. On he oher hand, he effecs on GDP end o be similar o he ones for he U.S.: hey are posiive and persisen. The evidence also suggess ha governmen spending shocks have a negaive effec on he price level. As for he long-erm ineres raes, he effecs are negaive, peaking a afer 1 quarers. In he case of asse markes, housing prices increase wih a lag of around 4 quarers and remain a a persisenly higher level. Regarding sock prices hey record a small fall following he spending shock, bu hey recover afer 8 quarers and reach a persisenly higher level. Conrary o he U.S., unemploymen iniially rises bu he effec mean revers afer 14 quarers and even becomes negaive. Conrary o he case in which we do no consider he deb feedback, he resuls sugges ha, following he shock in governmen spending, governmen revenues increase bu he effec is now less persisen and erodes afer around 8 quarers. Addiionally, whils here is sill a negaive impac on long-erm ineres raes, he effec is subsanially smaller in magniude and less pronounced. This, herefore, explains why GDP iniially falls and mean-revers a afer around 12 quarers, whils he price level iniially goes up and mean-revers a afer 8 quarers. The deb dynamics is also responsible for he response paerns of he asse markes: housing prices are now negaively impaced by he shock, whils he iniial negaive effec on sock prices becomes more pronounced. Finally, he rise in unemploymen is more persisen. Figure 3b shows he impulse-response funcions of all variables o a shock in governmen revenue. Similarly o a shock in governmen spending, he resuls show ha 14

16 governmen revenue declines following he shock, bu a a very slow pace so he effec vanishes only afer 2 quarers. The shock is also followed by a persisen fall in governmen spending. On he oher hand, he effecs on GDP are marginally posiive for around 12 quarers bu hen become negaive, whils he price level is negaively impaced by he shock on governmen revenue. Ineres raes fall afer he shock in governmen revenue bu he effec becomes posiive afer around 1 quarers. Regarding he reacion of asse markes, he empirical evidence suggess ha he effecs of revenue shocks end o be significan and posiive boh for housing and sock prices, alhough more persisen in he firs case: housing prices remain a a persisenly higher level afer 2 quarers wih he peak of he effec being reached a afer around 12 quarers; and sock prices increase for around 12 quarers, bu hen he effec disappears and becomes negaive as a resul of he downurn in GDP. The effec on unemploymen is negaive peaking a afer 8 quarers. A major difference relaive o he previous findings is ha he average cos of financing he deb is roughly unaffeced whils i is negaively affeced when he deb dynamics is no included as a resul of he smaller GDP growh. Figure 4a plos he forecas-error variance decomposiion of all variables in he VAR. Governmen spending shocks explains around 2% of he forecas-error variance decomposiion of governmen spending. In addiion, spending shocks explain around 5% of he forecas-error of sock prices, and only 2% of housing prices. Figure 4b displays he forecas-error variance decomposiions and shows ha governmen revenue shocks explain around 35% of he forecas-error in housing prices and 15% of he forecas-error in sock prices. Germany Figure 5a displays he impulse-response funcions of all variables o a shock in governmen spending in Germany. Similarly o he U.S., he resuls show ha governmen spending declines quickly afer he shock, eroding afer around 12 quarers. The shock is followed by a very shor bu posiive impac on governmen revenue. The effecs on GDP are posiive, peak a afer 4 quarers, and erode afer 12 quarers. On he oher hand, he evidence suggess ha governmen spending shocks have a negaive and persisen effec on he price level, alhough small in magniude. As for he long-erm ineres raes, here is a negaive effec ha persiss even afer 2 quarers. This aspec is also an imporan deerminan of he dynamics ha one observes in he asse markes: 15

17 housing prices go up persisenly; sock prices also rise bu he effec quickly disappears afer 4 quarers. Finally, he resuls sugges ha afer a governmen spending shock, he unemploymen slighly rises. When we include he deb feedback, he effec on GDP is smaller whils he cos of refinancing deb is posiively affeced, suggesing ha deb dynamics is imporan. As a resul, sock prices are negaively impaced (before he effec was posiive) and housing prices reac less posiively o he shock. Figure 5b shows he impulse-response funcions of all variables o a shock in governmen revenue. Similarly o he U.S., he resuls show ha governmen revenue declines quickly afer he shock, eroding afer 2 quarers, and being followed by a reducion in governmen spending ha persiss for around 8 quarers. On he oher hand, conrary o he U.S. and despie a very small and negaive iniial impac, he effecs on GDP are posiive alhough small. Addiionally, boh he price level and he long-erm ineres raes are posiively and persisenly impaced by he shock. Regarding he reacion of asse markes, he empirical evidence suggess ha he effecs of revenue shocks end o be posiive only for housing prices, which reac wih a lag of around 4 quarers. Sock prices iniially rise bu he effec laer mean revers and becomes negaive afer 8 quarers. Finally, he evidence suggess ha governmen revenue shocks have a very pronounced and negaive effec on he unemploymen rae, which peaks a afer around 8 quarers. Including he feedback from governmen deb implies ha he average cos of financing deb is now negaively impaced as a resul of he deb dynamics, ha is, he implici fall in he deb-o-gdp raio. The fall in long-erm ineres raes also affecs he reacion of asse markes: by including he deb feedback, boh housing and sock prices are posiively impaced, whils before ha happened only in he case of housing prices. Figure 6a shows he forecas-error variance decomposiions of all variables o a shock in governmen spending. I can be seen ha governmen spending shocks explain a large share (iniially, close o 8%) of he forecas-error for governmen spending. Moreover, i shows ha shocks o spending also play an imporan role for he forecaserror of he housing prices (around 2%), price level (1%), and jus a small share (less han 5%) of sock prices. Figure 6b shows he forecas-error variance decomposiions of all variables o a shock in governmen revenue. Ineresingly and conrary o governmen spending, 16

18 governmen revenue shocks explain a smaller percenage of he forecas-error variance decomposiion for he majoriy of he variables included in he sysem. Ialy Figure 7a displays he impulse-response funcions of all variables o a shock in governmen spending in Ialy. The resuls show ha governmen spending declines quickly afer he shock, eroding afer 2 o 4 quarers. The effecs on GDP are also similar: GDP (despie a very small posiive iniial effec) falls afer he shock in governmen spending, suggesing a crowding-ou effec. The empirical evidence also suggess ha governmen spending shocks have a posiive and persisen effec on boh he price level and he long-erm ineres rae. In wha concerns he reacion of asse markes, he shock in governmen spending has a posiive impac on housing prices (ha peaks a afer around 6 o 8 quarers) and negaive and very persisen effec on sock prices. The fall in sock prices peaks afer 2 quarers showing ha sock markes reac quickly. Finally, here is no evidence of a significan effec of governmen spending on he unemploymen rae. The resuls are similar in he case where he deb feedback is no included. Figure 7b shows he impulse-response funcions of all variables o a shock in governmen revenue. Similarly o a shock in governmen spending, he resuls show ha governmen revenue declines quickly afer he shock, eroding afer 2 quarers. Addiionally, he effecs on GDP are negaive, alhough no persisen as hey vanish afer 4 quarers. Regarding he reacion of asse markes, he empirical evidence shows ha he effecs of governmen revenue shocks end o be posiive for sock prices and negaive for housing prices. This suggess ha whils he credi channel (ha is, he fall in ineres raes) impacs posiively in sock markes, for housing markes ha channel is annihilaed by he crowding-ou effecs. Finally, he evidence suggess ha unemploymen rae rises afer he shock in governmen revenue, whils here are no significan effecs on he price level. The resuls are again similar o he case where he feedback from governmen deb is no included in he esimaion. Figure 8a shows he error-forecas variance decomposiions of all variables o a governmen spending shock and shows ha i plays a minor role for asse prices. 17

19 Similarly, Figure 8b displays he forecas-error variance decomposiions and shows ha governmen revenue shocks explain a small share of he forecas-error for he majoriy of he non-fiscal variables. 4.3 A VAR couner-facual exercise In his sub-secion, we describe a VAR couner-facual exercise aimed a describing he effecs of shuing down he shocks in governmen spending or governmen revenue. In pracice, afer esimaing he fully simulaneous sysem of equaions summarized by (1), (2) and (3), we consruc he couner-facual (CFT) series as follows: CFT CFT Γ { ( L) X + id 1 = Γ X + Γ1 X id 1 = c + ε 123 n n n 1 CFT CFT (5) d 1 i G T = + d + (1 + π )(1 + µ ) 1 PY (6) CFT 1 CFT ε v = Γ. (7) Since we are ineresed in analysing he role played by fiscal policy shocks, his is equivalen o consider he following vecor of srucural shocks CFT SP T Y P i U HP [ ε,,,,,,, ] ' ε ε ε ε ε ε SP G Y P i U HP [ ε, ε,, ε, ε, ε, ε, ε ] ' ε = (8) CFT ε = (9), where we shu down, respecively in (8) and in (9), he governmen spending shock and he governmen revenue shock, and hen use hese vecors of couner-facual srucural shocks o build he couner-facual series for all endogenous variables of he sysem. Figure 9a and 9b plo he acual and he couner-facual series for sock prices and housing prices in he U.S. and in he case of, respecively, a shock o governmen spending and a shock o governmen revenue. The resuls show ha fiscal policy shocks play a minor role as he difference beween he acual and he counerfacual series are negligible. Similarly, Figures 1a and 1b plo he acual and he couner-facual series for sock prices and housing prices in he U.K. and in he case of, respecively, a shock o governmen spending and a shock o governmen revenue. Conrary o he U.S., he resuls show ha fiscal policy shocks play an imporan role. In fac, i can be seen ha he acual and he couner-facual series are subsanially differen, in paricular: (i) 18

20 during he nineies, in he case of sock prices; and (ii) in he lae eighies and early nineies, for housing prices. Figures 11a and 11b depic he acual and he couner-facual series for sock prices and housing prices in Germany and in he case of, respecively, a shock o governmen spending and a shock o governmen revenue. The resuls sugges ha fiscal policy shocks are less relevan deerminans of asse markes. In fac, whils he difference beween acual and couner-facual series are negligible for sock prices, in he case of housing prices ha difference seems significan only afer 2 and conribued o a more sable performance of he marke. Figure 12a and 12b show he acual and he couner-facual series for sock prices and housing prices in Ialy and in he case of, respecively, a shock o governmen spending and a shock o governmen revenue. The resuls sugges ha fiscal policy shocks, in paricular, hose on he revenue side, are imporan deerminans of asse markes. Moreover, hey show ha unexpeced variance in he fiscal policy sance has a disurbing effec on hose markes, increasing heir volailiy. This is paricularly so afer he second half of he nineies and noably for a governmen revenue shock. 5. Conclusion This paper evaluaes he effecs of fiscal policy on economic aciviy, wih a paricular emphasis on he linkages beween fiscal policy and asse markes. The fiscal policy shocks are idenified using exernal informaion abou he elasiciy of fiscal variables o he economic aciviy as in Blanchard and Peroi (22). Moreover, we use a Fully Simulaneous Sysem approach in a Bayesian framework buil on he works of Leeper and Zha (23), and Sims and Zha (1999, 26), herefore, accouning for he poserior uncerainy of he impulse-response funcions. In addiion, we explicily include he feedback from he governmen deb in our framework. We show ha governmen spending shocks: (i) have a posiive and persisen effec on GDP in he U.S. and he U.K., while for Germany and Ialy, he (posiive) impac is emporary and becomes negaive afer 4 o 8 quarers; (ii) have a posiive and persisen effec on housing prices, alhough housing markes end o respond wih a lag of around 4 quarers; (iii) have a negaive effec on sock prices, alhough he ime of reacion is faser han for housing prices; (iv) have posiive effecs on he price level in he case of he U.K. and Ialy, and negaive effecs for he U.S. and Germany; and (v) reduce unemploymen only in he U.S. On he oher hand, governmen revenue shocks: 19

21 (i) have an iniial negaive effec on GDP ha laer becomes posiive; (ii) have a negaive impac on housing prices for he U.S. and Ialy, and a posiive impac for he U.K. and Germany; (iii) have a small and posiive effec on sock prices; (iv) have, in general, a negaive and persisen effec on he price level; and (iv) have a posiive and persisen impac on he unemploymen rae. Long-erm ineres raes and GDP become more responsive and he effecs on hese variables also become more persisen when we explicily include he deb feedback in he esimaions. Finally, in a VAR couner-facual exercise, we show ha: (i) fiscal policy shocks play a minor role in he paerns ha one observes for sock prices and housing prices in he U.S. and Germany; (ii) boh spending and revenue shocks have an imporan effec on asse markes in he U.K.; and (iii) for Ialy, governmen revenue shocks increased he volailiy in housing and sock prices, in paricular, in he nineies. A possible exension of he curren work is o inroduce a disaggregaed approach, by analyzing he effecs of shocks in he differen componens of governmen revenue (direc axes on households, direc axes on corporaions, indirec axes, and employers social securiy conribuions) and governmen spending (wages, non-wage expendiure). References AFONSO, A.; SOUSA, R. M. (28), The macroeconomic effecs of fiscal policy, ECB Working Paper, forhcoming. BARRO, R. J. (1974), "Are governmen bonds ne wealh?", Journal of Poliical Economy, 82(6), BLANCHARD, O.; PEROTTI, R. (22), "An empirical characerizaion of he dynamic effecs of changes in governmen spending and axes on oupu", Quarerly Journal of Economics, 117(4), BROOK, A.-M. (23), Recen and prospecive rends in real long-erm ineres: fiscal policy and oher drivers, OECD Working Paper #21. CANZONERI, M.; CUMBY, R.; DIBA, B. (21), "Is he price level deermined by he needs of fiscal solvency, American Economic Review, 91(5), CHRISTIANO, L. J.; EICHENBAUM, M.; EVANS, C. L. (25), "Nominal rigidiies and he dynamic effecs of a shock o moneary policy", Journal of Poliical Economy, 113(1),

22 EDELBERG, W.; EICHENBAUM, M.; FISHER, J. (1999), Undersanding he effecs of a shock o governmen purchases, Review of Economics Dynamics, 2, FATÁS, A.; MIHOV, I. (21), The effecs of fiscal policy on consumpion and employmen: heory and evidence, CEPR Discussion Paper # 276. FAVERO, C. (22), How do European moneary and fiscal auhoriies behave? CEPR Discussion Paper #3426. FAVERO, C.; GIAVAZZI, F. (28), "Deb and he effecs of fiscal policy", Universiy of Bocconi, Working Paper. GALE, W. G.; ORSZAG, P. R. (23), Economic effecs of susained budge deficis, Naional Tax Journal, 56, KORMENDI, R. C. (1983), Governmen deb, governmen spending, and privae secor behavior, American Economic Review, 73, LAUBACH, T. (23), New evidence on he ineres rae effecs of budge deficis and deb, Board of Governors of he Federal Reserve Sysem, Finance and Economics Discussion Paper #12. LEEPER, E. M.; ZHA, T. (23), Modes policy inervenions, Journal of Moneary Economics, 5(8), MOUNTFORD, A.; UHLIG, H. (25), "Wha are he effecs of fiscal policy shocks?", Humbold-Universiä zu Berlin Working Paper SFB #649. PEROTTI, R. (24), "Esimaing he effecs of fiscal policy in OECD counries", Universiy of Bocconi, Working Paper. RAMEY, V.; SHAPIRO, M. (1998), Cosly capial reallocaion and he effecs of governmen spending, Carnegie Rocheser Conference on Public Policy, 48, ROTEMBERG, J.; WOODFORD, M. (1992), Oligopolisic pricing and he effecs of aggregae demand on economic aciviy, Journal of Poliical Economy, 1, SIMS, C.; ZHA, T. (26), "Does moneary policy generae recessions?", Macroeconomic Dynamics, 1(2), SIMS, C.; ZHA, T. (1999), "Error bands for impulse-responses", Economerica, 67(5), WAGGONER, D. F.; ZHA, T. (1997), "Normalizaion, probabiliy disribuion, and impulse responses", Federal Reserve Bank of Alana Working Paper #

23 Appendix A. Assessing poserior uncerainy in a Fully Simulaneous SVAR To be able o idenify he srucural fiscal policy shocks we need a leas (n-1)n/2 linearly independen resricions. Wih enough resricions in he Γ marix and no resricions in he marix of coefficiens on he lagged variables, he esimaion of he model is numerically simple since he log-likelihood will be T 1 ' l ( B, a, Γ ) = + log Γ race[ S( B, a) Γ Γ ] (A.1) 2 2 T where S( B, a) = ( B( L) X a)( B( L) X a)'. Inegraing l B, a, Γ ) (or he poserior = 1 ( wih conjugae priors) wih respec o (B, a) he marginal log probabiliy densiy funcion of Γ is proporional o T k 1 ^ ^ ' log(2π ) + ( T k)log Γ race S( B OLS, a OLS ) Γ Γ 2 2. (A.2) The impulse-response funcion o a one sandard-deviaion shock is given by: 1 1 B ( L) Γ. (A.3) This implies ha o assess poserior uncerainy regarding he impulse-response funcion we need join draws for boh B(L) and Γ. Since equaion (A.2) is no in he form of any sandard probabiliy densiy funcion, we canno draw Γ from i direcly o make inference. Neverheless, aking a second order expansion around is peak, we ge he usual Gaussian approximaion o he asympoic disribuion of he elemens in Γ. In addiion, since his is no he rue form of he poserior probabiliy densiy funcion, we canno use i direcly o produce a Mone Carlo sample. Therefore, we follow an imporance sampling approach, in which we draw from he Gaussian approximaion bu weigh he draws by he raio of (A.2) o he probabiliy densiy funcion from which we draw. The weighed sample cumulaive densiy funcion hen approximaes he cumulaive densiy funcion corresponding o (A.2). Noe also ha he disribuion of B(L), given Γ, is he usual normal disribuion: ^ ' 1 vech( B( L)) Γ ~ Ν( vech( B OLS ), Γ ( Γ ) ( X ' X ) ). (A.4) Therefore, we can ake join draws using he following simple algorihm: (i) draw Γ using imporance sampling; and (ii) draw vech(b(l)) using he expression above. Error bands for he impulse-response funcion are hen consruced from he weighed perceniles of he Mone Carlo sample and compued as follows. Denoe ^ 1 1 H he numerical Hessian from he minimizaion rouine a he poin ^ esimae and Γ he maximum-likelihood esimaor, and follow he following algorihm: 1. Check ha all he coefficiens on he main diagonal of Γ are posiive. If hey are no, flip he sign of he rows ha have a negaive coefficien on he main diagonal (ha is, our poin esimaes are normalized o have posiive elemens on he main diagonal). 2. Se i=. 3. Draw vech(γ ) from a normal Ν ( vech( Γ ), V ), where V=H -1 and vech(.) vecorizes he unconsrained elemens of a marix. Tha is, his sep draws from he asympoic disribuion of Γ. There are 3 possible opions o handle draws in which some of he diagonal elemens of Γ are no posiive: 22

24 a. rejec he draw and go back o 2. o ake anoher draw (his is wha is done in Sims and Zha (26) and we follow heir approach); b. rejec he draw if and only if one of he negaive enries on he main diagonal are more han "alpha" sandard deviaions away from he maximum-likelihood esimaor; c. accep he draw and coninue. 4. Compue and sore he imporance sampling weigh, m, m i OLS OLS 2 1 ^ 1 2 ~ ~ ^ ~ ^ = ' exp logv +.5( vech( Γ ) vech( Γ )) V ( vech( Γ ) vech( Γ )) ~ 1 ^ T log de( Γ ) race S( B SCFT ^, a ~ ) Γ ' ~ Γ where SCFT is a scale facor ha prevens overflow/underflow (a good choice for i is normally he value of he likelihood a is peak) Draw ( ~ ' ' 1 vech B( L)) from a normal Ν( vech( B OLS ), Γ ( Γ ) ( X X ) ) o ge a ~ L draw for B ( ). 6. Compue he impulse-response funcion and sore i in a mulidimensional array. 7. If i < #draws, se i = i+1 and go back o 3. The sored draws of he impulse-response funcion, joinly wih he imporance sampling weighs, are used o consruc confidence bands from heir perceniles. Moreover, he draws of Γ are sored o consruc poserior confidence inerval for hese parameers from he poserior (weighed) quaniles. Normalized weighs ha sum up o 1 are simply consruced as: mi ωi =. # draws m i= 1 When he number of draws is sufficienly large for he procedure oulined above o deliver accurae inference, he plo of he normalized weighs should ideally show ha none of hem is oo far from zero - ha is, one single draw should no receive 9% of he weigh. 6 i ^ ~ 1 ~ 1 5 Confidence bands consruced using unweighed quaniles are asympoically jusified (due o he asympoic Gaussianiy), and are good o give a quick look a he shape of he impulse-response funcion using a small number of draws. The unweighed approach should be used wih cauion since: (i) i is likely o produce unrealisically igh bands in he presence of muliple local maxima; and (ii) will no capure asymmeries of he confidence bands (wha are imporan in deecing wheher and impulseresponse funcion is significanly differen from zero). 6 When he imporance sampling performs oo poorly (due o he variabiliy in he weighs), one can replace ha par of he algorihm wih he random walk Meropolis Markov-Chain Mone Carlo of Waggoner and Zha (1997) using also heir approach o handle swich in he sign of he rows of Γ (ha is, use a normalizaion for each draw ha minimizes he disance of Γ from he maximum likelihood esimae). 23

25 Appendix B. Daa sources B.1 U.S. Daa Housing Secor Housing prices are measured using wo sources: (a) he Price Index of New One-Family Houses sold including he Value of Lo provided by he U.S. Census, an index based on houses sold in 1996, available for he period 1963:1-26:3; and (b) he House Price Index compued by he Office of Federal Housing Enerprise Oversigh (OFHEO), available for he period 1975:1-27:4. Daa are quarerly, seasonally adjused. Housing Marke Indicaors Oher Housing Marke Indicaors are provided by he U.S. Census. We use he Median Sales Price of New Homes Sold including land and he New Privaely Owned Housing Unis Sared. We seasonally adjus quarerly daa for he Median Sales Price of New Homes Sold including land using Census X12 ARIMA, and he series comprise he period 1963:1-27:4. The daa for he New Privaely Owned Housing Unis Sared are quarerly (compued by he sum of corresponding monhly values), seasonally adjused and comprise he period 1959:1-27:4. GDP The source is Bureau of Economic Analysis, NIPA Table 1.1.5, line 1. Daa for GDP are quarerly, seasonally adjused, and comprise he period 1947:1-27:4. Price Deflaor All variables were deflaed by he GDP deflaor. Daa are quarerly, seasonally adjused, and comprise he period 1967:1-27:4. The source is he Bureau of Economic Analysis, NIPA Tables and 1.1.6, line 1. Sock Marke Index Sock Marke Index corresponds o S&P 5 Composie Price Index (close price adjused for dividends and splis). Daa are quarerly (compued from monhly series by using end-of-period values), and comprise he period 195:1-27:4. Governmen Spending The source is Bureau of Economic Analysis, NIPA Table 3.2. Governmen Spending is defined as primary governmen expendiure, obained by subracing from oal Federal Governmen Curren Expendiure (line 39) ne ineres paymens a annual raes (obained as he difference beween line 28 and line 13). Daa are quarerly, seasonally adjused, and comprise he period 196:1-27:4. Governmen Revenue The source is Bureau of Economic Analysis, NIPA Table 3.2. Governmen Revenue is defined as governmen receips a annual raes (line 36). Daa are quarerly, seasonally adjused, and comprise he period 1947:1-27:4. Deb Deb corresponds o he Federal governmen deb held by he public. The source is he Federal Reserve Bank of S Louis (series FYGFDPUN ). Daa are quarerly, seasonally adjused, and comprise he period 197:1-27:4. Average Cos of Financing Deb 24