URL:

Transcription

1 Consumpion, Wealh, Sock and Governmen Bond Reurns: Inernaional Evidence Anónio Afonso Ricardo M. Sousa 09/ 2011

2 Consumpion, Wealh, Sock and Governmen Bond Reurns: Inernaional Evidence Anónio Afonso Ricardo M. Sousa NIPE * WP 09/ 2011 URL: hp:// * NIPE Núcleo de Invesigação em Políicas Económicas is suppored by he Poruguese Foundaion for Science and Technology hrough he Programa Operacional Ciência, Teconologia e Inovação (POCI 2010) of he Quadro Comuniário de Apoio III, which is financed by FEDER and Poruguese funds.

3 Consumpion, Wealh, Sock and Governmen Bond Reurns: Inernaional Evidence * Anónio Afonso a, $, #, and Ricardo M. Sousa b a ISEG/TULisbon, Technical Universiy of Lisbon, Deparmen of Economics, UECE - Research Uni on Complexiy and Economics, R. Miguel Lupi 20, Lisbon, Porugal. European Cenral Bank, Direcorae General Economics, Kaisersraße 29, D Frankfur am Main, Germany b Universiy of Minho, Deparmen of Economics and Economic Policies Research Uni (NIPE), Campus of Gualar, Braga, Porugal. London School of Economics, Financial Markes Group (FMG), Houghon Sree, London WC2 2AE, Unied Kingdom. Absrac In his paper, we show, from he consumer s budge consrain, ha he residuals of he rend relaionship among consumpion, aggregae wealh, and labour income should predic boh sock reurns and governmen bond yields. We use daa for several OECD counries and find ha when agens expec fuure sock reurns o be higher, hey will emporarily allow consumpion o rise. Regarding governmen bond yields, when bonds are seen as a componen of asse wealh, hen invesors reac in he same way. If, however, he increase in he yields is perceived as signalling a fuure rise in axes, hen hey will emporarily reduce heir consumpion. JEL classificaion: E21, E44, D12. Keywords: consumpion, wealh, sock reurns, bond reurns. * We are graeful o Jerry Coakley, Ad van Rie, o paricipans o an ECB workshop, o he Money, Macro and Finance Research Group 41 s Annual Conference, and o wo anonymous referees for helpful commens. Ricardo Sousa would like o hank he Fiscal Policies Division of he ECB for is hospialiy. The opinions expressed herein are hose of he auhors and do no necessarily reflec hose of he ECB or he Eurosysem. $ Corresponding auhor. European Cenral Bank, Direcorae General Economics, Kaisersraße 29, D Frankfur am Main, Germany. addresses: anonio.afonso@ecb.europa.eu, aafonso@iseg.ul.p (A. Afonso), rjsousa@eeg.uminho.p, rjsousa@alumni.lse.ac.uk. (R. Sousa). # UECE and NIPE are suppored by FCT (Fundação para a Ciência e a Tecnologia, Porugal).

4 Conens 1. Inroducion Theoreical framework and empirical mehodology Theory Long-run relaionship beween consumpion, wealh and income Empirical resuls Forecasing sock reurns Forecasing governmen bond reurns Forecasing consumpion growh Robusness analysis Addiional conrol variables Nesed forecas comparisons Conclusion...23 References

5 1. Inroducion Differences in expeced reurns across asses are explained by differences in risk, and he risk premium is generally considered as reflecing he abiliy of an asse o insure agains consumpion flucuaions (Sharpe, 1964). Despie his belief, a measure such as he covariance of reurns across porfolios and conemporaneous consumpion growh did no prove o be sufficien o explain he differences in expeced reurns (Breeden e al., 1989). In fac, he lieraure on asse pricing has concluded ha inefficiencies of financial markes 1 and he raional response of agens o ime-varying invesmen opporuniies 2 help jusifying why expeced excess reurns appear o vary wih he business cycle. In addiion, differen macro-financially moivaed variables ha capure imevariaion in expeced reurns have been developed. For insance: he consumpion-wealh raio (Leau and Ludvigson, 2001); he long-run risk (Bansal and Yaron, 2004; Bansal e al., 2005); he labour income risk (Julliard, 2004); he housing collaeral risk (Lusig and van Nieuwerburgh, 2005); 3 he ulimae consumpion risk (Parker and Julliard, 2005); and he composiion risk (Yogo, 2006; Piazzesi e al., 2007). Addiional variables include he (adjused) dividend or cash flow yield (Goyal and Welsh, 2003; Roberson and Wrigh, 2006; Boudoukh e al., 2007); he raio of excess consumpion (i.e. consumpion in excess of labour income) o observable asses (Whelan, 2008); and he wealh composiion risk (Sousa, 2010a). 4 In conras wih he lieraure on he predicabiliy of sock reurns, only a few sudies ried o explain he facors behind sovereign bond risk premia. Among hese: Fama and Bliss (1987) focus on he spread beween he n-year forward rae and he one-year yield; Campbell and Shiller (1991) emphasize he Treasury yield spreads; Campbell and Cochrane (1999) and Wacher (2006) sress he role of shocks o aggregae consumpion, while Brand and Wang (2003) address he imporance of shocks o inflaion; Cochrane and Piazzesi (2005) highligh a single facor, a single en-shaped linear combinaion of forward raes; and Ludvigson and Ng (2009) find marked counercyclical variaion in bond risk premia. The curren paper argues ha he wealh and macroeconomic daa can be combined o address he issue of predicabiliy of asse reurns. More specifically, we derive an 1 See Fama (1998), Fama and French (1996), and Farmer and Lo (1999). 2 See Sundaresan (1989), Consaninides (1990), Campbell and Cochrane (1999), Duffee (2005), and Sanos and Veronesi (2006). 3 In he same spiri, Sousa (2007a) shows ha housing can be used as an hedge agains wealh shocks. 4 Sousa (2007b) provides a comparison of he forecasing power of several empirical proxies, bu disinguishes beween heir expeced and unexpeced componens. 3

6 equilibrium relaion beween he ransiory deviaion from he common rend in consumpion, aggregae wealh and labour income, labelled as cay (cday when disaggregae wealh is used) and sock reurns as well as governmen bond yields. The above-menioned common rends summarize agen's long-erm expecaions of sock reurns, governmen bond yields and/or consumpion growh. In paricular, when forward-looking invesors expec fuure sock reurns o be higher, hey will allow consumpion o rise above is common rend wih aggregae wealh and labour income. As in Leau and Ludvigson (2001) and Sousa (2010a), in his way, invesors insulae fuure consumpion from flucuaions in sock reurns. As for governmen bond reurns, one needs o undersand he way governmen deb is perceived by agens. If governmen bonds are seen as a componen of asse wealh, hen invesors allow consumpion o rise above is equilibrium relaionship wih aggregae wealh and labour income when hey have expecaions of higher governmen bond yields. However, if he issuance of governmen deb is seen as a sympom of deerioraing public finances, hen invesors will allow consumpion o fall below is common rend wih aggregae wealh and labour income. Using daa for a se of 16 OECD counries, we show ha he predicive power of he cay and cday measures for real sock reurns is paricularly imporan for horizons from hree o four quarers for Ausralia, Belgium, Canada, Denmark, Finland, he UK and he US. As for Germany, Ireland, and Spain, hose proxies do no seem o rack well imevariaion in sock reurns. In wha concerns governmen bond reurns, our analysis suggess ha we can cluser he counries in wo groups. In he firs group (which includes Ausralia, Finland and he Neherlands), boh cay and cday have an associaed coefficien wih posiive sign in he forecasing regressions. Therefore, his corroboraes he idea ha governmen deb is a componen of asse wealh. In he second group (which includes Canada, Denmark, Germany, Ireland, Ialy and he US), he forecasing regressions show ha boh cay and cday have an associaed negaive coefficien. Consequenly, agens in hese counries undersand he rise in governmen bond reurns raher as signalling an increase in fuure axaion. Finally, assessing he robusness of our resuls, we show ha: (i) addiional conrol variables do no change he predicive power of cay and cday; and (ii) models ha include cay and cday perform beer han oher benchmark models. The paper is organized as follows. Secion wo describes he heoreical framework and presens he empirical mehodology. Secion hree provides he esimaion resuls of 4

7 he forecasing regressions for asse reurns. Secion four provides he robusness analysis. Secion five concludes. 2. Theoreical framework and empirical mehodology 2.1 Theory Consider he case of a represenaive consumer. The budge consrain can be wrien as W 1 ( 1 Rw, 1)( W C ), (1) where W represens aggregae wealh, C denoes privae consumpion, and R w,+1 corresponds o he reurn on aggregae wealh beween period and +1. Campbell and Mankiw (1989) show ha, under he assumpion ha he i consumpion-aggregae wealh is saionary and ha lim ( c w ) 0, 5 one can approximae equaion (1) by a Taylor expansion as follows i i w w, i w i w i1 i1 i c w r c k, (2) where c logc, w logw, and k w is a consan. This equaion shows ha deviaions of consumpion from is equilibrium relaionship wih aggregae wealh can reflec changes in he reurns on aggregae wealh or in consumpion growh. The aggregae reurn on wealh can be decomposed as R R (1- ) R, (3) w, 1 a, 1 h, 1 where is a ime varying coefficien and R a,+1 is he reurn on asse wealh. Campbell (1996) shows ha he las expression can be approximaed as r r (1- ) r k, (4) w, a, h, r where k r is a consan, and r w, is he log reurn on asse wealh. Similarly, he log oal wealh can be approximaed as w a (1- )h k, (5) a where a is he log asse wealh, h is he log human wealh, ω is he mean of, and k a is a consan. Campbell (1996) and Jagannahan and Wang (1996) sugges ha labour income, Y, can be hough as he dividend on human capial, H. Consequenly, he reurn o human capial can be defined as: w i i 5 The auhors assume ha ( W C) / W 1. w 5

8 H 1 Y 1 1 Rh, 1. (6) H Log-linearizing his relaion around he seady sae, 6 one ges h, 1 h h h r (1- ) k ( h - y ) - ( h - y ) y, (7) where r log(1+r), h logh, y logy, k h is a consan of no ineres, and he variables wihou ime subscrip are evaluaed a heir seady sae value. Assuming ha lim i i h ( h i y i ) 0, he log human capial income raio can be rewrien as a linear combinaion of fuure labour income growh and fuure reurns on human capial: i1 - h ( i - h, i) h. i1 h y y r k (8) Replacing equaion (4), (7) and (8) ino (2), one obains c a (1 )( y i1 i1 h y i ) i i i1 i r (1 ) ( ) r c, (9) i1 w a, i w h h,i w i k i1 i1 where k is a consan. This equaion holds ex-pos as a direc consequence of agen's budge consrain, bu i also has o hold ex-ane. Taking ime condiional expecaion of boh sides gives c a - (1-) y E cay i1 i1 i i1 i wra, i (1 ) E h y i E wc i k, i i1 where (1 ) ( ) r, is a saionary componen. w h h,i When he lef hand side of equaion (10) is high, consumers expec high fuure reurns on marke wealh. Based on ha equaion, cay may carry relevan informaion abou marke expecaions of fuure asse reurns, r a,+i, fuure labour income growh, y i, and fuure consumpion growh, c i i1 i1 (10). In paricular, when fuure sock reurns are expeced o be higher, forward-looking invesors will allow consumpion o rise above is common rend wih aggregae wealh and labour income. In wha concerns governmen bond reurns: (i) invesors behave in he same fashion when governmen bonds are seen as 6 This is rue under he assumpion ha he seady sae human capial-labour income raio is consan, ha is, Y H, where 0 < ρ h < 1. 1 / h 1 6

9 a componen of asse wealh; bu (ii) invesors will reduce consumpion when higher governmen bond reurns are perceived as signalling a deerioraion of public finances. Finally, Sousa (2010a) highlighs he imporance of wealh composiion in pricing risk premium. 7 By disaggregaing wealh, a, ino is major componens (financial wealh, f, housing wealh, u ) and aggregae reurns, r a,, ino reurns on financial asses, r f,, and reurns on housing asses, r u,, one can link he rend deviaion cday wih he marke expecaions of fuure financial and housing asse reurns as follows: E f i1 cday c f f uu - (1- f u ) y i i i1 i rf, i E ru, i (1- ) E y E c k, w u u i1 i1 i i1 where (1 ) ( ) r. f w w h 8 h,i f u i1 h i i1 w i (11) 2.2. Long-run relaionship beween consumpion, wealh and income In order o assess he long-run relaionship beween consumpion, (dis)aggregae wealh and labour income, we sar by using he augmened Dickey-Fuller and he Phillips-Perron ess. This allows one o deermine he exisence of uni roos in he series and he ess sugges ha all series are firs-order inegraed, I(1). Nex, we analyze he exisence of coinegraion among he series, using he mehodology of Engle-Granger and Phillips-Ouliaris, and find evidence ha suppors ha hypohesis. Finally, we esimae he rend relaionship among consumpion, wealh and labour income following Davidson and Hendry (1981) and Blinder and Deaon (1985). Since he impac of differen asses caegories on consumpion can vary (Poerba and Samwick, 1995; Sousa, 2008), we also disaggregae wealh ino is main componens: financial wealh and housing wealh. Following Sock and Wason (1993), we use a dynamic leas squares (DOLS) echnique, specifying he following equaion 7 Sousa (2010b) also shows ha moneary policy can indeed have a srong impac on he composiion of wealh in he euro area as a whole. 8 From a heoreical poin of view, some auhors argue ha housing wealh effecs should be small. For insance, Buier (2008) susains ha an increase in value of housing leads o higher housing consumpion coss, which offse he housing wealh effec on non-housing consumpion. Muellbauer (2008) suggess ha he posiive effec on non-housing consumpion from an increase in housing prices is counerbalanced by a fall on housing consumpion. Calomiris e al. (2009) emphasize ha changes in housing wealh are ypically correlaed wih changes in expeced permanen income. 7

10 c a a y y k ik a i k b, a-i b y-i (12) ik y, i where he parameers a and y represen, respecively, he long-run elasiciies of consumpion wih asse wealh and labour income, Δ denoes he firs difference operaor, is a consan, and is he error erm. In he esimaion of he long-run relaionships among consumpion, (dis)aggregae wealh and labour income, we use quarerly daa, pos-1960, for 16 counries (Ausralia, Ausria, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Ialy, Japan, he Neherlands, Spain, Sweden, he UK, he US). The consumpion daa are he privae consumpion expendiure and were aken from he daabase of he NiGEM model of he NIESR Insiue, he Main Economic Indicaors of he OECD and DRI Inernaional. The labour income daa correspond o he compensaion series of he NIESR Insiue. In he case of he US, labour income series was consruced following Leau and Ludvigson (2004). The wealh daa were aken from he naional cenral banks or Eurosa. For he G-8 counries, he wealh series were compared wih alernaive sources, namely, Berau (2002), Pichee and Tremblay (2003), Tan and Voss (2003), Cae e al. (2004), and he Bank of Japan. The sock reurn daa were compued using he share price index provided by he Inernaional Financial Saisics of he IMF and he dividend yield raio provided by Daasream. The 10-year governmen bond yield daa were obained from he Inernaional Financial Saisics of he IMF. The governmen finance daa normally refers o he Cenral Governmen, herefore, wih he exclusion of he Local and/or he Regional Auhoriies. I is ypically disseminaed hrough he monhly publicaions of he General Accouning Offices, Minisries of Finance, Naional Cenral Banks and Naional Saisical Insiues of he respecive counries. The laes figures are also published in he Special Daa Disseminaion Sandard (SDDS) secion of he Inernaional Moneary Fund (IMF) websie. Finally, he populaion series were aken from he OECD's Main Economic Indicaors and inerpolaed (from annual daa), and all series were deflaed wih consumpion deflaors and expressed in logs of per capia erms. The series were seasonally adjused using he X-12 mehod where necessary and he ime frames were chosen based on he availabiliy of reliable daa for each counry. 8

11 Table 1.1 shows he esimaes for he shared rend among consumpion, asse wealh, and income. I can be seen ha, despie some heerogeneiy, he long-run elasiciies of consumpion wih respec o aggregae wealh and labour income imply roughly a share of one hird for asse wealh and wo hirds for human wealh in aggregae wealh, in accordance wih he values ha one would expec in a producion funcion wih Cobb-Douglas echnology. This is paricularly rue for Ausralia, Canada, Finland, France, Ireland, he UK and he US. Moreover, he disaggregaion beween asse wealh and labour income is saisically significan for all counries (wih he excepions of Finland and Ialy). The able also presens he uni roo ess o he residuals of he coinegraion relaionship based in he Engle and Granger (1987) mehodology and reveals heir saionariy. Table Long-run relaionship beween consumpion, aggregae wealh, and labour income. cay = c - β 1 a - β 2 y. a Y ADF - Criical values saisic Lags: 1 5% 10% Ausralia 0.35*** (13.39) 0.54*** (8.03) Ausria -0.08*** 1.46*** (-5.10) (23.48) Belgium 0.16*** 0.56*** (8.02) (13.01) Canada 0.36*** 0.56*** (13.16) (10.82) Denmark 0.09*** 0.65*** (6.12) (19.10) Finland 0.38*** (6.88) (0.98) France 0.25*** 0.55*** (16.95) (18.03) Germany 0.13* 1.16*** (1.71) (35.01) Ireland 0.36*** 0.46*** (9.17) (10.03) Ialy *** (-0.20) (11.32) Japan 0.08*** 0.89*** (3.74) (25.99) Neherlands 0.17*** 0.53*** (12.92) (10.30) Spain 0.06* 0.76*** (1.67) (16.10) Sweden -0.13** 1.12*** (-2.45) (9.06) UK 0.32*** 0.66*** (13.84) (12.84) US 0.28*** 0.79*** (17.14) (35.75) Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. *, **, *** denoe saisical significance a he 10, 5, and 1% level, respecively. Table 1.2 repors he esimaes of he long-run elasiciies of consumpion wih respec o financial wealh, f, housing wealh, u, and labour income. Firs, i shows ha he disaggregaion beween financial and housing wealh is saisically significan for almos 9

12 all counries. Moreover, consumpion is, in general, more sensiive o financial wealh han o housing wealh, as he elasiciies of consumpion wih respec o financial wealh are larger in magniude. Second, i ells us ha consumpion is very responsive o financial wealh in he case of Belgium (0.11), Canada (0.30), Finland (0.14), Germany (0.31), Ialy (0.24), Sweden (0.12) and he UK (0.17). Third, he long-run elasiciy of consumpion wih respec o housing wealh is paricularly srong for Ausralia (0.27), France (0.10), Ireland (0.13) and he Neherlands (0.10). Finally, he coinegraion ess sugges ha he residuals of he common rend beween consumpion, financial wealh, housing wealh and labour income are saionary. Table Long-run relaionship beween consumpion, financial wealh, housing wealh, and labour income. cday = c - β 1 f - β 2 u - β 3 y. f U y ADF - Criical values saisic Lags: 1 5% 10% Ausralia 0.07*** (10.26) 0.27*** (9.63) 0.59*** (10.44) Ausria -0.05*** *** (-2.67) (-1.21) (19.10) Belgium 0.11*** -0.06** 0.80*** (8.98) (-2.13) (11.45) Canada 0.30*** 0.06*** 0.49*** (14.43) (2.98) (11.37) Denmark 0.02** *** (2.45) (0.84) (18.88) Finland 0.14*** *** (12.09) (-1.00) (6.53) France 0.08*** 0.10*** 0.62*** (17.22) (4.23) (22.74) Germany 0.31*** 0.09*** 0.33*** (22.10) (3.41) (9.60) Ireland 0.13*** 0.13*** 0.53*** (9.11) (3.06) (9.97) Ialy 0.24*** *** (15.87) (-1.10) (9.57) Japan 0.07*** *** (3.25) (-1.52) (13.95) Neherlands 0.08*** 0.10*** 0.53*** (14.80) (0.02) (7.39) Spain 0.08*** *** (5.60) (0.93) (13.80) Sweden 0.12*** -0.15*** 0.61*** (10.28) (-8.11) (9.43) UK 0.17*** 0.06*** 0.76*** (18.87) (2.99) (16.14) US 0.04*** *** (2.66) (-0.46) (22.53) Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. *, **, *** denoe saisical significance a he 10, 5, and 1% level, respecively. 3. Empirical resuls 3.1. Forecasing sock reurns Equaion (10) shows ha ransiory deviaions from he long-run relaionship among consumpion, aggregae wealh and income, cay, mainly reflec agens expecaions of fuure changes in asse reurns. 10

13 Moreover, given he disaggregaion of asse wealh ino is main componens (financial and housing wealh), we argue ha cday should provide a beer forecas for asse reurns han a variable like cay in Leau and Ludvigson (2001). We look a real sock reurns (denoed by SR ) for which quarerly daa are available and should provide a good proxy for he non-human componen of asse wealh. Table 2.1 summarizes he forecasing power of cay a differen horizons. I repors esimaes from OLS regressions of he H-period real sock reurn, SR SR +H, on he lag of cay. Ausralia 0.64* (1.73) Ausria 0.51 (1.42) Belgium 1.56*** (4.37) [0.11] Canada 0.72*** (3.31) Denmark 0.38*** (2.84) Finland 0.74** (2.09) France 1.51*** (2.76) Germany (-0.99) Table 2.1 Forecasing real sock reurns: esimaed effec of cay. SR +1 + SR SR +H = f(cay -1 ), H=1, 2, 3, 4, 8. Forecas Horizon H Forecas Horizon H ** 1.44** 1.88*** 3.03*** Ireland (2.02) (2.35) (2.61) (3.71) (0.56) (0.22) (-0.19) (-0.33) [0.10] 1.01 (1.59) 2.72*** (4.31) [0.11] 1.27*** (3.55) [0.11] 0.82*** (3.58) [0.11] 1.67*** (3.06) 3.24*** (3.94) [0.13] (-1.20) 1.45* (1.66) 3.39*** (3.63) [0.12] 1.51*** (3.08) 1.31*** (4.09) [0.14] 2.67*** (3.83) [0.11] 4.75*** (4.85) [0.18] (-1.33) 1.99* (1.87) 3.31*** (2.94) 1.50*** (2.52) [0.17] 1.77*** (4.43) [0.15] 3.64*** (4.16) [0.14] 6.05*** (5.34) [0.20] (-1.55) 2.92* (1.82) (-0.07) 0.68 (0.95) 3.57*** (5.24) [0.25] 6.30*** (4.87) [0.18] 10.51*** (7.00) [0.29] -1.85** (-2.36) Ialy 0.23 (0.97) Japan 0.86** (2.32) Neherlands 0.65* (1.80) Spain 0.27 (0.25) Sweden 0.05 (0.19) UK 0.94*** (3.81) [0.09] US 0.71** (2.24) 0.43 (1.23) 1.40** (2.31) 1.15** (2.23) (-0.02) 0.09 (0.19) 1.88*** (4.19) [0.15] 1.55*** (2.89) 0.67 (1.59) 1.58** (1.95) 1.91*** (2.86) 0.35 (0.28) (-0.07) 2.66*** (4.41) [0.15] 2.21*** (3.13) 0.91* (1.82) 1.76* (1.84) 2.62*** (3.37) 0.47 (0.29) (-0.59) 3.27*** (4.54) [0.18] 2.77*** (3.24) Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. Adjused R-square is repored in square brackes. *, **, *** denoe saisical significance a he 10, 5, and 1% level, respecively (-0.57) 1.69** (2.48) (-0.33) 3.47*** (2.78) (-1.10) -1.81* (-1.68) 3.42*** (4.27) [0.12] 5.69*** (4.86) [0.15] I shows ha cay is saisically significan for almos all counries and he poin esimae of he coefficien is large in magniude. Moreover, is sign is posiive. These resuls are in line wih he heoreical framework presened in Secion 3, suggesing ha invesors will emporarily allow consumpion o rise above is equilibrium level in order o smooh i and insulae i from an increase in real sock reurns. Therefore, deviaions in he long-erm rend among c, a and y should be posiively relaed o fuure sock reurns. Moreover, hey explain an imporan fracion of he variaion in fuure real reurns (as described by he adjused R-square), in paricular, a horizons spanning from hree o four quarers. In fac, a he four quarer horizon, cay explains 20% (France), 18% (UK), 17% (Canada), 15% (Denmark), 14% (Finland), 8% (Belgium and US) and 7% (Ausralia) 11

14 of he real sock reurn. In conras, is forecasing power is poor for counries such as Germany, Ireland, Spain and Sweden. Table 2.2 summarizes he forecasing power of cday a differen horizons. I repors esimaes from OLS regressions of he H-period real sock reurn, SR SR +H, on he lag of cday. Ausralia 0.66* (1.67) Ausria 0.39 (1.14) Belgium 2.25*** (4.50) [0.18] Canada 1.21*** (4.71) Denmark 0.36*** (2.61) Finland 0.92* (1.94) France 1.83*** (2.87) Germany -1.52*** (-3.10) Table 2.2 Forecasing real sock reurns: esimaed effec of cday. SR +1 + SR SR +H = f(cday -1 ), H=1, 2, 3, 4, 8. Forecas Horizon H Forecas Horizon H * 1.53** 2.18** 3.72*** Ireland (1.92) (2.19) (2.47) (3.55) (0.63) (0.35) (-0.06) (-0.20) (0.11) [0.10] 0.75 (1.24) 4.18*** (5.97) [0.24] 2.34*** (5.48) [0.19] 0.76*** (3.27) [0.10] 1.68** (2.06) 3.99*** (4.13) [0.14] -2.16*** (-2.65) 1.04 (1.26) 5.74*** (7.15) [0.28] 3.08*** (4.95) [0.19] 1.21*** (3.74) [0.12] 2.27** (2.18) 5.91*** (5.20) [0.19] -2.15** (-1.99) 1.44 (1.43) 6.67*** (7.28) [0.26] 3.40*** (4.22) [0.17] 1.62*** (4.07) [0.13] 2.74** (2.11) 7.70*** (6.04) [0.22] (-1.51) 2.21 (1.39) 6.96*** (4.27) [0.12] 3.44*** (3.14) [0.10] 3.18*** (4.80) [0.21] 4.01* (1.86) 13.51*** (8.12) [0.32] 1.19 (0.68) Ialy 0.65 (1.09) Japan 0.80** (2.24) Neherlands 0.69* (1.90) Spain (-0.30) Sweden 0.87 (1.58) UK 0.95*** (2.49) US 0.50 (0.88) 1.27 (1.46) 1.33** (2.21) 1.29** (2.48) 0.04 (0.02) 2.20*** (2.77) 2.08*** (4.25) [0.12] 1.14 (1.38) 1.76* (1.78) 1.58** (1.96) 2.13*** (3.17) 1.76 (0.81) 3.45*** (3.60) [0.11] 3.13*** (5.30) [0.17] 1.93* (1.79) 2.21* (1.95) 1.77* (1.89) 2.90*** (3.69) 3.84 (1.45) 3.98*** (3.68) [0.11] 4.13*** (5.65) [0.22] 2.41* (1.93) Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. Adjused R-square is repored in square brackes. *, **, *** denoe saisical significance a he 10, 5, and 1% level, respecively. 5.50*** (3.06) [0.09] (-0.18) 4.09*** (3.05) 4.91 (1.46) 4.65*** (3.73) 5.68*** (4.66) [0.22] 5.42*** (3.68) [0.09] In accordance wih he findings for cay, i shows ha cday is saisically significan for almos all counries, he poin esimae of he coefficien is large in magniude and is sign is posiive. Therefore, deviaions in he long-erm rend among c, f, h and y should be posiively linked wih fuure sock reurns. In addiion, i can be seen ha he rend deviaions explain a subsanial fracion of he variaion in fuure real reurns. A he four quarer horizon, cday explains 26% (Belgium), 22% (France and UK), 17% (Canada), 13% (Denmark), 7% (Ausralia), 6% (Neherlands), 4% (Finland and US) of he real sock reurn. However, i does no seem o exhibi forecasing power for counries such as Germany, Ireland, and Spain. Noiceably, i is imporan o emphasize ha, in general, cday performs beer han cay, also in accordance wih he findings of Sousa (2010a), reflecing he abiliy of cday o rack he changes in he composiion of asse wealh. Porfolios wih differen composiions of asses are subjec o differen degrees of liquidiy, axaion, or ransacion 12

15 coss. For example, agens who hold porfolios where he exposure o housing wealh is larger face an addiional risk associaed wih he (il)liquidiy of hese asses and he ransacion coss involved in rading hem. Wealh composiion is, herefore, an imporan source of risk ha cday bu no cay is able o capure Forecasing governmen bond reurns We now look a he power of cay (Table 3.1) and cday (Table 3.2) in predicing bond reurns (proxied by he governmen bond yields and denoed by BR ) for which quarerly daa are available. As menioned before, one needs o keep in mind ha, in conras wih socks, an increase in he governmen bond reurn may no be seen as a rise in wealh, bu may be perceived as a signal of a fuure increase in axes. Therefore: (i) when agens see governmen deb as a componen of wealh, one should observe a posiive poin coefficien for cay and/or cday in he forecasing regressions; and (ii) when agens inerpre he rise in governmen deb as a signal of deerioraion of public finances, hen cay and cday should be negaively relaed o fuure governmen bond reurns. Table 3.1 shows ha cay is saisically significan for almos all counries and he poin esimae of he coefficien is large in magniude. I can also be seen ha he rend deviaions srongly predic fuure real governmen bond yields, in paricular, a horizons spanning from hree o four quarers. Indeed, a he four quarer horizon, cay explains 64% (Ialy), 31% (Sweden), 33% (Ausralia), 27% (Canada), 23% (Germany), 13% (Belgium), 11% (Denmark), 10% (Ireland) and 8% (Japan) of he real governmen bond reurns. As for France, Spain and he UK, he forecasing power of cay is virually nil. 13

16 Ausralia 0.54*** (7.81) [0.30] Ausria (-1.08) Belgium -0.47*** (-3.06) Canada -0.42*** (-5.15) [0.21] Denmark -0.40*** (-3.72) Finland 0.39** (2.36) France 0.22* (1.91) Germany -0.17*** (-4.39) [0.11] Table 3.1 Forecasing real bond reurns: esimaed effec of cay. BR +1 + BR BR +H = f(cay -1 ), H=1, 2, 3, 4, 8. Forecas Horizon H Forecas Horizon H *** 1.62*** 2.15*** 4.26*** Ireland -0.48*** -1.02*** -1.61*** -2.17*** (8.61) (9.23) (9.29) (9.20) (-2.96) (-3.24) (-3.43) (-3.52) [0.32] [0.33] [0.33] [0.33] [0.10] [0.10] (-1.39) -1.00*** (-3.58) [0.11] -0.86*** (-5.54) [0.23] -0.81*** (-4.06) [0.09] 0.75** (2.44) 0.40* (1.73) -0.33*** (-4.52) [0.15] -0.27* (-1.70) -1.60*** (-4.01) [0.12] -1.33*** (-5.97) [0.26] -1.25*** (-4.05) [0.10] 1.09** (2.43) 0.54 (1.55) -0.51*** (-4.98) [0.18] -0.36* (-1.86) -2.17*** (-4.26) [0.13] -1.80*** (-6.32) [0.27] -1.79*** (-4.23) [0.11] 1.36** (2.32) 0.65 (1.39) -0.73*** (-5.93) [0.23] -0.84** (-2.25) -4.11*** (-4.03) [0.12] -3.48*** (-6.76) [0.28] -4.11*** (-4.61) [0.15] 2.13* (1.86) 0.79 (0.85) -1.59*** (-7.51) [0.32] Ialy -0.44*** (-11.10) [0.57] Japan (-0.27) Neherlands 0.33*** (4.79) [0.10] Spain 0.10 (0.46) Sweden -0.30*** (-2.66) UK (-0.97) US -0.27** (-2.13) -0.91*** (-12.34) [0.60] (-0.60) 0.57*** (4.44) [0.09] 0.17 (0.40) -0.56*** (-3.79) [0.13] (-1.34) -0.54** (-2.15) -1.38*** (-13.02) [0.62] (-1.19) 0.80*** (4.47) 0.23 (0.36) -0.94*** (-5.25) [0.21] (-1.54) -0.82** (-2.23) -1.86*** (-13.61) [0.64] -0.98*** (-3.52) 0.99*** (4.10) 0.24 (0.28) -1.31*** (-7.31) [0.31] (-1.58) -1.14** (-2.34) Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. Adjused R-square is repored in square brackes. *, **, *** denoe saisical significance a he 10, 5, and 1% level, respecively *** (-3.06) -3.86*** (-15.52) [0.70] -1.72*** (-3.70) 1.31*** (2.71) 0.58 (0.31) -2.70*** (-8.28) [0.34] -1.01** (-2.01) -2.16** (-2.24) Ineresingly he resuls sugges ha he sign of he coefficien of cay is posiive for Ausralia, Finland, and he Neherlands and negaive for Belgium, Canada, Denmark, Germany, Ireland, Ialy and he US. This piece of evidence corroboraes he idea ha agens allow consumpion o rise above is equilibrium relaionship wih asse wealh and labour income when hey expec governmen bond yields o increase in he fuure. As for he second se of counries, agens perceive he rise in governmen bond reurns as a deerioraion of he public finances and an increase in fuure axaion. In pracice, hese resuls largely reflec higher susainabiliy of public finances in he firs se of counries. 9 As for he second se of counries, hey characerize well he relaively frequen swings in public deficis and governmen deb and he concerns abou he long-erm susainabiliy of public finances. Table 3.2 describes he resuls from forecasing regressions of cday a differen horizons. 9 Ineresingly, Afonso (2005) finds ha i is no possible o rejec he hypohesis of coinegraion beween governmen spending and revenue for Ausria, Finland, and he Neherlands. 14

17 Ausralia 0.67*** (7.60) [0.32] Ausria (-1.37) Belgium 0.45*** (2.72) Canada -0.52*** (-5.45) [0.21] Denmark -0.36*** (-3.43) Finland 0.80*** (4.63) [0.12] France 0.54*** (3.62) [0.10] Germany (-0.37) Table 3.2 Forecasing real bond reurns: esimaed effec of cday. BR +1 + BR BR +H = f(cday -1 ), H=1, 2, 3, 4, 8. Forecas Horizon H Forecas Horizon H *** 2.07*** 2.76*** 5.56*** Ireland (8.46) (9.20) (9.28) (9.48) (-0.60) (-0.76) (-0.88) (-0.89) [0.35] [0.36] [0.37] [0.39] -0.21* (-1.75) 0.90*** (2.74) -1.08*** (-5.94) [0.24] -0.71*** (-3.65) 1.58*** (5.40) [0.14] 1.03*** (3.43) [0.09] (-0.00) -0.32** (-2.06) 1.25** (2.44) -1.66*** (-6.42) [0.26] -1.08*** (-3.58) 2.32*** (5.67) [0.14] 1.46*** (3.23) (-0.06) -0.43** (-2.21) 1.62** (2.39) -2.32*** (-6.79) [0.27] -1.52*** (-3.71) [0.09] 2.92*** (5.53) [0.18] 1.84*** (3.07) (-0.23) -0.95*** (-2.64) 3.35*** (2.55) -3.91*** (-5.87) [0.22] -3.37*** (-3.99) [0.11] 4.56*** (4.27) [0.09] 3.02*** (2.54) -.17 (-0.25) Ialy (-0.05) Japan (-0.01) Neherlands 0.48*** (6.84) [0.21] Spain 0.11 (0.30) Sweden 0.07 (0.30) UK 0.02 (0.17) US 0.01 (0.05) (-0.18) (-0.07) 0.88*** (6.27) [0.21] 0.04 (0.05) 0.30 (1.20) (-0.02) 0.01 (0.04) (-0.30) (-0.50) 1.27*** (6.17) [0.20] (-0.27) 0.29 (0.88) (-0.07) (-0.01) (-0.43) -0.63** (-2.46) 1.63*** (6.09) [0.19] (-0.58) 0.34 (0.98) (-0.01) (-0.09) Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. Adjused R-square is repored in square brackes. *, **, *** denoe saisical significance a he 10, 5, and 1% level, respecively (-0.45) (-1.01) -1.23*** (-3.13) 2.68*** (5.14) [0.13] (-0.86) 1.03 (1.55) 0.68 (0.88) (-1.03) In accordance wih he findings for cay, Table 3.2 shows ha cday is saisically significan for a reasonable se of counries. A he four quarer horizon, cday explains 37% (Ausralia), 27% (Canada), 19% (Neherlands), 18% (Finland), 9% (Denmark), 8% (France), 6% (Belgium), 5% (Ausria) and 4% (Japan) of he real governmen bond reurns. As for Germany, Ireland, Ialy, Spain, Sweden, he UK and he US, he forecasing power of cday is negligible. The resuls also sugges ha he sign of he coefficien of cday is posiive for Ausralia, Belgium, Finland, France and he Neherlands, herefore, supporing he idea ha agens behave in a non-ricardian manner. As for Ausria, Canada, Denmark and Japan, he sign of he coefficien of cday is negaive and, herefore, indicaes ha agens behave in a Ricardian fashion. In fac, a rise in governmen bond yields is perceived as a signal of deerioraion of public finances. 10 Under he Ricardian Equivalence hypohesis, forward-looking consumers save he proceeds from a deb-financed fiscal simulus in anicipaion of he fuure ax increase needed o repay he addiional governmen deb. Such Ricardian behaviour implies ha consumers ne wealh is invarian o a deb-financed governmen expendiure increase. 10 Afonso (2008) also repors empirical evidence regarding he exisence of Ricardian fiscal regimes in he European Union. By is urn, Sousa (2010c) shows ha invesors in he euro area as a whole seem o consider governmen deb as a componen of asse wealh. 15

18 However, some raher resricive assumpions need o be in place, noably: infiniely living households; lump-sum axes; efficien capial markes; and absence of credi consrains. 11 Neverheless, if fiscal expansions are perceived as permanen, and lead o expecaions of much higher governmen deb, he imporance of Ricardian behaviour may acually increase. In addiion, he possible negaive reacions of financial markes o sizeable increases in governmen deb may undermine he expeced posiive economic effec from a fiscal expansion. Indeed, an increased risk of governmen deb defaul and he poenial rise in ineres raes will dampen or even offse he relaed economic simulus. Finally, we assess he exisence of a poenial bias in he coefficiens associaed wih cay and cday in he forecasing regressions of sock reurns and governmen bond yields. Sambaugh (1986, 1999) sugges ha when he empirical proxies cay and cday used as regressors in he forecasing equaions are auocorrelaed and he shocks o regressors are correlaed wih shocks o reurns, he dependen variable is no independen of all leads and lags of he error erms. This, in urn, generaes an upward bias in he esimaions, which is, approximaely, equal o ( 1 3) / T under he normaliy assumpion, where is he coefficien obained from regressing he residual in he reurns regression on he residual from an AR(1) regression for he forecasing variables (cay or cday), is he AR coefficien for he forecasing variables, and T is he sample size. The magniudes of he bias are shown in Table 3.3 and sugges ha i does no seem o affec he predicive power of boh cay and cday. In fac, he bias is small in he forecasing regressions a differen horizons. Consequenly, even afer making his adjusmen, he empirical proxies cay and cday are saisically significan and imporan predicors of sock reurns and governmen bond yields. This is also in accordance wih he works Leau and Ludvigson (2001) and Sousa (2010a). Similarly, Whelan (2008) finds ha he bias does no impac on he predicive abiliy of he excess consumpion o observable asses. 11 See noably Barro (1976) and Seaer (1993). 16

19 Table 3.3 Forecasing asse reurns: checking for poenial bias Sock reurns, Forecas Horizon H Sock reurns, Forecas Horizon H cay cday Ausralia Ausralia Ausria Ausria Belgium Belgium Canada Canada Denmark Denmark Finland Finland France France Germany Germany Ireland Ireland Ialy Ialy Japan Japan Neherlands Neherlands Spain Spain Sweden Sweden UK UK US US Gov. bond Forecas Horizon H Gov. bond Forecas Horizon H reurns, cay reurns, cday Ausralia Ausralia Ausria Ausria Belgium Belgium Canada Canada Denmark Denmark Finland Finland France France Germany Germany Ireland Ireland Ialy Ialy Japan Japan Neherlands Neherlands Spain Spain Sweden Sweden UK UK US US Noes: he magniude of he bias is, approximaely, equel o )/T, under he normaliy assumpion; is he coefficien from regressing he residual in he reurns regression on he residual from an AR(1) regression for he forecasing variables; is he AR coefficien for he forecasing variables; T is he sample size. (Sambaugh, 1986, 1999) Forecasing consumpion growh In principle, cay and cday could be a proxy for expeced fuure consumpion growh, asse reurns, or boh. Table 4.1 presen he resuls of he regressions of consumpion growh over horizons spanning 1 o 8 quarers, on he lag of rend deviaion cday. Table 4.2 provides a summary of he findings when he lag of cday is used as he explanaory variable. In he esimaion of he regressions of consumpion growh, he dependen variable is, herefore, he H-period consumpion growh rae Δc Δc +H. Consisen wih he findings of Leau and Ludvigson (2001), he resuls shown in Table 4.1 sugges ha, in general, cay has no predicive power for fuure consumpion growh. The individual coefficiens are no saisically significan and he adjused R-square is close o zero. A few excepions include he cases of Canada, France, Neherlands, Spain, Sweden and he UK, where he rend deviaion cay exhibis some predicive power. Neverheless, one should noe ha he coefficiens are sill very 17

20 small in magniude. As a resul, he residuals from he conegraing relaionship among consumpion, asse wealh and labour income can be generally described as a good predicor of asse reurns (and no of fuure consumpion growh). Ausralia -0.05* (1.71) Ausria (-1.00) Belgium (-0.26) Canada 0.13*** (5.03) [0.18] Denmark 0.03 (0.65) Finland (-0.40) France (-1.41) Germany 0.01 (0.44) Table 4.1 Forecasing consumpion growh: esimaed effec of cay. c +1 + c c +H = f(cay -1 ), H=1, 2, 3, 4, 8. Forecas Horizon H Forecas Horizon H * -0.09* Ireland (-1.71) (-1.69) (-1.57) (-0.31) (-0.40) (-0.67) (-0.67) (-0.58) (0.02) (-1.08) (-0.94) 0.27*** (7.45) [0.33] 0.07 (1.08) 0.00 (0.01) -0.14*** (-2.53) 0.06 (1.44) (-1.39) -0.15** (-2.16) 0.39*** (8.56) [0.41] 0.07 (0.91) 0.04 (0.59) -0.20*** (-2.93) 0.10** (1.95) (-0.81) -0.18* (-1.94) 0.51*** (8.69) [0.46] 0.10 (1.22) 0.09 (1.18) -0.24*** (-2.92) 0.13** (2.23) (-0.42) (-0.46) 0.85*** (9.35) [0.44] 0.03 (0.21) 0.46*** (3.44) [0.10] -0.34*** (-2.72) 0.31*** (3.33) [0.10] Ialy (-0.28) Japan 0.04 (1.10) Neherlands -0.13*** (-2.72) Spain 0.12*** (2.64) [0.09] Sweden 0.04*** (2.59) UK 0.10*** (3.72) US 0.02 (0.44) (-0.15) 0.09 (1.57) -0.25*** (-3.60) [0.12] 0.26*** (3.37) [0.16] 0.07*** (2.97) 0.20*** (4.91) [0.12] 0.03 (0.58) 0.00 (0.12) 0.10 (1.22) -0.34*** (-4.09) [0.14] 0.41*** (4.04) [0.22] 0.11*** (3.17) 0.28*** (5.73) [0.15] 0.02 (0.08) 0.01 (0.32) 0.15 (1.36) -0.41*** (-4.05) [0.14] 0.60*** (5.06) [0.30] 0.14*** (3.36) 0.39*** (6.80) [0.20] 0.02 (0.16) Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. Adjused R-square is repored in square brackes. *, **, *** denoe saisical significance a he 10, 5, and 1% level, respecively (0.65) 0.15 (0.86) -0.67*** (-4.04) [0.12] 1.20*** (7.71) [0.49] 0.31*** (4.84) [0.13] 0.73*** (8.04) [0.29] (-0.11) In conras, Table 4.2 shows ha cday conains some relevan informaion abou fuure consumpion growh. The coefficiens associaed wih he deviaions of consumpion from is rend relaionship wih financial wealh, housing wealh, and labour income are saisically significan for Canada, France, Germany, Ireland, Ialy, Neherlands and Sweden. In he case of Canada, his is also refleced in a relaively large adjused R-square. These findings may be relaed wih he fac ha cday racks changes in he composiion of asse wealh and, in paricular, in housing wealh. Given ha housing wealh changes end o have small bu also very persisen effecs on consumpion, 12 cday may, herefore, be capuring ime-variaion in expeced reurns and in consumpion growh. Finally, one should emphasize ha despie his, he coefficiens associaed wih cday in he regressions of consumpion growh are quie 12 See Sousa (2010d) for a review of he opic and, in paricular, an applicaion of he wealh effecs on consumpion o he euro area as a whole. Similarly, Sousa (2009) finds evidence of imporan wealh effecs from moneary policy. 18

21 small, which is in conras wih he findings in he regressions of real sock reurns and real governmen bond yields. Consequenly, cday mainly forecass asse reurns. Ausralia -0.07** (-2.09) Ausria (-0.98) Belgium (-0.92) Canada 0.15*** (4.75) [0.14] Denmark 0.02 (0.43) Finland (-1.36) France (-1.50) Germany -0.18** (-1.96) Table 4.2 Forecasing consumpion growh: esimaed effec of cday. c +1 + c c +H = f(cday -1 ), H=1, 2, 3, 4, 8. Forecas Horizon H Forecas Horizon H ** -0.12** -0.12* Ireland -0.18** -0.40*** -0.61*** -0.78*** (-2.09) (-1.96) (-1.92) (-1.33) (-2.12) (-2.58) (-2.81) (-3.02) [0.09] [0.12] [0.13] [0.13] (-1.04) -0.10* (-1.82) 0.30*** (7.04) [0.26] 0.04 (0.67) (-1.19) -0.18*** (-2.76) -0.26*** (-2.79) (-1.29) -0.19*** (-2.71) 0.44*** (8.07) [0.34] 0.02 (0.24) (-0.94) -0.27*** (-3.29) -0.34*** (-3.19) (-0.73) -0.25*** (-2.74) 0.56*** (7.31) [0.36] 0.03 (0.35) (-0.87) -0.33*** (-3.36) -0.50*** (-5.04) [0.11] (-0.21) (1.62) 0.83*** (6.42) [0.26] (-1.33) 0.04 (0.26) -0.61*** (-3.99) [0.12] -0.68*** (-3.56) Ialy -0.07*** (-3.49) Japan 0.05 (1.39) Neherlands -0.13*** (-2.90) Spain -0.23*** (-2.68) [0.12] Sweden -0.11*** (-2.74) [0.10] UK 0.05 (1.42) US (-0.15) -0.14*** (-3.92) 0.12** (1.94) -0.26*** (-3.78) [0.14] -0.34*** (-2.83) [0.10] -0.21*** (-3.38) [0.14] 0.07 (1.27) (-0.29) -0.17*** (-3.31) 0.15* (1.81) -0.36*** (-4.24) [0.16] -0.40** (-2.30) -0.31*** (-3.66) [0.15] 0.07 (0.93) (-0.63) -0.17*** (-2.55) 0.22** (2.06) -0.44*** (-4.21) [0.16] (-1.35) -0.38*** (-3.79) [0.15] 0.14 (1.56) (-0.67) Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. Adjused R-square is repored in square brackes. *, **, *** denoe saisical significance a he 10, 5, and 1% level, respecively. 4. Robusness analysis 4.1. Addiional conrol variables Campbell and Shiller (1988), Fama and French (1988) and Lamon (1998) show ha he raios of price o dividends or earnings or he raio of dividends o earnings have predicive power for sock reurns. More recenly, Goyal and Welsh (2003) argue ha because he dividend yield follows a random walk i canno predic sock prices. However, Roberson and Wrigh (2006) and Boudoukh e al. (2007) menion ha a change in ax legislaion in he US in 1983 ha legalised share buybacks implies an adjusmen of he dividend yield for hese and similar effecs. Consequenly, his adjused saisic is mean revering and a good predicor of sock reurns. Table 5.1 repors he esimaes from forecasing regressions ha include addiional variables shown o conain predicive power for real sock reurns, in paricular, he dividend yield raio (DivYld ). We also add he lag of he real sock reurns (SR -1 ) as a conrol variable. The resuls show ha boh he poin coefficien esimaes of cay and cday and heir saisical significance do no change wih respec o he findings of Tables 2.1 and -1.24*** (-3.78) [0.12] (-0.31) 0.28 (1.57) -0.76*** (-4.48) [0.16] 0.10 (0.24) -0.47*** (-2.80) 0.28** (2.17) (-1.41) 19

22 2.2 where only cay and cday were included as explanaory variables. Moreover, he lag of he dependen variable is no saisically significan, a feaure ha is in accordance wih he forward-looking behaviour of sock reurns. Finally, he dividend yield raio (DivYld ) seems o provide relevan informaion abou fuure asse reurns since i is saisically significan in pracically all regressions and i improves he adjused R- square. Table 5.1 Forecasing real sock reurns: addiional conrol variables. SR +1 = f(cay -1,...); SR +1 = f(cday -1,...). SR -1 cay -1 DivYld -1 Adj. R-square SR -1 cday -1 DivYld -1 Adj. R-square Ausralia ** 7.36** [0.11] ** 7.48*** (0.79) (2.40) (2.22) (-0.44) (2.02) (2.86) Ausria (1.18) (0.95) (1.30) (0.60) Belgium *** 3.85** [0.20] *** [0.23] (-0.26) (6.08) (1.89) (0.05) (4.75) (-1.32) Canada *** 7.89*** [0.13] *** 5.54** [0.17] (0.68) (4.42) (2.91) (1.00) (5.36) (2.10) Denmark 0.43*** 0.23** [0.23] 0.43*** 0.20* [0.22] (2.86) (2.00) (2.90) (1.71) Finland ** ** 4.29 (1.23) (2.42) (1.41) (1.15) (0.60) (1.20) France *** 3.64** [0.10] *** 2.34 (-0.69) (3.60) (2.04) (-0.31) (3.22) (1.29) Germany ** 6.97* (0.19) (1.32) (1.33) (-0.01) (-2.49) (1.89) Ireland (-0.72) (0.39) (-0.70) (0.39) Ialy *** 29.41*** [0.16] * 23.33*** [0.10] (-0.32) (3.27) (4.09) (0.56) (1.89) (2.98) Japan ** (-0.62) (1.99) (1.61) (-0.39) (1.54) (1.30) Neherlands [0.10] 0.32* [0.12] (1.05) (-1.33) (0.21) (1.80) (0.97) (-0.57) Spain (0.88) (0.20) (0.77) (0.05) Sweden *** 29.68*** [0.19] *** [0.13] (-1.61) (2.66) (3.96) (-0.04) (1.07) (3.26) UK -0.21* 0.60*** 4.73** [0.13] *** 4.31** [0.12] (-1.80) (3.20) (2.44) (-1.51) (2.65) (2.11) US 0.05 (0.57) 0.76** (2.42) 2.32 (1.32) 0.05 (0.66) 0.47 (0.81) 0.91 (0.53) Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. Adjused R-square is repored in square brackes. *, **, *** denoe saisical significance a he 10, 5, and 1% level, respecively. On he oher hand, Table 5.2 repors he esimaes from forecasing regressions ha include addiional variables shown o conain predicive power for long-erm ineres raes, in paricular, he inflaion rae (Inflaion) and he defici-o-gdp raio (Defici). Gale and Orszag (2003) argue ha here are wo imporan reasons why governmen budge deficis may raise nominal ineres raes: (i) budge deficis reduce aggregae savings when privae savings do no increase by he same amoun and here 20

23 are no compensaing foreign capial inflows; and (ii) budge deficis increase he sock of governmen deb and, consequenly, he ousanding amoun of governmen bonds. In his case, here is a porfolio effec, as a higher ineres rae on governmen bonds would be required in order for invesors o hold he addiional bonds. Table 5.2 Forecasing real bond reurns: addiional conrol variables. BR +1 = f(cay -1,.. ); BR +1 = f(cday -1,.. ). cay -1 Inflaion -1 Defici -1 Adj. R-square cday -1 Inflaion -1 Defici -1 Adj. R-square Ausralia 0.60*** 0.00*** 0.01 [0.35] 0.71*** 0.00** [0.35] (9.45) (3.20) (0.20) (8.66) (2.23) (-0.13) Ausria *** [0.11] *** [0.12] (-1.03) (2.90) (-1.32) (2.91) Belgium -0.41*** *** [0.52] *** [0.46] (-3.83) (-0.25) (7.35) (1.08) (-0.42) (6.61) Canada -0.17** 0.00* [0.10] -0.48*** ** [0.33] (-2.04) (1.91) (-1.27) (-5.92) (1.09) (-2.13) Denmark -0.27** 0.01*** [0.17] -0.26** 0.01*** [0.17] (-2.27) (3.81) (-2.40) (4.12) Finland 0.35** ** [0.14] (2.01) (-1.50) (-1.22) [2.36] (-1.29) (-1.06) France 0.61*** 0.01*** 0.01 [0.38] 0.87*** 0.01*** 0.00 [0.46] (5.95) (6.71) (0.54) (7.59) (7.09) (0.03) Germany -0.16*** 0.00*** 0.07 [0.16] ** (-4.44) (3.79) (0.63) (-0.96) (2.29) (-1.31) Ireland -0.48*** (-2.96) (-0.60) Ialy -0.32*** 0.01*** 0.06 [0.75] *** -0.23*** [0.65] (-8.47) (6.33) (1.45) (-0.59) (7.37) (-4.40) Japan *** -2.78** [0.36] *** -3.47** [0.32] (0.05) (4.72) (-2.35) (0.57) (4.73) (-2.17) Neherlands 0.18** *** [0.24] 0.36*** *** [0.30] (2.24) (1.31) (-3.76) (4.01) (1.28) (-2.73) Spain 0.35** 0.02*** -0.43** [0.30] *** -0.39** [0.28] (2.03) (2.74) (-2.32) (0.71) (2.61) (-2.00) Sweden -0.38*** * [0.10] (-3.15) (0.07) (-1.81) (0.29) (0.37) (-0.89) UK -0.18*** 0.00*** 0.01 [0.18] -0.18** 0.00*** [0.16] (-3.63) (2.51) (0.23) (-2.25) (3.01) (-1.37) US 0.03 (0.21) 0.01*** (3.79) (-0.03) [0.11] 0.16 (1.17) 0.02*** (5.77) -0.66*** (-3.23) [0.34] Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. Adjused R-square is repored in square brackes. *, **, *** denoe saisical significance a he 10, 5, and 1% level, respecively. While some sudies find ha ineres raes end o increase afer a rise in he defici, ohers do no (Barro and Sala-i-Marin, 1990; Engen and Hubbard, 2004). The empirical findings seem o depend on wheher expeced or curren budge deficis are used as explanaory variables (Upper and Worms, 2003; Laubach, 2009), and also on wheher yield differenials in Europe wih respec o Germany (Codogno e al., 2003) or ineres rae swap spreads are used as he dependen variable (Goodhar and Lemmen, 1999; Afonso and Srauch, 2007). For Europe, he exising evidence poins eiher o a significan (alhough small) effec (Bernoh e al., 2003; Codogno e al., 2003; Afonso and Srauch, 2007; Faini, 21