Wealth, Stock Returns, Government Bond Yields and Systemic Risk

Transcription

1 Wealh, Sock Reurns, Governmen Bond Yields and Sysemic Risk Ricardo M. Sousa $ London School of Economics, FMG, Universiy of Minho, and FMG Sepember 2010 Absrac In his paper, I assess he role of collaeralizable wealh and sysemic risk in explaining fuure asse reurns. I show ha he residuals of he rend relaionship among housing wealh and labour income predic boh sock reurns and governmen bond yields. Using daa for a se of indusrialized counries, I find ha when he housing wealh-o-income raio falls, invesors demand a higher risk premium for socks. As for governmen bonds reurns: (i) when hey are seen as a componen of asse wealh, invesors reac in he same manner; and (ii) if, however, invesors perceive he increase in governmen bond reurns as signalling a fuure rise in axes or a deerioraion of public finances, hen invesors inerpre he fall in he housing wealh-o-income raio as a fall in fuure bond premia. Finally, I show ha he occurrence of crises episodes amplifies he ransmission of housing marke shocks o financial markes. Keywords: housing wealh, labor income, sock reurns, governmen bond reurns, crises. JEL classificaion: E21, E44, D12. $ London School of Economics, Financial Markes Group (FMG), Houghon Sree, London WC2 2AE, Unied Kingdom; Universiy of Minho, Deparmen of Economics and Economic Policies Research Uni (NIPE), Campus of Gualar, Braga, Porugal. s: rjsousa@alumni.lse.ac.uk; rjsousa@eeg.uminho.p.

2 1. Inroducion The heoreical and empirical lieraure has shown ha credi markes are no perfec, and are characerized by he lack of arbirage and raioning (Sigliz and Weiss, 1981). Besanko and Thakor (1987) argue ha hese problems could be avoided if borrowers had enough collaeralizable wealh. In fac, banks would be able o offer wo differen conracs o prospecive cusomers: (i) one requiring a high collaeral (and a low ineres rae), herefore, aracing low-risk individuals; and (ii) anoher one requiring less collaeral (and a high ineres rae), hus favoring high-risk enrepreneurs. In addiion, he efficiency of he housing finance sysem is of key ineres o financial insiuions, homeowners, and policy makers. Liquidiy and collaeralizable wealh play, herefore, a major role for asse pricing. Firs, liquidiy shocks are posiively correlaed wih shocks o reurns (Jones 2002). Second, asses have higher expeced reurns when hey are posiively correlaed wih aggregae marke liquidiy (Pasor and Sambaugh, 2003; Acharya and Pedersen, 2005). Third, asses wih high ransacion coss or illiquid asses normally rade a a discoun (Brennan and Subrahmanyam, 1996). While differences in expeced reurns are ypically explained by differences in risk, he covariance of porfolio reurns and conemporaneous consumpion growh does no fully explain he cross-secional variaion (Mankiw and Shapiro, 1986; Breeden e al., 1989). As a resul, he idenificaion of he economic sources of risk remains an imporan issue. Moreover, given he srong linkages beween housing marke developmens and sock marke dynamics, many auhors sared o consider feaures of hose markes in asse pricing models (Lusig and Van Nieuwerburgh, 2005; Yogo, 2006; Fernandez-Corugedo e al., 2007; Piazzesi e al., 2007; Sousa, 2007). The curren paper addresses he role of collaeralizable wealh in analysing predicabiliy of boh sock and governmen bond reurns for a se of indusrialized counries. Specifically, I assess he forecasing power of he nonlinear deviaions of housing wealh from is coinegraing relaionship wih labor income, hwy, for expeced fuure reurns. The raionale behind his linkage lies on he fac ha a decrease in housing prices reduces he value of housing in providing collaeral services and, herefore, increases household s exposure o idiosyncraic risk. Consequenly, a decrease in he raio of asse wealh o human wealh predics higher sock reurns. As for governmen bond reurns, one needs o undersand he way governmen deb is perceived by he 2

3 agens. If governmen bonds are seen as a componen of asse wealh, hen invesors demand a higher bond risk premium when hey face a fall in he raio of collaeralizable wealh. If, however, he issuance of governmen deb is undersood as leadig o an increase in fuure axaion or as a sympom of public finance deerioraion, hen invesors will inerpre he fall in he housing wealh-o-income raio as predicing a decrease in fuure governmen bond reurns. I show ha he raio of housing wealh-o-income, hwy, predics boh sock and governmen bond reurns, which highlighs he characerisic of housing as providing collaeral o he banking sysem. I also emphasizes he imporan channel by which shocks originaed in he housing secor are ransmied o risk premium in asse markes. The empirical findings sugges ha he predicive power of hwy for real sock reurns is subsanial, ranging beween 6% (US), 8% (Finland and UK), and 10% (Ausralia) over he nex 4 quarers. Wih regards o governmen bond reurns, he analysis shows ha one can cluser he se of counries ino wo groups. In he firs group (which includes Ausralia, Denmark, Finland, Neherlands and Spain), hwy has an associaed coefficien wih negaive sign in he forecasing regressions. The predicive power is, paricularly, large for Neherlands (11%), Finland (13%) and Spain (49%). This, herefore, corroboraes he idea ha governmen deb is seen as par of he invesor s asse wealh, which implies ha agens exhibi a non-ricardian behaviour. In he second group (which includes Belgium, Canada, France, Germany, Ireland, Ialy, Japan, Sweden, he UK and he US), he forecasing regressions show ha hwy has a posiive coefficien. Specifically, he predicive abiliy of hwy is large for Germany (11%), Ireland (12%), Belgium (28%) and he US (29%). Consequenly, agens in hese counries perceive he rise in governmen bond reurns raher as a deerioraion of public finances and as signalling an increase in fuure axaion, ha is, hey behave in a Ricardian way. Finally, I ask abou he imporance of episodes of crisis in amplifying he ransmission of shocks in he housing marke o he financial sysem. In paricular, I assess wheher he occurrence of sysemic versus non-sysemic crises can help improving our undersanding abou he linkages beween housing and financial markes. I show ha he predicive power of fuure asse reurns is indeed improved when one akes ino accoun he presence of such phenomena, especially, he sysemic ones. The robusness of he resuls is analysed in several direcions. In fac, I show ha: (i) he inclusion of addiional conrol variables does no change he predicive 3

4 power of hwy; and (ii) models ha include hwy perform beer han he auoregressive and he consan expeced reurns benchmark models. The research presened in his paper is indebed o he work of Leau and Ludvigson (2001). However, he auhors use he consumer s ineremporal budge consrain o explore he predicive abiliy of he deviaions of consumpion from is long-run relaionship wih aggregae wealh and labour income, cay, for sock reurns. In conras, I use he srucure of he preferences of he represenaive agen o assess he forecasing power of he deviaions of housing wealh from is equilibrium relaionship wih labour income, hwy, for boh sock reurns and governmen bond yields. This work is organized as follows. Secion 2 reviews he lieraure on he predicabiliy of asse reurns. Secion 3 describes he heoreical approach. Secion 4 discusses he empirical resuls from he forecasing regressions for sock reurns and governmen bond yields. Secion 5 provides he robusness analysis. Secion 6 analyses he role of sysemic. Finally, in Secion 7, I conclude and discuss he implicaions of he findings. 2. Lieraure review In his secion, I review he lieraure on he predicabiliy of sock reurns and governmen bond reurns, in paricular, by highlighing he works ha focus on he ransmission of housing marke developmens o he financial sysem Predicabiliy of sock reurns Risk premium is generally considered as reflecing he abiliy of an asse o insure agains consumpion flucuaions. The empirical evidence has, however, shown ha he covariance of reurns across porfolios and conemporaneous consumpion growh is no sufficien o jusify he differences in expeced reurns. In fac, he lieraure on asse pricing has emphasized he role of marke inefficiencies (Fama, 1998; Fama and French, 1996), he raional response of agens o ime-varying invesmen opporuniies ha is driven by variaion in risk aversion (Consaninides, 1990) and by changes in he join disribuion of consumpion and asse reurns (Duffee, 2005), and differen models of economic behaviour. These explanaions also jusify why expeced excess reurns on asses appear o vary wih he business cycle. Therefore, differen economically moivaed variables have been developed o capure ime-variaion in expeced reurns and documen long-erm predicabiliy. Leau 4

5 and Ludvigson (2001) show ha he ransiory deviaion from he common rend in consumpion, aggregae wealh and labour income is a srong predicor of sock reurns, as long as he expeced reurns o human capial and consumpion growh are no oo volaile. Bansal and Yaron (2004) find ha he long-run risk, ha is, he exposure of asses' cash flows o consumpion is an imporan deerminan of risk premium. Julliard (2004) emphasize he role of labor income risk, while Parker and Julliard (2005) measure he risk of a porfolio by is ulimae risk o consumpion, ha is, he covariance of is reurn and consumpion growh over he quarer of he reurn and many following quarers. Lusig and Van Nieuwerburgh (2005) show ha he housing collaeral raio can shif he condiional disribuion of asse prices and consumpion growh. Yogo (2006) and Piazzesi e al. (2007) emphasize he role of non-separabiliy of preferences in explaining he counercyclical variaion in equiy premium, while Fernandez-Corugedo e al. (2007) focus on he relaive price of durable goods. Sousa (2007) shows ha housing can be used as a hedge agains wealh shocks. Chien and Lusig (2010) find ha accouning for he imporance of collaeralizable wealh, namely, by allowing agens o file for bankrupcy, allows one o improve asse pricing predicions Predicabiliy of bond reurns In conras wih he lieraure on he predicabiliy of sock reurns, here are jus a few sudies ha ry o explain he facors undermining bond risk premia. Fama and Bliss (1987) show ha he spread beween he n-year forward rae and he one-year yield can forecas he n-year excess bond reurns. Campbell and Shiller (1991) find ha excess bond reurns can be prediced by he Treasury yield spreads. More recenly, Cochrane and Piazzesi (2005) highligh ha a linear combinaion of forward raes explains up o 44% of he variaion in nex year s excess reurns on bonds wih mauriies ranging from one o five years. While hese findings imply ha bond risk premium is ime-varying, hey are, in general, silen regarding is relaionship wih macroeconomic magniudes. Campbell and Cochrane (1999) sugges ha risk premia on equiy reflecs a slow-moving habi ha is driven by shocks o aggregae consumpion. Despie he linkages beween equiy risk premia and he macroeconomic fundamenals addressed in he above-menioned works, heir imporance for bond risk premia has been ypically negleced. Moreover, he exising empirical evidence ends o 5

6 show ha excess bond reurns can be forecased no by macroeconomic variables such as aggregae consumpion or inflaion, bu raher by pure financial indicaors, such as forward spreads and yield spreads. For insance, Ludvigson and Ng (2009) find marked counercyclical variaion in bond risk premia. 3. Theoreical framework and empirical approach 3.1. Theoreical consideraion: housing wealh and risk premium consumpion, services), I assume ha here is a coninuum of agens who consume nondurable c, and housing services (from which hey derive uiliy or collaeral hw, and are endowed wih sochasic labor income, y i, a ), where i represens he idiosyncraic even and a denoes he aggregae even. The household maximizes uiliy, ha is ( U( c, hw) p( s s ) uc ( s ), hw ( s ), (1) s s0 0 0 where is he ime discoun facor, s represens he sae of he economy, s s ) p( 0 denoes he probabiliy of sae specified by s given he iniial sae s 0, and preferences are ( 1) / ( 1) / (1 ) /( 1) c hw /(1 ), u ( c, hw ) (2) where >0 capures he imporance of housing wealh in he uiliy funcion, ε is he inraemporal elasiciy of subsiuion beween consumpion and services from housing wealh, and is he coefficien of risk aversion. The solvency consrains are resricions on he value of he household s consumpion claim ne of is labour income claim, ha is: s c s ) ( a ) hw ( s ) y ( s ), ( s (3) where s [ d ( s )] represens he price of a claim o d ( s ), and is he renal price of housing services. The srengh of hese consrains is deermined by he raio of asse wealh o human wealh (i.e., he housing wealh-o-income raio), hwy, a a hw / c. hwy ( a ) (4) z where hw a and c a correspond, respecively, o aggregae housing wealh and aggregae consumpion. z 6

7 Equilibrium allocaions and prices will depend on he consumpion weigh as follows: 1) if he household does no swich o a sae wih a binding consrain, i is '(, s ) ; and 2) if i swiches, hen he new weigh is he cuoff level y, a ). ( In order o obain aggregae consumpion, one inegraes over he new household a weighs, ha is, a ) ' (, s ) d ( ; a ), where ; a ) represens he ( ( disribuion over weighs a he sar of period. The consumpion share of an agen can hen be represened as he raio of his consumpion weigh o he aggregae consumpion a a weigh, c, s ) ' (, s ) c ( a ) / ( a ), and, similarly, for he housing wealh ( a a a share of an agen, hw, s ) ' (, s ) hw ( a ) / ( a ), where a ) defines a ( nondecreasing sochasic process. As he raio of housing wealh-o-income, hwy, decreases, he cuoff levels for a he consumpion weighs increase, y, a ) / ( a ), and, if he consumer moves o a ( sae where he consrain is binding, hen he cuoff level for he consumpion share equals he household s labour income share. As a resul, when he raio of housing wealh-o-income, hwy, decreases, he household s exposure o labor income shocks increases and a higher risk premium is demanded. Consequenly, i should predic a rise in fuure sock reurns. In conras wih socks, an increase in governmen bond yields may no be seen as a rise in wealh, bu merely perceived as signalling a fuure increase in axes. Therefore, when agens see governmen deb as a wealh componen, one should observe a behavior similar o he one found for socks; oherwise, deviaions in he long-erm rend among housing wealh and income should be posiively relaed wih fuure governmen bond reurns. ( 3.2. Empirical counerpar: housing wealh-o-income raio Real per capia housing wealh, hw, and labor income, y, are nonsaionary. As a resul, I esimae he following vecor error-correcion model (VECM): log( hw ) log( y ) K log( hw ) k k 1 log( y k ) (5) k log( hw ) log( y ) D, where denoes he ime rend and is a consan. The K error correcion erms allow one o eliminae he effec of regressor s endogeneiy on he disribuion of he leassquares esimaors of 1,,,. 7

8 The componens log (hw) and log (y) are sochasically coinegraed wih he coinegraing vecor 1,,. I also impose he resricion ha he coinegraing vecor eliminaes he deerminisic rends, so ha log( hw ) log( y is saionary. ) Then, he raio of housing wealh-o-income, hwy, is measured as he deviaion from he coinegraion relaionship, i.e.: ^ hwy log( hw ) log( y ). (6) Given ha he OLS esimaors of he coinegraion parameers are superconsisen, one can use he raio of housing wealh-o-income, hwy, as a regressor wihou needing an errors-in-variables sandard error-correcion. ^ ^ 4. Resuls 4.1. Daa The daa are quarerly, pos-1960, and include sixeen counries (Ausralia, since 1970:1; Ausria, since 1978:2; Belgium, since 1980:2; Canada, since 1965:1; Denmark, since 1977:1; Finland, since 1979:1; France, since 1970:2; Germany, since 1965:1; Ireland, since 1975:4; Ialy, since 1971:4; Japan, since 1965:1; he Neherlands, since 1975:1; Spain, since 1978:1; Sweden, since 1977:1; he UK, since 1961:2; and he US, since 1965:1). I, herefore, cover he las 30 o 50 years of daa. All series wih he obvious excepions of sock reurns and governmen bond yields - are deflaed wih consumpion deflaors, expressed in logs of per capia erms and seasonally adjused. Labour income is approximaed by he compensaion series of he NIESR Insiue. In he case of he US, I follow Leau and Ludvigson (2001). As for he UK, I follow Sousa (2010). Wealh includes financial and housing wealh and daa come from Naional Cenral Banks, he Eurosa, he Bank for Inernaional Selemens (BIS), he Unied Naion s Bullein of Housing Saisics for Europe and Norh America. Sock reurns are compued using he share price index provided by he Inernaional Financial Saisics (IFS) of he Inernaional Moneary Fund (IMF) and he dividend yield raio provided by Daasream. The 10-year governmen bond yield daa are also provided by he IFS of he IMF. The governmen finance daa normally refers o he Cenral Governmen, ha is, i excludes 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 8

9 counries. The laes figures are also published in he Special Daa Disseminaion Sandard (SDDS) secion of he Inernaional Moneary Fund (IMF) websie. Daa for populaion are aken from OECD's Main Economic Indicaors and inerpolaed from annual series The long-run relaion I firs use he Augmened Dickey and Fuller (1979), he Phillips and Perron (1988) and he Kwiakowski e al. (1992) ess o deermine he exisence of uni roos in he series of housing wealh and labor income and conclude ha hey are firs-order inegraed, I(1). Nex, I analyze he exisence of coinegraion among he wo series using he mehodology of Engle and Granger (1987), Johansen and Juselius (1990), Phillips and Ouliaris (1990) and MacKinnon (1996), and find evidence ha suppors ha hypohesis. Finally, I esimae he vecor error-correcion model (VECM) as expressed in (5). Table 1 shows he esimaes (ignoring he coefficien esimaes on he consan and he ime rend) for he shared relaionship among housing wealh and income. I can be seen ha, wih he excepions of Canada, France and Spain, he long-run elasiciy of housing wealh wih respec o labour income is posiive, implying ha he wo aggregaes end o share a posiive long-run pah. The able also presens he uni roo ess o he residuals of he coinegraion relaionship, and suppors he idea ha hey are saionary. [ PLACE TABLE 1 HERE ] 4.3. Forecasing sock reurns Secion 3 shows ha ransiory deviaions from he long-run relaionship among housing wealh and income, hwy, mainly reflec agens expecaions of fuure changes in asse reurns. Therefore, I look a real sock reurns (denoed by SR ) for which quarerly daa are available. They should provide a good proxy for he non-human componen of asse wealh. Table 2 summarizes he forecasing power of hwy for differen horizons. I repors esimaes from OLS regressions of he H-period real sock reurn, SR SR +H, on he lag of hwy. I esimae he following model: 9

10 H SR h hwy 1 h1. (7) Noe ha long-horizon reurns are calculaed by summing he (coninuously compounded) quarerly reurns. This implies ha he observaions on long-horizon reurns overlap which possibly biases he differen es saisics owards rejecing he null hypohesis of no predicabiliy more ofen han is correc (Nelson and Kim, 1993; Sambaugh, 1999; Valkanov, 2003; Ang and Bekaer, 2006). Neverheless, one should emphasize ha hese works focus on he predicive abiliy of he dividend yield and he price-earnings raio which are very persisen regressors. In conras, I assess he forecasing power of he deviaions from he equilibrium relaionship beween housing wealh and labor income, hwy, which exhibi much less persisence. Thus, he abovemenioned problems become less severe. Addiionally, Leau and Ludvigson (2001), Whelan (2008) and Sousa (2010) find ha he bias does no impac on he predicive abiliy of a wide range of variables in he forecasing regressions for sock reurns. Finally, he adoped mehodology is sandard in he empirical finance lieraure (Leau and Ludvigson, 2001; Julliard, 2004; Lusig and Van Nieuwerburgh, 2005; Sanos and Veronesi, 2006; Yogo, 2006; Fernandez-Corugedo e al., 2007; Piazzesi e al., 2007; Sousa, 2010). Keeping hese quesions in mind, Table 2 shows ha hwy is saisically significan for a large number of counries and he poin esimaes of he coefficien are large in magniude. Moreover, is sign is negaive and saisically significan for Ausralia, Germany, Finland, Ialy, he UK, and he US. These resuls are in line wih he framework presened in Secion 3, suggesing ha invesors expec a fall in fuure sock reurns when hey observe a rise in he housing wealh-o-income raio. I can also be seen ha he rend deviaions explain an imporan fracion of he variaion in fuure real reurns (as described by he adjused R 2 ), in paricular, a horizons spanning from 4 o 8 quarers. In fac, a he 4-quarer horizon, hwy explains 5% (Japan), 6% (US), 8% (Finland and UK) and 17% (Belgium) of he real sock reurn. In conras, is forecasing power is poor for counries such as Ausria, Canada, France, Ireland, Japan, Neherlands and Spain. [ PLACE TABLE 2 HERE ] 10

11 4.4. Forecasing governmen bond reurns I now look a he power of hwy in predicing governmen bond yields (denoed by BR ) for which quarerly daa are available. Table 3 repors he forecasing abiliy of hwy for differen horizons. I provides esimaes from OLS regressions of he H-period real governmen bond reurn, BR BR +H, on he lag of hwy, as described by he model: H BR h hwy 1 h1. (8) One can see ha hwy is saisically significan for almos all counries (wih he excepion of Ausria) and he associaed coefficien are, in general, large in magniude. A he 4-quarer horizon, hwy explains 11% (Germany and Neherlands), 12% (Ireland), 13% (Finland), 28% (Belgium), 29% (US) and 49% (Spain) of he real bond reurns. Ineresingly he resuls sugges ha he sign of he coefficien associaed o hwy is negaive for Ausralia, Finland, Neherlands and Spain, and posiive for Belgium, Canada, France, Germany, Ireland, Ialy, Japan, Sweden, he UK and he US. This piece of evidence corroboraes he idea ha governmen deb is seen as par of invesor s wealh for he firs se of counries: in he oucome of a fall in he raio of housing wealh-o-income, agens allow consumpion o rise as hey expec fuure yields o increase. As for he second se of counries, agens perceive he rise in governmen bond reurns as a deerioraion of public finances and an increase in fuure axaion. As a resul, hey reduce consumpion when hey observe a fall in he raio of housing wealho-income. In pracice, hese resuls largely reflec higher susainabiliy of public finances in he firs se of counries. Addiionally, hey characerize he frequen swings in public deficis and governmen deb and he concerns abou he long-erm susainabiliy of public finances in he second group of counries. [ PLACE TABLE 3 HERE ] 5. Robusness analysis 5.1. Addiional conrol variables In his Secion, I assess he robusness of he previous resuls, namely, by considering addiional conrol variables. 11

12 Shiller (1984) and Campbell and Shiller (1988) show ha he price-o-dividend raio and he price-o-earnings raio have predicive power for sock reurns. Lamon (1998) suggess ha he raio of dividends o earnings is also a good predicor of sock reurns a quarerly frequency. The relaive T-bill rae, he erm spread and he defaul spread are also shown o have forecasing power (Fama and French, 1989; Hodrick, 1992). Table 4.1 repors he esimaes from one-quarer-ahead forecasing regressions ha include he dividend yield raio (DivYld ) as an addiional variable. I only displays informaion abou counries for which daa on he dividend yield raio is available. The resuls show ha he coefficien esimaes of hwy and heir saisical significance do no change wih respec o he findings of Table 2 where only hwy was included in he se of explanaory variables. Moreover, he dividend yield raio (DivYld ) seems o provide some relevan informaion abou fuure asse reurns: i is saisically significan in a large number of regressions. By is urn, Table 4.2 summarizes he esimaes from one-quarer-ahead forecasing regressions ha include he inflaion rae (Inflaion) and he defici-o-gdp raio (Defici) as poenial deerminans of fuure governmen bond yields. Brand and Wang (2003) argue ha he risk premium is driven by shocks o inflaion and o aggregae consumpion. Gale and Orszag (2003) highligh ha budge deficis may raise nominal ineres raes, because hey reduce aggregae savings and increase he sock of governmen deb. Despie his, he lieraure has no provided a consensual answer ye (Engen and Hubbard, 2005). The resuls show do no show subsanial changes vis-a-vis he findings repored in Table 3, where only hwy was considered in he se of regressors. In addiion, boh inflaion and he defici-o-gdp raio help forecasing bond reurns. Therefore, his suggess ha invesors use governmen bonds o hedge agains he risk of inflaion. I also reveals ha a deerioraion of he fiscal sance is ypically associaed wih a rise in fuure governmen bond yields. [ PLACE TABLE 4.1 HERE ] [ PLACE TABLE 4.2 HERE ] 12

13 5.2. Nesed forecas comparisons Some recen sudies (Bossaers and Hillion, 1999; Goyal and Welch, 2003, 2004) expressed concerns abou he apparen predicabiliy of sock reurns because, while a number of financial variables display significan in-sample forecasing power, hey seem o have negligible ou-of-sample predicive properies. In addiion, he forecasing resuls presened so far could suffer from he "look-ahead" bias ha arises from a long-erm relaionship esimaed using he full sample (Brennan and Xia, 2005). In his conex, some robus saisics such as he Clark and McCracken's (2001) encompassing es (ENC-NEW), he McCracken's (2006) equal forecas accuracy es (MSE-F) and he modified Diebold and Mariano (1995) encompassing es proposed by Harvey e al. (1998) could allow one o explore he ou-of-sample performance of he forecasing model. Noe, however, ha he in-sample and he ou-of-sample ess are equally reliable under he null of no predicabiliy (Inoue and Killian, 2004). Moreover, he resuls from ou-of-sample forecass where he coinegraing vecor is reesimaed every period using only he daa available a he ime of he forecas could srongly undersae he predicive power of he regressor (Leau and Ludvigson, 2001). Therefore, i would make i difficul for hwy o display forecasing power when he heory is rue. Finally, Hjalmarsson (2006) shows ha ou-of-sample forecasing exercises are unlikely o generae evidence of predicabiliy, even when he correc model is esimaed and here is, in fac, predicabiliy. Wih hese caveas in mind and as a final robusness check, I make nesed forecas comparisons, in which I look a he mean-squared forecasing error from a series of one-quarer-ahead ou-of-sample forecass obained from a predicion equaion ha includes hwy as he sole forecasing variable and he mean-squared forecasing error from a variey of forecasing equaions ha do no include hwy. As a resul, he unresriced model ness he benchmark model. I consider wo benchmark models: he auoregressive benchmark and he consan expeced reurns benchmark. In he auoregressive benchmark, I compare he mean-squared forecasing error from a regression ha only includes he lagged asse reurn as he predicive variable wih he mean-squared error from regressions ha include, in addiion, hwy. In he consan expeced reurns benchmark, I compare he mean-squared forecasing error from a regression ha includes a consan (as he only explanaory variable) wih he mean-squared error from regressions ha include, addiionally, hwy. 13

14 Table 5 summarizes he nesed forecas comparisons for he equaions of real sock reurns and governmen bond yields. I shows ha including hwy in he forecasing regressions, in general, improves over he benchmark models. This is paricularly imporan when he benchmark model is he consan expeced reurns benchmark, and, herefore, suppors he exisence of ime-variaion in expeced reurns. [ PLACE TABLE 5 HERE ] 6. Does sysemic risk maer? Financial crises can be conagious and damaging, and promp quick policy responses, as hey ypically lead economies ino recessions and sharp curren accoun imbalances. Among he many causes of financial crises, one can refer: (i) credi booms; (ii) currency and mauriy mismaches; (iii) large capial inflows; and (iv) unsusainable macroeconomic policies. Honohan and Laeven (2005) and Laeven and Valencia (2008) idenify episodes of financial crises, and sysemic crises include currency, deb and banking crises. A sysemic currency crisis corresponds o a nominal depreciaion of he currency of a leas 30% and, simulaneously, a leas a 10% increase in he rae of depreciaion compared o he year before. A sysemic deb crisis describes a siuaion where here are sovereign defauls o privae lending and deb rescheduling programs. In a sysemic banking crisis, here is a large number of defauls on corporae and financial secors, non-performing loans increase sharply, asse prices evenually depress, and real ineres raes increase dramaically Sysemic crises In order o assess he imporance of sysemic crises, I esimae he following models: H SR h 1 1 * SysemicCr hwy hwy isis, (9) h1 H BR h 1 1 * SysemicCr hwy hwy isis, (10) h1 where SysemicCrisis is a dummy variable ha akes he value of 1 in he presence of a sysemic crisis and 0 oherwise, and H refers o he number of quarers-ahead of he forecasing exercise. Given ha he effecs of sysemic crises may no be immediae, I 14

15 consider H=4, herefore, allowing for a ime lag from he dae of he occurrence of he crisis and he emergence of is effecs. Tables 6.1 and 6.2 repor he esimaes from 4-quarers-ahead forecasing regressions as expressed by equaions (9) and (10), respecively. The resuls show ha boh he coefficien esimaes of hwy does no change relaive o he previous findings. Moreover, he coefficien associaed o he ineracion beween hwy and he dummy variable for he sysemic crisis is, in general, saisically significan. In addiion, i has an opposie sign of he one associaed wih hwy, implying ha invesors demand a higher risk premium for boh socks and governmen bonds during sysemic crises. [ PLACE TABLE 6.1 HERE ] [ PLACE TABLE 6.2 HERE ] 6.2. Non-sysemic crises Finally, I analyse he impac of non-sysemic sysemic crises, and regress he following equaions: H SR h 1 1 * NonSysemi hwy hwy ccrisis, (11) h1 H BR h 1 1 * NonSysemi hwy hwy ccrisis, (12) h1 where NonSysemicCrisis is a dummy variable ha akes he value of 1 in he presence of a non-sysemic crisis and 0 oherwise, and H refers o he number of quarers-ahead of he forecasing exercise. Similarly o he case of sysemic crisis, I allow for a lag in he ransmission of he effecs of non-sysemic crises o financial markes and consider H=4. Tables 7.1 and 7.2 summarize he resuls from 4 quarers-ahead forecasing regressions. In general, he coefficien associaed wih he ineracion beween hwy and he dummy variable for he non-sysemic crisis is saisically significan and has, wih he excepion of Finaland, he opposie sign of he one associaed wih hwy. Therefore, in he oucome of a non-sysemic crisis, invesors demand a higher risk premium. [ PLACE TABLE 7.1 HERE ] 15

16 [ PLACE TABLE 7.2 HERE ] 7. Conclusion This paper explores he predicive power of he nonlinear deviaions of housing wealh from is equilibrium relaionship wih labour income (summarized by he variable hwy) for expeced fuure asse reurns. The above-menioned common rend summarizes agen's expecaions of boh sock reurns and governmen bond yields. In paricular, when he housing wealh-oincome raio falls (increases), forward-looking invesors will demand a higher (lower) risk premium given ha hey will be exposed o larger (smaller) idiosyncraic shocks. As for bond yields, if governmen bonds are undersood as anoher wealh componen, hen invesors behave in he same way as for socks. However, if he increase in governmen bond yields is perceived as a sympom of he deerioraion of he fiscal sance, invesors will inerpre he fall in he wealh-o-income raio as a fall in fuure bond risk premium. Using daa for sixeen indusrialized counries, I show ha he predicive power of hwy for real sock reurns is paricularly srong a horizons from 4 o 8 quarers. In wha concerns bond reurns, he analysis suggess ha one can consider wo ses of counries: (i) hose where invesors seem o behave in a non-ricardian way (Ausralia, Denmark, Finland, Neherlands and Spain); and (ii) hose where invesors seem o be forward-looking and o have a Ricardian behavior (Belgium, Canada, France, Germany, Ireland, Ialy, Japan, Sweden, he UK and he US). Finally, I show ha sysemic crises amplify he linkages beween shocks in collaeralizable wealh and financial markes. Therefore, he curren work opens new and challenging avenues for undersanding he dynamics of he relaionship beween he housing secor, sock marke and governmen bond developmens, and he banking sysem. Acknowledgemens I am exremely graeful o Nicholas Apergis, Lloyd Blenman, Michael Brennan and Javier Esrada for very helpful commens. I am also indebed o he paricipans of he Firs Inernaional Symposium in Compuaional Economics and Finance, he Firs World Finance Conference, he Third Annual Brunel Universiy Conference in 16

17 Economics and Finance, and he 6 h fruiful discussions. Poruguese Finance Nework Conference for References Acharya, V. V., and Pedersen, L. H., 2005, Asse pricing wih liquidiy risk, Journal of Financial Economics, Vol. 77, pp Ang, A., and Bekaer, G., 2006, Sock reurn predicabiliy: Is i here?, Review of Financial Sudies, Vol. 20 No. 3, pp Bansal, R., and Yaron, A., 2004, "Risks for he long run: a poenial resoluion of asse pricing puzzles", Journal of Finance, Vol. 59, pp Baxer, M., and Jermann, U. J., 1997, "The inernaional diversificaion puzzle is worse han you hink", American Economic Review, Vol. 87, pp Besanko, D. A., and Thakor, A. V., 1987, Collaeral and raioning: soring equilibria in monopolisic and compeiive credi markes, Inernaional Economic Review, Vol. 28, pp Beser, H., 1985, Screening versus raioning in credi markes wih imperfec informaion, American Economic Review, Vol. 75, pp Bossaers, P., and Hillion, P., Implemening saisical crieria o selec reurn forecasing models: Wha do we learn?, Review of Financial Sudies, Vol. 12 No. 2, pp Brand, M. W., and Wang, K. Q., 2003, Time-varying risk aversion and unexpeced inflaion, Journal of Moneary Economics, Vol. 50, pp Breeden, D. T., Gibbons, M. R., and Lizenberger, R. H., 1989, "Empirical ess of he consumpion-oriened CAPM", Journal of Finance, Vol. 44, pp Brennan, M. J., and Subrahmanyam, A., 1996, Marke microsrucure and asse pricing: on he compensaion for illiquidiy in sock reurns, Journal of Financial Economics, Vol. 41, pp Brennan, M., and Xia, Y., 2005, ay's as good as cay, Finance Research Leers, Vol. 2, pp Campbell, J., and J. Cochrane, 1999, "By force of habi: a consumpion-based explanaion of aggregae sock marke behaviour", Journal of Poliical Economy, Vol. 107, pp Campbell, J., and Mankiw, N., 1989, "Consumpion, income, and ineres raes: reinerpreing he imes series evidence", in Blanchard, O., and Fischer, S. (Eds.), 17

18 NBER Macroeconomics Annual, MIT Press, Cambridge, Massachusses, pp Campbell, J. Y., and Shiller, R. J., 1991, Yield spreads and ineres raes: a bird s eye view, Review of Economic Sudies, Vol. 58, pp Campbell, J. Y., and Shiller, R. J., 1988, "The dividend-price raio and expecaions of fuure dividends and discoun facors", Review of Financial Sudies, Vol. 1, pp Chien, Y., and Lusig, H., 2010, The marke price of aggregae risk and he wealh disribuion, Review of Financial Sudies, Vol. 23, pp Clark, T., and McCracken, M., 2001, Tess of equal forecas accuracy and encompassing for nesed models, Journal of Economerics, Vol. 105, pp Cochrane, J. H., and Piazzesi, M., 2005, Bond risk premia, American Economic Review, Vol. 95 No. 1, pp Consaninides, G., 1990, "Habi-formaion: a resoluion of he equiy premium puzzle", Journal of Poliical Economy, Vol. 98, pp Dickey, D. A., and Fuller, W. A., 1979, Disribuions of he esimaors for auoregressive ime series wih a uni roo, Journal of American Saisical Associaion, Vol. 74, pp Diebold, F., and Mariano, R., 1995, Comparing predicive accuracy, Journal of Economics and Business Saisics, Vol. 13, pp Duffee, G., 2005, "Time variaion in he covariance beween sock reurns and consumpion growh", Journal of Finance, Vol. 60 No. 4, pp Engen, E. M., and Hubbard, R. G., 2005, Federal governmen deb and ineres raes, NBER Macroeconomics Annual, Vol. 19. Engle, R., and Granger, C., 1987, Co-inegraion and error-correcion: represenaion, esimaion and esing, Economerica, Vol. 55 No. 2, pp Fama, E., 1998, "Marke efficiency, long erm reurns and behavioral finance", Journal of Financial Economics, Vol. 49 No. 3, pp Fama, E. F., and Bliss, R. H., 1987, The informaion in long-mauriy forward raes, American Economic Review, Vol. 77 No. 4, pp Fama, E., and French, K., 1996, "Mulifacor explanaions of asse pricing anomalies", Journal of Financial Economics, Vol. 51 No. 1, pp Fama, E. F., and French, K. R., 1989, Business condiions and expeced reurns on socks and bonds, Journal of Financial Economics, Vol. 25, pp

19 Fernandez-Corugedo, E., Price, S., and Blake, A., 2007, The dynamics of consumers expendiure: he UK consumpion ECM redux, Economic Modelling, Vol. 24, pp Gale, W. G., and Orszag, P. R., 2003, Economic effecs of susained budge deficis, Naional Tax Journal, Vol. 56, pp Goyal, A., and Welch, I., 2003, Predicing he equiy premium wih dividend raios, Managemen Science, Vol. 49, pp Goyal, A., and Welch, I., 2004, A comprehensive look a he empirical performance of equiy premium predicion, NBER Working Paper No Harvey, D., Leybourne, S., and Newbold, P., 1998, Tess for forecas encompassing, Journal of Business and Economic Saisics, Vol. 16, pp Hjalmarsson, E., 2006, Should we expec significan ou-of-sample resuls when predicing sock reurns?, Board of Governors of he Federal Reserve Sysem, Inernaional Finance Discussion Paper No Hodrick, R. J., 1992, "Dividend yields and expeced sock reurns: alernaive procedures for inference and measuremen", Review of Financial Sudies, Vol. 5, pp Honohan, P., and Laeven, L., 2005, Sysemic financial disress: conainmen and resoluion, Cambridge: Cambridge Universiy Press. Inoue, A., and Kilian, L., 2004, In-sample or ou-of-sample ess of predicabiliy: which one should we use?, Economeric Reviews, Vol. 23, pp Jagannahan, R., and Wang, Z., 1996, "The condiional CAPM and he cross-secion of expeced reurns", Journal of Finance, Vol. 51, pp Johansen, S., and Juselius, K., 1990, Maximum likelihood esimaion and inference on coinegraion wih applicaions o he demand for money, Oxford Bullein of Economics and Saisics, Vol. 52, pp Julliard, C., 2004, "Labor income risk and asse reurns", Princeon Universiy, Primary Job Marke Paper. Kwiakowski, D., Phillips, P. C. B., Schmid, P., Shin, Y., 1992, Tesing he null hypohesis of saionariy agains he alernaive of a uni roo: How sure are we ha economic ime series have a uni roo?, Journal of Economerics, Vol. 54 No. 1-3, pp Laeven, L., and Valencia, F., 2008, Sysemic banking crises: a new daabase, IMF Working Paper No

20 Lamon, O., 1998, Earnings and expeced reurns, Journal of Finance, Vol. 53, pp Leau, M., Ludvigson, S., 2004, "Undersanding rend and cycle in asse values: reevaluaing he wealh effec on consumpion", American Economic Review, Vol. 94, pp Leau, M., and S. Ludvigson, 2001, Consumpion, Aggregae Wealh, and Expeced Sock Reurns, Journal of Finance, Vol. 41 No. 3, pp Ludvigson, S., and Ng, S., 2009, Macro facors in bond risk premia, Review of Financial Sudies, Vol. 22 No. 12, pp Lusig, H., and van Nieuwerburgh, S., 2005, "Housing collaeral, consumpion insurance, and risk premia: an empirical perspecive", Journal of Finance, Vol. 60, pp Mackinnon, J., 1996, Numerical disribuion funcions for uni-roo and coinegraion ess, Journal of Applied Economerics, Vol. 11, pp Mankiw, N. G., and Shapiro, M. D., 1986, "Risk and reurn: consumpion bea versus marke bea", Review of Economics and Saisics, Vol. 68, pp McCracken, M., 2006, Asympoics for ou-of-sample ess of causaliy, Journal of Economerics, Vol. 140, pp Nelson, C. R., and Kim, M. J., 1993, Predicable sock reurns: The role of small sample bias, Journal of Finance, Vol. 48 No. 2, pp Newey, W., and Wes, K., 1987, A simple posiive semi-definie, heerokedasiciy, and auocorrelaion consisen covariance marix, Economerica, Vol. 55 No. 3, pp Pasor, L., and Sambaugh, R. F., 2003, Liquidiy risk and expeced sock reurns, Journal of Poliical Economy, Vol. 3, pp Parker, J. A., and Julliard, C., 2005, "Consumpion risk and he cross secion of expeced reurns", Journal of Poliical Economy, Vol. 113, pp Phillips, P. C. B., and Ouliaris, S., 1990, Asympoic properies of residual based ess for coinegraion, Economerica, Vol. 58, pp Piazzesi, M., Schneider, M., and Tuzel, S., 2007, "Housing, consumpion and asse pricing", Journal of Financial Economics, Vol. 83, pp Richards, A., 1995, "Comovemens in naional sock marke reurns: evidence of predicabiliy, bu no coinegraion", Journal of Moneary Economics, Vol. 36, pp

21 Sanos, T., and Veronesi, P., 2006, Labor income and predicable sock reurns, Review of Financial Sudies, Vol. 19, pp Shiller, R. J., 1984, "Sock prices and social dynamics", Brookings Papers on Economic Aciviy, Vol. 84, pp Sousa, R. M., 2010, Consumpion, (dis)aggregae wealh, and asse reurns, Journal of Empirical Finance, Vol. 17 No. 4, pp Sousa, R. M., 2007, Wealh shocks and risk aversion, Universiy of Minho, NIPE Working Paper Nº. 28. Sambaugh, R. F., 1999, Predicive regressions, Journal of Financial Economics, Vol. 54 No. 3, pp Sigliz, J., and Weiss, A., 1981, Credi raioning in markes wih imperfec informaion, American Economic Review, Vol. 71 No. 3, pp Valkanov, R., 2003, Long-horizon regressions: heoreical resuls and applicaions, Journal of Financial Economics, Vol. 68, pp Whelan, K., 2008, Consumpion and expeced asse reurns wihou assumpions abou unobservables, Journal of Moneary Economics, Vol. 55, No. 7, pp Yogo, M., 2006, "A consumpion-based explanaion of expeced sock reurns", Journal of Finance, Vol. 61 No. 2, pp Zeldes, S., 1989, Opimal consumpion wih sochasic income: deviaions from cerainy equivalence, Quarerly Journal of Economics, Vol. 104 No. 2, pp

22 Lis of Tables Table 1 Coinegraion esimaions. hwy log( hw ) log( y ). ^ Augmened Dickey and Fuller (1979) -saisic Lags: Auomaic based on SIC MacKinnon (1996) Criical values Kwiakowski e al. (1992) LM-saisic 5% 10% Bandwidh: Newey-Wes using Barle kernel ^ Kwiakowski e al. (1992) Criical values 5% 10% Ausralia 1.89*** (2.57) Ausria 27.75*** (4.62) Belgium 4.73*** (8.37) Canada *** (-2.93) Denmark 12.38*** (3.42) Finland 1.80*** (3.81) France -4.01*** (-2.95) Germany 0.54*** (2.87) Ireland 4.09*** (5.58) Ialy 1.25*** (3.00) Japan 2.18*** (5.79) Neherlands 4.17*** (8.31) Spain * (-1.40) Sweden 4.63*** (3.47) UK 2.59*** (3.73) US 4.48*** (9.31) Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. *, **, *** - saisically significan a he 10, 5, and 1% level, respecively. ^ ^ 22

23 Ausralia -0.20*** (-2.61) [0.05] Ausria 0.00 (0.59) Belgium 0.17*** (2.92) [0.08] Canada (-0.34) Denmark 0.03*** (2.33) Finland -0.11** (-2.07) France (-0.43) Germany -0.27** (-2.38) [0.04] Table 2 Forecasing real sock reurns: esimaed effec of hwy. Forecas Horizon H Forecas Horizon H *** -0.53*** -0.64*** -0.99*** Ireland (-3.23) (-3.58) (-3.70) (-3.83) (0.87) (1.31) (1.53) (1.50) [0.13] [0.16] [0.10] [0.12] 0.00 (0.43) 0.34*** (3.80) [0.12] (-0.13) 0.05*** (2.76) [0.04] -0.25*** (-2.88) [0.05] (-0.61) -0.56*** (-3.27) [0.06] 0.00 (0.46) 0.49*** (4.28) [0.15] 0.00 (0.13) 0.08*** (2.85) [0.04] -0.39*** (-3.12) [0.06] (-0.51) -0.87*** (-4.03) [0.08] 0.00 (0.20) 0.62*** (4.41) [0.17] 0.01 (0.36) 0.11*** (3.18) [0.05] -0.52*** (-3.24) [0.08] (-0.22) -1.18*** (-4.75) [0.11] -0.02* (-1.92) 0.93*** (4.71) [0.16] 0.04 (1.37) 0.23*** (3.96) [0.11] -1.06*** (-3.53) [0.13] 0.10 (1.06) -2.10*** (-7.39) [0.16] Ialy -0.25** (2.14) [0.06] Japan 0.08 (1.20) Neherlands 0.02 (0.23) Spain (-1.38) Sweden 0.16*** (2.80) [0.07] UK -0.20* (-1.75) [0.06] US -0.16* (-1.81) -0.41** (-2.07) [0.05] 0.13 (1.18) 0.00 (0.04) (-1.24) 0.29*** (3.27) [0.09] -0.34* (-1.71) [0.06] -0.33** (-2.16) [0.04] -0.47* (1.92) [0.04] 0.17 (1.22) (-0.18) (-1.08) 0.38*** (3.44) [0.09] -0.46* (-1.78) [0.08] -0.45** (-2.32) [0.05] (-1.63) 0.19 (1.11) (-0.30) (-0.98) 0.47*** (3.65) [0.10] -0.54* (-1.82) [0.08] -0.62*** (-2.71) [0.06] 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 (0.10) 0.23 (0.87) 0.07 (0.31) (-0.95) (-0.39) 0.86*** (5.25) [0.18] -0.59** (-2.13) [0.05] -1.42*** (-4.89) [0.16] Ausralia -0.05** (-1.96) Ausria (-0.30) Belgium 0.11*** (6.62) [0.25] Canada 0.01*** (3.12) [0.05] Denmark (-1.12) Finland -0.10*** (-4.37) [0.09] France 0.01** (2.35) Germany 0.05*** (2.70) Table 3 Forecasing real governmen bond reurns: esimaed effec of hwy. Forecas Horizon H Forecas Horizon H ** -0.16* -0.21** -0.38*** Ireland 0.08*** 0.17*** 0.25*** 0.34*** (-1.95) (-1.94) (-1.96) (-1.87) (4.63) (4.59) (4.57) (4.61) [0.11] [0.11] [0.12] [0.12] (-1.62) 0.22*** (6.92) [0.28] 0.02*** (3.46) [0.06] (-1.29) -0.21*** (-5.25) [0.11] 0.03** (2.36) 0.12*** (4.07) [0.06] (-1.24) 0.32*** (6.92) [0.28] 0.02*** (3.81) [0.07] (-1.50) -0.32*** (-5.73) [0.12] 0.04** (2.36) 0.20*** (4.90) [0.08] (-1.26) 0.43*** (6.92) [0.28] 0.04*** (4.22) [0.08] -0.07* (-1.72) -0.43*** (-5.95) [0.13] 0.05** (2.31) 0.30*** (5.78) [0.11] 0.00 (0.07) 0.80*** (6.65) [0.27] 0.09*** (6.02) [0.14] -0.18** (-2.32) -0.91*** (-6.68) [0.15] 0.10** (2.26) 0.81*** (9.42) [0.26] Ialy 0.05** (2.01) Japan 0.05 (0.96) Neherlands -0.06*** (-3.04) [0.08] Spain -0.02*** (-7.69) [0.44] Sweden 0.04* (1.82) UK 0.01 (0.54) US 0.21*** (7.44) [0.30] 0.08* (1.79) 0.10 (1.38) -0.13*** (-3.94) [0.11] -0.03*** (7.61) [0.46] 0.05 (1.62) 0.03 (0.77) 0.42*** (7.56) [0.31] 0.10 (1.50) 0.14** (2.15) [0.06] -0.20*** (-4.07) [0.11] -0.05*** (-7.98) [0.47] 0.08* (1.84) 0.06 (1.14) 0.63*** (7.60) [0.30] 0.11 (1.18) 0.17*** (3.41) [0.09] -0.25*** (-3.90) [0.11] -0.06*** (-8.58) [0.49] 0.10** (2.20) [0.04] 0.10 (1.46) 0.81*** (7.38) [0.29] 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. 0.79*** (5.30) [0.17] 0.04 (0.22) 0.32*** (3.25) [0.09] -0.46*** (-3.59) [0.10] -0.12*** (-10.57) [0.49] 0.13 (1.35) 0.31*** (2.59) 1.41*** (6.28) [0.25] 23

24 Table 4.1 Forecasing real sock reurns: addiional conrol variables. hwy -1 DivYld -1 Adj. R-square hwy -1 DivYld -1 Adj. R-square Ausralia -0.19** (2.34) 5.55** (2.22) [0.08] Ireland 0.04 (0.87) Ausria 0.00 Ialy -0.23** 20.48*** [0.14] (0.59) (-2.02) (3.20) Belgium 0.12* [0.04] Japan ** [0.04] (1.85) (-0.17) (0.12) (2.00) Canada Neherlands 0.73** 6.24 [0.15] (-0.24) (1.20) (2.44) (0.66) Denmark 0.03*** Spain (2.33) (-1.38) Finland -0.22*** [0.05] Sweden 0.13** 12.48*** [0.12] (-2.61) (-0.66) (2.46) (2.73) France UK *** (-0.33) (0.70) (-0.01) (-0.01) Germany -0.68*** (-2.98) 11.33*** (2.75) [0.11] US 0.16* (1.68) (-0.04) Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. *, **, *** - saisically significan a he 10, 5, and 1% level, respecively. Table 4.2 Forecasing real governmen bond reurns: addiional conrol variables. hwy -1 Inflaion -1 Defici -1 Adj. R-square hwy -1 Inflaion -1 Defici -1 Adj. R-square Ausralia (-1.59) 0.00 (1.10) 0.01 (0.26) Ireland 0.05** (2.01) Ausria Ialy 0.06*** 0.01*** 0.33*** [0.77] (-0.35) (0.85) (3.88) (9.55) (10.02) Belgium 0.04** *** [0.49] Japan *** 3.53** [0.31] (2.49) (-0.28) (-4.93) (0.47) (4.67) (2.06) Canada ** 0.04 [0.07] Neherlands -0.05*** *** [0.28] (-0.51) (2.08) (0.05) (-2.59) (1.33) (5.83) Denmark *** [0.16] Spain -0.01*** 0.01** 0.29*** [0.58] (-1.15) (5.41) (-6.42) (2.19) (3.11) Finland -0.05* -0.00*** 0.35*** [0.20] Sweden (-1.84) (-3.87) (4.22) (1.51) (0.62) (0.30) France 0.02*** 0.01*** 0.02 [0.30 UK *** 0.03 [0.13] (4.01) (4.89) (0.43) (-0.51) (3.48) (0.46) Germany 0.05** (2.47) 0.00** (1.98) 0.19** (2.02) [0.08] US 0.17*** (5.97) 0.02*** (7.49) 0.28 (1.35) [0.52] Noes: Newey-Wes (1987) correced -saisics appear in parenhesis. *, **, *** - saisically significan a he 10, 5, and 1% level, respecively. Table 5 One-quarer ahead forecass of reurns: hwy model vs. consan/ar models. Real sock reurns Real governmen bond reurns MSE hwy /MSE consan MSE hwy /MSE AR MSE hwy /MSE consan MSE hwy /MSE AR Ausralia Ausria Belgium Canada Denmark Finland France Germany Ireland Ialy Japan Neherlands Spain Sweden UK US Noes: MSE mean-squared forecasing error. *, **, *** - saisically significan a he 10, 5, and 1% level, respecively. 24