Government Debt Spillovers in a Federation*

Transcription

1 Government Debt Spllovers n a Federaton* Stuart Landon and Constance E. Smth** Department of Economcs Unversty of Alberta Edmonton, Alberta, Canada T6G 2H4 Aprl 2006 Abstract Ths paper presents estmates of the mpact of debt ssued by one government n a monetary unon on the yelds of bonds ssued by other governments n the unon. These debt spllovers may occur f there s a rsk of monetary accommodaton, mplct or explct nter-jursdctonal balout provsons, or nterdependent revenue sources. The analyss emprcally dstngushes between two channels through whch debt spllovers may affect bond yelds: currency deprecaton rsk and default rsk. Data on the yelds of ndvdual Canadan provncal government bonds for the perod are employed. No evdence s found of debt spllover effects between provnces, but a 1 percentage pont ncrease n the central government s debt to GDP rato rases the yeld on provncal government bonds by 4.3 bass ponts 3.0 bass ponts by ncreasng the expected deprecaton rate of the Canadan dollar and 1.3 bass ponts by rasng the rsk of provncal government default. These results mply that a rse n the Canadan central government debt to GDP rato from.25 to.58, equvalent to the rse that occurred between 1983 and 1995, would lead to an ncrease n provncal government bond yelds of 142 bass ponts. JEL Classfcaton: F33, F36, F42, H63. Keywords: bond yelds; asset returns; government debt; government default rsk; spllovers. * The authors thank the Faculty of Arts (Support for the Advancement of Scholarshp Fund) and the Vce- Presdent Research at the Unversty of Alberta for fnancal support. Carmen Stefanescu and Mchael Januska provded excellent assstance wth the data. Valuable comments were receved from Paul Boothe and semnar partcpants at the Unversty of Alberta. **Correspondng author. Phone: (780) ; fax: (780) , emal: constance.smth@ualberta.ca

2 1. Introducton The creaton of the European Monetary Unon (EMU) and proposals to establsh currency unons n Asa, Afrca and the Amercas have generated ncreased nterest n the role of fscal arrangements and fscal nsttutons n federatons and monetary unons. One ssue that has been of partcular nterest s the role of government debt spllovers the mpact of debt accumulaton by one government n a federaton or currency unon on the yelds of the bonds ssued by other member governments. These debt spllovers may arse f there s a rsk of monetary accommodaton, mplct or explct nter-jursdctonal balout provsons, or nterdependent revenues. Debt spllover effects may be mportant snce, as noted n the theoretcal studes of Bergn (2000), Canzoner, Cumby and Dba (2001), and Char and Kehoe (2003), the ncentves nherent n currency unons can lead ndvdual governments to choose a hgher level of debt than s optmal from the perspectve of the unon as a whole. For example, Char and Kehoe (2003) show that there s a tme nconsstency problem n monetary polcy f the monetary authorty cannot commt, because t has an ncentve to nflate away nomnal debt when member states debt levels are hgh. As ndvdual fscal authortes take account only of the costs of nflaton to ther own resdents, each fscal authorty ssues too much debt. 1 The exstence of ntergovernmental debt spllovers mples that the yelds on bonds ssued by one government n a monetary unon wll rse wth the debt levels of other governments. Hence, spllovers may mpose costs on the group s more fscally prudent members. Although debt spllovers may have potentally sgnfcant pecunary externaltes, few emprcal studes attempt to 1 Devereux (1993) emphaszes ths type of spllover effect n the Canadan context. Some observers argue that the defct to GDP and debt to GDP lmts mposed on members of the European Monetary Unon are meant to prevent suboptmal debt level choces by member states (Bovenberg, Kremers and Masson (1991, 375)). These rules reman n place, but on 20 March 2005 the European Councl ntroduced changes to the enforcement of the debt and defct rules to allow longer adjustment tmes and more exceptons (European Unon, 2005). See Feldsten (2005) for a dscusson of the new enforcement gudelnes. 1

3 verfy the presence of these spllovers or determne ther magntude. 2 Evdence that ndrectly supports the exstence of debt spllovers n the EMU s reported n Bernoth, von Hagen and Schuknecht (2004). They show that EMU membershp reduces the effect of own-country debt accumulaton on nterest rates, a result they vew as beng consstent wth markets antcpatng fscal support for EMU countres n fnancal dstress. Usng data for Canada, Landon and Smth (2000) fnd that central government debt accumulaton reduces the credtworthness of ndebted provncal governments, but the focus of ther study s on credt ratngs, so the fndngs cannot be used to quantfy the cost of spllovers n terms of bond yelds. 3 The purpose of the current analyss s to contrbute to the understandng of the lnk between debt accumulaton, spllovers and bond yelds. Specfcally, usng data for Canada, estmates are presented of the mpact on ndvdual provncal government bond yelds of debt accumulaton by other provncal (subcentral) governments and the federal (central) government. Whle much of the nterest n the relatonshp between government debt and bond yelds has focused on the European Unon, several features of the Canadan case facltate the analyss of debt spllovers. For example, provncal governments n Canada exert consderable fscal ndependence and there are no federal government or consttutonal restrctons on provncal government borrowng. Debt to GDP ratos vary sgnfcantly across provnces and across tme (from a low of to a hgh of 0.51 between 1983 and 2003), and provncal government debt s often szable, so the rsk of provncal government default s not trval. 4 As well, aggregate provncal government expendtures (equal to 2 Whle there s relatvely lttle research on government debt spllover effects, numerous studes nvestgate whether a rse n own-country debt ncreases own-country bond yelds. See, for example, Alesna, De Broeck, Prat and Tabelln (1992), Codogno, Favero, and Mssale (2003), Ardagna, Casell and Lane (2004), and Chnn and Frankel (2004) for cross-country studes. Laubach (2003) and Engen and Hubbard (2005) are recent US studes. 3 In other work on spllovers, Gande and Parsley (2005) dentfy a spllover from a soveregn debt credt ratng change for one country onto nterest rate spreads n other countres. However, ther focus s on the lnk mpled by nternatonal trade and captal flows, rather than government debt accumulaton. Gulodor and Beetsma (2004) use a vector autoregressve model to analyze spllover effects of fscal shocks on trade n Europe, but do not examne the mpact on nterest rates. 4 In the early 1990s, the Saskatchewan government faced mountng nterest payments and speculaton that t mght default on ts debt (Drummond and Burleton, 2004). One provncal government defaulted n 1936, but eventually receved federal support (Ascah, 1999). 2

4 more than 20 percent of GDP) are larger than those of the federal government (approxmately 16 percent of GDP). Snce Canada s a sngle naton state, comparable and consstent output and government debt data are avalable and nsttutons are more smlar than would be the case for a cross secton of countres. 5 Furthermore, the market for Canadan provncal government bonds has consderable sze and lqudty (Kahn and Gulrajan, 1993). 6 Fnally, fnancal markets n Canada are relatvely free and open, and there are few restrctons on the moblty of captal across provncal or nternatonal boundares. Usng a panel of annual provncal bond yeld data for , ths study fnds that debt accumulaton by the federal government of Canada has a sgnfcant spllover effect on the yelds of bonds ssued by Canadan provncal governments. A one percentage pont ncrease n the federal government s debt to GDP rato s predcted to rase the yeld on provncal government bonds by 4.3 bass ponts 3.0 bass ponts from an ncrease n the expected deprecaton rate of the Canadan dollar and 1.3 bass ponts from a rse n the rsk of provncal government default. Between 1983 and 1995, the federal government debt to GDP rato ncreased from.25 to.58. Accordng to the estmates presented below, an ncrease of ths magntude would rase provncal government bond yelds by 142 bass ponts. In contrast to the sgnfcant mpact of central government debt on provncal government bond yelds, provncal government debt accumulaton does not have a statstcally sgnfcant effect on the yelds of the bonds ssued by ndvdual provnces. A possble explanaton for the dfferent mpact on yelds of federal and provncal debt s that, n Canada, snce the federal government has exclusve responsblty for monetary polcy, fnancal markets beleve that only federal government debt accumulaton wll lead to nflaton-nducng monetzaton. One mplcaton of the absence of 5 Lemmen and Goodhart (1999, 92) argue that future research efforts should rely on the comparson of federal and provncal debt yelds wthn exstng federal states snce studes that use cross-country data provde a less clean comparson, due to dfferent country-specfc factors that may confound the estmated default rsk measure. Obstfeld (1992) also recognzes ths advantage of data for a sngle country. 6 In 2002, the outstandng stock of Canadan provncal government bonds was C$256 bllon. See Statstcs Canada, Cansm database, seres V

5 evdence of provncal government debt spllover effects s that spllovers do not provde a ratonale for the mposton of restrctons on provncal (sub-central) government debt accumulaton. The outlne of the paper s as follows. The next secton presents arguments for a lnk between the debt ssued by one government n a federaton (or currency unon) and the yelds on the bonds ssued by the other governments n the federaton. In Secton 3, the emprcal specfcaton s developed, whle the data are descrbed n Secton 4. Results are presented n Secton 5 and concludng remarks are gven n Secton Potental Causes of Government Debt Spllovers The ssue of government debt spllovers s closely related to the large lterature on the mpact of own-government debt on own-government borrowng costs. 7 As noted by Alesna, De Broeck, Prat and Tabelln (1992), own-government debt accumulaton can affect own-government borrowng costs by alterng two types of rsk: default rsk and devaluaton rsk. Snce governments wth larger debt levels are less lkely to be able to meet ther requred nterest and prncpal payments, debt accumulaton rases the rsk of default. The rsk of a currency deprecaton also ncreases wth own debt accumulaton f hgh debt countres are more lkely to nflate away ther domestc currency denomnated debt (Sargent and Wallace, 1981; Woodford, 2001). 8 In contrast to the effects of own debt accumulaton, debt spllovers occur when the debt accumulaton of one government causes a change n the default rsk or devaluaton rsk assocated wth another government n a federaton or currency unon and ths, n turn, rases the borrowng 7 Recent studes for the US nclude Canzoner, Cumby and Dba (2001), Laubach (2003), and Engen and Hubbard (2005). For studes of subcentral governments wthn a federaton or currency unon, such as the US, Canada and the European Unon, see Lane (1993), Goldsten and Woglom (1992), Bayoum, Goldsten and Woglom (1995), Poterba and Reuben (2001), and Lemmen (1999). Cross-country studes nclude Alesna, De Broeck, Prat and Tabelln (1992), Codogno, Favero, and Mssale (2003), Ardagna, Casell and Lane (2004), Bernoth, von Hagen and Schuknecht (2004), and Chnn and Frankel (2004). 8 Sargent and Wallace (1981) and Woodford (2001) argue that, f a government s budget constrant s to hold n present value terms, the government s approach to debt accumulaton and fscal solvency can lmt the optons open to the central bank, even f the central bank has been granted legal ndependence. If the monetary authorty s to have the functonal ndependence necessary to acheve prce stablty, the government must ensure fscal solvency for any potental path of the prce level. 4

6 costs of the other government. Ths secton presents several explanatons for why such a change may occur. These explanatons relate to the possblty of monetary accommodaton, the exstence of mplct or explct balout provsons, and the nterdependence of government revenues. If a government has control of monetary polcy, t can meet ts domestc currency debt oblgatons by monetzng ts debt. The greater the rsk of debt monetzaton, the hgher the yelds on all debt denomnated n the domestc currency, both debt ssued by the monetzng government as well as debt ssued by other domestc governments, due to the ncreased rsk of nflaton and currency deprecaton. The sze of the debt of the government n control of monetary polcy s lkely to be an mportant determnant of the rsk of monetzaton snce, wth a larger debt, governments may be vewed as beng more nclned to use an nflaton tax rather than an explct tax. In the context of a federaton, ths means that, f the central government has control of monetary polcy, ncreases n the magntude of central government debt wll rase the rsk of monetary accommodaton and ncrease the yelds on all debt nstruments denomnated n the federaton s currency. As a result, central government debt accumulaton wll spll over onto the yelds of sub-central government debt. It seems lkely that the market wll vew the probablty of monetzaton to be hgher the greater s the accumulated debt of the government n control of monetary polcy. However, f there are mplct or explct balout provsons n a federaton or currency unon, the market may beleve that monetary accommodaton could be used to reduce the real value of the debt of ndvdual member governments, as an alternatve to an outrght balout, even f these governments do not drectly control monetary polcy (McKnnon, 1995; Lemmen and Goodhart, 1999). As well, f a government or group of governments experence fnancal dstress, pressure may be put on the central bank to keep nterest rates low, n order to mnmze the costs of a fnancal crss, and ths could eventually lead to nflaton and a currency deprecaton. Thus, the market may vew the rsk of monetary accommodaton to be greater the larger s the debt of the member governments of a 5

7 federaton, both those wth and those wthout drect control of monetary polcy. 9 Even n the absence of monetary accommodaton, the exstence of mplct or explct balout provsons can lead to debt spllovers. Ths follows because the nterest rate demanded by lenders wll reflect the ablty to pay of the government(s) that s (are) lkely to provde the balout. That s, the rsk premum ncorporated by the market n bond yelds of borrowng governments wll depend, to some extent, on the debt oblgatons of the guarantor government. In a federaton, f lenders expect that the ndvdual jursdctons of the federaton wll be supported by the fnancal resources of the entre federaton, the yelds on the debt of ndvdual members wll depend on the debt oblgatons of the entre federaton. If the guarantor s the central government only, the yelds on sub-central government debt wll depend on the central government s debt commtments as these wll determne the ablty of the central government to fulfll ts guarantor role. Revenue nterdependence s an addtonal explanaton for why the accumulaton of government debt n one regon may affect the bond yelds of other jursdctons. Interdependent revenues exst to the extent that the economes of the members of a federaton are nterconnected as a result of trade flows, for example. Excess debt accumulaton that causes a fscal crss n one jursdcton may adversely affect output and ncomes n other jursdctons. Ths would decrease the tax revenues generated n the affected jursdctons and nhbt the ablty of the governments of these jursdctons to meet ther debt servce oblgatons. Further, f the central government s debt grows, the market may perceve an ncrease n the lkelhood of a cut n transfers to sub-central governments and, therefore, an ncrease n the probablty that lower-level governments wll have dffculty meetng ther debt oblgatons. 9 For example, to prevent bank falures when the natonal bankng system owns a large proporton of domestc debt, McKnnon (1995, 474) argues that solvent members of the EMU mght feel pressured to bal out an nsolvent government by askng the European Central Bank to buy the troubled government s bonds. Further, Lemmen and Goodhart (1999, 81) note that, although the Maastrcht Treaty does not allocate a lender of last resort role to the European System of Central Banks, t s yet to be determned how fnancal crses would be detected, montored and resolved. They also pont out that the Maastrcht Treaty s no-balout provson may be at odds wth other functons of the ECB, such as the smooth functonng of the payments system. 6

8 3. Specfcaton of the Emprcal Model Ths secton descrbes the dervaton of an estmable model of provncal government bond yelds that ncorporates nter-government debt spllover effects. Suppose provncal (subcentral) government bonds are purchased n a compettve fnancal market by rsk-neutral lenders. The relatonshp between the expected return on a domestc currency-denomnated provncal government bond ssued by provnce, r, and the yeld on a default rsk-free domestc currencydenomnated natonal (central) government bond, r C, can be expressed as: where C (1 + )(1 ) = (1 + r ), (1) r m ρ m m s the probablty of default by provnce between the current perod and perod m, so the subscrpt m represents maturty length. The central government bond s default rsk-free snce the central government has control of monetary polcy and so can always repay ts domestc currency debt. ρ m In addton to provncal and central government bonds, nvestors can hold foregn bonds whch are denomnated n foregn currency. The dfference between the returns on domestc and foregn bonds nvolves exchange rate rsk, snce the value of the exchange rate when the bonds mature s uncertan. Gven rsk neutral nvestors, the yeld on a central government default rsk-free domestc currency government bond must equal the expected yeld on a foregn currency default rsk-free bond when returns are converted to a common currency: C 1 + r m = e f Sm (1 + rm ), (2) S C f where s the yeld on a domestc central government bond wth m perods to maturty, s the r m yeld on a foregn currency bond, S s the current value of the exchange rate (the domestc currency prce of one unt of foregn currency), e S m s the expected value of the exchange rate when the bonds mature n m perods, and there are assumed to be neglgble transactons costs and no restrctons on r m foregn ownershp. 7

9 Takng log approxmatons of equatons (1) and (2) and rearrangng yelds: C r = r +, (3) m m C f r = r + m m ρ m εm, (4) where ε m s the expected log change n the exchange rate (the expected rate of currency deprecaton) between the current perod and perod m, so ε [ln( S e m m) ln(s)]. 10 The probablty of default, ρ n equaton (3), s determned by government debt spllover factors that ncrease the lkelhood of default by provnce, as well as by non-spllover factors that affect the provnce s probablty of default. Thus, the default rsk n perod t of a bond ssued by provnce wth maturty m can be expressed as: ρ ρ m,t = α Z Z t + α X X t + αwwt, (5) where Z s a vector of spllover varables that depend on the debt levels of the other provnces and the central government, X s a vector of factors specfc to provnce that may affect ts rsk of default, and ρ W s a vector of non-spllover factors, not specfc to provnce, that may affect the provnce s probablty of default. Changes n the antcpated future value of the exchange rate, ε m n equaton (4), may also depend on central and subcentral government debt levels because an ncrease n debt accumulaton by these governments may rase the rsk of a monetary expanson. As a result, ε m s specfed to depend on the spllover factors, Z, as well as on non-spllover factors, denoted expected rate of currency deprecaton:, that affect the ε ε m,t = β Z Z t + βwwt. (6) ε W 10 If nvestors are rsk averse, factors representng the varance of expected returns would also enter equatons (3) and (4). See, for example, the portfolo balance models of Solnk (1974) and Glassman and Rddck (1996) and, for the specfc case of soveregn debt rsk, see Bernoth, von Hagen and Schuknecht (2004). 8

10 Usng equaton (4) to substtute for C r m n equaton (3) yelds: f r = + ρ + ε. (3 ) m r m m m Rearrangng equatons (3) and (3 ), and usng equatons (5) and (6) to substtute for gves: ρm and ε m, C ρ r m,t rm,t = α Z Zt + α X X t + αwwt, (7) f r r = α Z Z + α X X t + m,t m,t t α ρ WW t + ε β ZZt + βwwt. (8) Estmates of the vectors of parameters α Z and β Z n equatons (7) and (8) can be used to determne the sgnfcance and magntude of spllover effects on bond yelds. The parameters n α Z represent the mpact of the spllover varables on the probablty of default ( ρ ), whle the parameters n reflect the mpact of the spllover varables on the expected rate of currency deprecaton ( ε ). By specfyng the dependent varables n equatons (7) and (8) as yeld dfferentals, t s possble to control for factors that may affect yelds on all bonds. Estmaton usng a yeld dfferental s standard n studes of own-government debt effects, such as, for example, Alesna, de Broeck, Prat and Tabelln (1992), Favero, Gavazz and Spaventa (1997), Lemmen and Goodhart (1999), Codogno, Favero and Mssale (2003), Ardagna, Casell and Lane (2004), and Bernoth, von Hagen, and Schuknecht (2004). m m βz 4. Emprcal Implementaton and the Data In order to estmate equatons (7) and (8), data are requred for the bond yelds (r, r C, r f ), the spllover varables (Z), and the non-spllover varables (X, ρ W, ). Data on the yelds of ndvdual provncal government bonds, publshed by the Fnancal Post Corporaton, are used to represent the provncal government bond yeld varable, r. Whle bond yeld data are generally avalable for several outstandng bonds at any one tme for each provnce, the yeld on only one bond for each provnce s used because the yelds on the dfferent bonds ssued by each provnce are 9 ε W

11 hghly correlated. 11 However, snce the data for a sngle bond dd not span the entre length of the sample for any provnce (as bonds matured durng the sample) the tme seres for r for each provnce ncorporated the yelds of several bonds. Bonds were chosen, from those avalable, n order to mnmze the number of dfferent bonds used for each provnce. Table A1 n the Appendx gves a complete lst of the ndvdual bonds employed. 12 Whle ndvdual bond yelds are avalable at a monthly frequency, only one observaton per year s utlzed n the emprcal analyss as data on the stock of provncal government debt are avalable only at an annual frequency as of fscal year-end 31 March. To avod possble endogenety between the level of provncal debt and provncal government bond yelds, the bond yeld observaton employed s the observaton for the last busness day n Aprl, the frst yeld observaton avalable followng 31 March. The average tme to maturty of the bonds n the sample s 10 years, whle the average ssue amount s approxmately C$250 mllon. The Appendx provdes detaled notes on the bond yeld data, as well as descrptons and sources for all the other data. The central government bond yeld, C r m, s represented by the yeld on a Government of Canada bond of maturty length m. Gven the open fnancal markets n Canada and the US, and the large sze of the US bond market, the natural choce for the return on the alternatve foregn bond, f r m, s the yeld on a US Treasury bond wth maturty m. 4.1 The Spllover Explanatory Varables, Z Two spllover varables are employed: the federal government s debt to GDP rato, denoted FEDDEBT, and the aggregate debt to GDP rato of the provnces, denoted PROVDEBT. 13 The use 11 It s not possble to use a benchmark bond or a bond ndex, snce these are not avalable for bonds ssued by ndvdual provncal governments. Further, the two yelds that form the dependent varables n equatons (7) and (8) are matched by maturty, and ths would be mpossble wth an ndex. 12 To check the robustness of ths selecton procedure, the parameters were re-estmated usng a completely dstnct sample of bond yelds. Ths sample yelded estmates very smlar to those reported for the orgnal sample. 13 Government debt s defned as fnancal assets mnus total drect labltes, excludng the labltes of government employee penson plans. Employee penson labltes are not ncluded n the debt measure because they are lkely to be 10

12 of PROVDEBT mples that the relevant subcentral government debt spllover varable s the aggregate level of provncal borrowng and that, as t relates to spllovers, a dollar of debt s treated the same no matter whch provncal government s the ssuer. 4.2 Non-Spllover Explanatory Varables To determne the magntude of debt spllover effects on provncal government bond yelds, ρ ε t s necessary to control for the non-spllover factors ( X, W, W ) that may also affect yelds through ther mpact on default rsk or the expected change n the exchange rate. The control varables were chosen to represent these effects n a parsmonous fashon and to be consstent wth the lterature that examnes (non-spllover) debt effects on yelds Provnce-Specfc Non-Spllover Explanatory Varbles, Own-government debt exposure and regonal ncome are among the most frequently employed varables used to explan government bond yelds (Edwards, 1984; Alesna, De Broeck, Prat and Tabelln, 1992; Bayoum, Goldsten and Woglom, 1995; Cantor and Packer, 1996; Echengreen and Mody, 1998). Each provnce s debt exposure s represented by ts own debt to X GDP rato, OWNDEBT. 14 Two varables are used to represent the determnants of regonal ncomes: the change n provncal real GDP, Y, and federal government transfers to provnce as a share of the provnce s total revenues, TRANSFERS. A jursdcton wth growng ncome s of a dfferent nature than other drect government labltes. For example, t may be easer for a government to renege on these commtments than to renege on ther bond labltes. In addton, the employee penson labltes of the provncal governments may not have been measured as accurately as other government labltes, snce these labltes were ether not assessed or assessed only perodcally durng much of the sample perod. The defnton of debt employed here also does not nclude the labltes of the Canada and Quebec penson plans, because they are not drect government labltes; nor does t nclude guaranteed labltes (such as the labltes of government enterprses) snce these are condtonal labltes. As reported n Secton 5 below, ncorporatng guaranteed labltes n the debt defnton has lttle mpact on the results. 14 Sx provncal governments have ntroduced legslatve restrctons to control debt accumulaton. Whle there s some evdence that strong tax and expendture lmts have had an mpact on borrowng n US states (Bayoum, Goldsten and Woglom, 1995; Poterba and Reuben, 2001), smlar results do not appear to hold for state debt lmts. Also, Mllar (1997, 6-7) notes that, wth the excepton of Alberta, none of the acts seems to clearly restrct the ablty of governments to create new debt and the provncal governments could amend, or even repeal ths ant-defct legslaton f complance becomes partcularly costly. 11

13 lkely to have more revenue flexblty, so faster growth should reduce yelds. 15 A provnce that s more dependent on federal transfers s lkely to be more vulnerable to a unlateral reducton n these transfers. The TRANSFERS varable may also reflect the revenue rasng capacty of a provnce, as those provnces wth poor revenue rasng capacty tend to receve the largest transfers. For smlar reasons, ths varable may proxy the general long run economc health of a provnce. As a result, hgher values for TRANSFERS should be assocated wth hgher yelds. To account for effects that dffer across bonds, but are constant through tme, the vector also ncludes a dummy varable for each bond. Because bonds are ssued by only one provnce, the set of bond fxed effects s perfectly colnear wth a set of provnce fxed effects and, thus, the bond fxed effects also control for the socal, poltcal, economc and nsttutonal characterstcs that are specfc to a provnce, but constant over tme. ρ Addtonal Non-Spllover Explanatory Varables, W and 16 ρ ε The elements of the vectors W and W represent non-spllover non-provnce specfc factors that, respectvely, determne common movements n provncal bond rsk prema and changes n the expected rate of currency deprecaton. Studes that examne long term bond rates typcally nclude controls for monetary polcy, 17 as changes n monetary polcy may cause movements n bond yelds by alterng both the rsk of default and the expected rate of currency ρ deprecaton. To capture the relatve tghtness of Canadan and US monetary polcy, W and both nclude the polcy rate dfferental r BankR r FFR, where r BankR s the Bank of Canada s Bank rate, ε W X ε W 15 Alesna, De Broeck, Prat and Tabelln (1992) fnd strong and very robust evdence that government bond yelds are affected by the level of economc actvty n all of the 12 OECD countres they consder. 16 Durng the sample perod, the Quebec government held a referendum on the queston of Quebec soveregnty, whch mght be expected to affect provncal bond yelds. However, Johnson and McIlwrath (1998) fnd that the referendum had only a transent mpact on Quebec and Ontaro provncal government bond yelds (they examne a 3-month perod). Gven ths transent effect, the referendum s unlkely to have affected the results here snce, as noted above, the bond yeld data employed n ths study are annual and measured at the end of Aprl, whle the referendum campagn began on 7 September 1995 (when the referendum queston was offcally announced) and the vote was held on 30 October See Echengreen and Mody (1998), Arora and Cersola (2001), Ardagna, Casell and Lane (2004), and Engen and Hubbard (2005). 12

14 used to capture the stance of Canadan monetary polcy, and r FFR s the US Federal funds rate. The polcy rate dfferental s ncluded n the vector ε W snce a relatve rse n the Canadan monetary polcy rate would be expected to lead to an ncrease n Canadan bond rates relatve to comparable US rates. The dfferental s ncluded n the vector ρ W as relatvely tghter Canadan monetary polcy condtons may rase expectatons of provncal government default rsk by reducng the avalablty of credt or rasng the debt servce burden of borrowers n Canadan currency. As shown n numerous emprcal studes (Meese and Rogoff, 1983, for example), t s dffcult to dentfy factors that can accurately forecast future exchange rate movements. For ths reason, factors that may determne expected changes n exchange rates ( ε ) are also dffcult to dentfy. However, lterature revews by Taylor and Taylor (2004) and Rogoff (1996) conclude that, over the longer term, purchasng power party may provde some gudance for predctons of exchange rate movements. Ths mples that the nflaton dfferental s a lkely determnant of long-term expected changes n exchange rates. To capture ths effect, the dfferentals between the Can US Can US Canadan and US nflaton rates over each of the two prevous years, π π and π π, 1 1 W 18 ε respectvely, are ncluded n the vector. Hgher values of these varables are expected to rase the yeld dfferental by ncreasng the expected rate of deprecaton ( ε ). Also ncluded as an element of ε W s the natonal growth rate, denoted Can Y, snce stronger natonal economc growth may reduce the probablty of a currency deprecaton and, thus, reduce the yeld dfferental. The bond yeld equatons, (3) and (4), and thus the estmatng equatons (7) and (8), are specfed for bonds wth maturty m. To account for maturty dfferences across bonds, a varable representng the log of the number of months to maturty for each bond, denoted MATURITY, s ncluded n both estmatng equatons. 18 Ardagna, Casell and Lane (2004) also nclude an nflaton varable n ther nterest rate equaton. 13

15 5. Emprcal Results Equatons (7) and (8) are estmated jontly usng panel data for the ten Canadan provnces over the perod Jont estmaton allows the mposton of the restrcton that the vectors of parameters, α, α and α, are the same n both equatons. As well, snce the Z X W dependent varables n equatons (7) and (8) both ncorporate r, the error terms are lkely to be correlated and jont estmaton takes ths correlaton nto account. From the estmates of equatons (7) and (8), t s possble to derve estmates of the parameters n equatons (5) and (6), the equatons that descrbe the rsk of default ( ρ ) and the expected rate of deprecaton ( ε ), respectvely. These estmates are reported n Columns I and II of Table 1. The coeffcent estmates assocated wth the control varables have the antcpated sgns, the estmates are not rejected by a test for autocorrelaton or a Reset test for model msspecfcaton, and the estmates explan a large proporton of the varaton n the yeld dfferental. One result of note n Table 1 s that a provnce s own debt to GDP rato (OWNDEBT ) has a statstcally nsgnfcant mpact on ts own bond yeld. A smlar fndng s not uncommon n studes that measure the mpact of own-debt on own-government bond yelds, but whch do not consder spllover effects. 20 The estmates presented n Table 1 show that a rse n the aggregate provncal government debt to GDP rato (PROVDEBT) has a statstcally nsgnfcant mpact on the rsk premum of provncal government bonds, ρ. One mplcaton of ths result s that the provncal governments are not perceved to be mplct guarantors of the debt ssued by the other provnces. The 19 The number of observatons used n the emprcal analyss s 203 rather than 210 (21 years multpled by 10 provnces) as, for 7 observatons, approprate provncal government bond yelds were not avalable. See the Appendx. Due to these mssng observatons, the sample s unbalanced. 20 Studes that have found no sgnfcant relatonshp between own-government debt and own-government bond yelds, or that fnd a postve relatonshp only at relatvely hgh levels of debt, nclude Ford and Laxton (1999), Alesna, De Broeck, Prat and Tabelln (1992), and Codogno, Favero and Mssale (2003). For Canada, Lemmen (1999) fnds some evdence that hgher provncal government debt rases provncal government bond yelds, but he uses a farly short sample perod (1992Q3-97Q4). None of these studes, nor those that fnd a sgnfcant postve effect, examne spllovers. 14

16 nsgnfcant coeffcent assocated wth the aggregate provncal debt varable n the equaton for the expected change n the exchange rate, ε, ndcates that the market does not expect the federal government to choose a more nflatonary polcy n response to hgher provncal debt levels. An mplcaton of these results s that, as there s no evdence of statstcally sgnfcant nterprovncal government debt spllovers, debt spllovers cannot be used as a ratonale to justfy restrctons on provncal government debt accumulaton. 21 Such fscal rules have been proposed as an alternatve to market-based fscal dscplne 22 when the possblty of spllovers exsts (Bergn 2000; Canzoner, Cumby and Dba, 2001; Char and Kehoe, 2003; European Commsson, 2002; Devereux, 1993). In contrast to the results for provncal government debt, as can be seen from Table 1, a rse n the federal government debt to GDP rato (FEDDEBT) has a sgnfcant effect on both the expected rate of currency deprecaton, ε, and the rsk of default,. The sgnfcant postve coeffcent assocated wth the federal debt varable n the exchange rate deprecaton ( ε ) equaton mples that the market beleves the federal government, through ts control of monetary polcy, may use nflaton to reduce the real value of ts own debt. The sgnfcant effect of federal debt on the provncal government debt rsk premum ( ρ ) s consstent wth the federal government beng n a weaker poston to provde balout funds to the provnces, or more lkely to cut ntergovernmental transfers, as ts debt rses. Further, the sgnfcant effect of the federal debt on the provncal debt rsk premum mples that a rse n the federal government debt to GDP rato has a greater mpact on provncal government bond yelds than on federal government yelds (snce federal yelds rse only due to the ncrease n the expected rate of currency deprecaton, whle ρ 21 Ths result does not mply that fscal rules could not be useful to meet objectves other than the preventon of government debt spllovers. For example, Fatás and Mhov (2006) consder the crcumstances under whch fscal rules n US states could reduce macroeconomc volatlty. 22 Market dscplne encourages fscal restrant by causng a wdenng of the dfferental between the market yelds pad by fscally mprudent and fscally prudent debtors. For the market to send the correct sgnal, the yeld on government debt should reflect the fscal polcy stance of the borrowng government alone, and there should be no spllover or externalty effects of the debt of one government on the yelds pad by other governments. Bshop, Damrau and Mller (1989), for example argue that market dscplne s an effectve alternatve to budget rules for the EMU. 15

17 provncal bond yelds rse as a result of changes n both the rsk premum and the expected rate of deprecaton). The coeffcents assocated wth the federal debt varable, FEDDEBT, n columns I and II of Table 1 ndcate that the largest component of the mpact on bond yelds of federal government debt accumulaton s through the expected change n the exchange rate, ε. Through ths channel, a one percentage pont ncrease n the federal government s debt to GDP rato ncreases the yeld on provncal government bonds by.0305 percentage ponts. A smlar ncrease n the federal debt would rase the default rsk premum on provncal bonds,, by only.0126 percentage ponts. One reason why default rsk may be relatvely small s the potentally hgh costs assocated wth default (loss of borrower reputaton, possble fnancal sector nstablty or crss). Alesna, De Broeck, Prat and Tabelln (1992, 431) note that, n OECD countres, the costs of default are lkely to greatly outwegh the benefts, even for hghly ndebted countes, so the default premum s lkely to be small. The larger mpact of federal debt accumulaton on deprecaton rsk s consstent wth studes of the mpact of own-debt on own-yelds whch fnd currency deprecaton rsk to be large relatve to default rsk. 23 Accordng to the results n Table 1, takng deprecaton and default rsk together, a one percentage pont rse n the federal government s debt to GDP rato s predcted to ncrease the yeld on provncal government bonds by 4.3 bass ponts. Gven that movements n Canadan federal government debt levels have often been substantal, federal debt accumulaton may have had a large mpact on provncal bond yelds. For example, from 1983 to 1995 the federal government debt to GDP rato rose from.248 to.583. Based on the coeffcent estmates n Table 1, an ncrease of ths ρ 23 See the study by Govannn and Pga (1994) for Italy, and the studes for European countres by Favero, Gavazz and Spaventa (1997) and Codogno, Favero and Mssale (2003). Also, Ardagna, Casell and Lane (2004) do not fnd a sgnfcant effect of the government debt to GDP rato or the prmary defct to GDP rato on the 10-year government bond yeld over a swap contract wth the same maturty and currency of denomnaton (a dependent varable that allows for government default rsk but excludes currency rsk). However, ther fscal polcy varables have a sgnfcant effect on nterest rate spreads relatve to German rates. Ardagna, Casell and Lane (2004, 13) note that These results may suggest that the mpact of fscal polcy on nterest rates s not lkely to be va default rsk drectly, but could be through expected nflaton. 16

18 magntude would lead to a rse n provncal bond yelds of 142 bass ponts (equvalent to 17 percent of the average provncal bond yeld n the sample for 1995). Smlarly, a declne n the federal government debt to GDP rato, equvalent to the fall from.583 n 1995 to.346 n 2003, s predcted to cause provncal bond yelds to fall by 102 bass ponts. For provnces wth a consderable burden of outstandng debt, a porton of whch must be rolled over every year, a yeld declne of 102 bass ponts could represent sgnfcant savngs (almost $2 bllon per year f the yeld declne s appled to total provncal net debt). Whle the beneft to the provnces from a reducton n federal debt, n terms of lower debt servcng costs, s consderable, back-of-the-envelope calculatons ndcate that the provnces would prefer a drect $1 transfer from the federal government to a $1 cut n the federal debt. In 2003, a $10 bllon declne n federal government debt would have reduced the federal debt to GDP rato by.86 percentage ponts, a declne whch, gven the estmates n Table 1, s predcted to cause a fall n provncal bond yelds of 3.7 bass ponts. If ths declne appled to all outstandng provncal government net debt, provncal debt nterest payments would fall by approxmately $70 mllon per year. Wth a dscount rate of 5 percent, ths savng would be equvalent to $1.4 bllon n present value terms. Thus, the value of provncal government nterest savngs from a reducton n the federal debt s much smaller than the magntude of the debt reducton that would be requred to realze these savngs. 24 The sze of the effect of the federal government debt to GDP rato on provncal government bond yelds found here s of smlar magntude to several estmates of the mpact of own-debt on own-government bond yelds found n studes for other countres (none of whch consder spllover effects). For example, Engen and Hubbard (2005) fnd approxmately a 3 bass pont rse n long 24 From the perspectve of the provncal governments alone, ths calculaton may overestmate the dfference between the cost and beneft of a reducton n federal debt. Cash transfers to the provnces made up only about 23 percent of federal program spendng (20 percent of all federal spendng) n If a federal debt reducton of $10 bllon s mplemented through proportonal cuts to all types of federal program spendng, the drect cost to the provnces would be only $2.3 bllon, whch s much closer to the estmated $1.4 bllon gan that would result from the reducton n provncal debt servcng costs. 17

19 term nterest rates n the US followng a 1 percentage pont ncrease n the federal government debt to GDP rato, whle Laubach (2003) estmates the mpact to be 4 to 5 bass ponts. Chnn and Frankel (2004, 20-21) obtan a smlar estmate for a sample of European countres. The estmates of the coeffcents assocated wth the federal government debt to GDP rato are robust, n terms of magntude and sgnfcance, to a number of varatons n the model s specfcaton. Whle not reported to conserve space, the followng alternatve specfcatons were consdered. Frst, the model was augmented by the addton of a varable representng the rato of provncal government foregn currency debt to total outstandng debt, snce the market may vew foregn currency debt dfferently than domestc debt. 25 Second, to capture possble nonlneartes n the model, equatons (7) and (8) were re-estmated wth varous combnatons of the three debt varables (FEDDEBT, PROVDEBT, OWNDEBT ) squared and nteracted wth one another. 26 Thrd, snce t s possble that negatve levels of debt have a dfferent mpact on yelds than postve debt, the model was re-estmated usng only those observatons for whch the OWNDEBT varable was postve. Fourth, the two provncal debt varables (OWNDEBT and PROVDEBT) were redefned to nclude the guaranteed labltes of provncal governments. 27 Fnally, the aggregate PROVDEBT varable was modfed to exclude the debt of provnce. 28 None of these alternatve specfcatons lead to mportant changes n the results n terms of coeffcent sgnfcance or the magntude of the margnal effects. 25 In ths case, the estmaton perod s , due to the more lmted avalablty of the foregn currency debt data. One reason debt ssued n foregn currency mght not be a sgnfcant determnant of domestc currency yelds s that, as noted by Wooldrdge (1996, 25), two of the larger provnces, Ontaro and Brtsh Columba, swap vrtually all of ther foregn currency borrowngs nto Canadan dollars and, whle Alberta held substantal amounts of unhedged U.S. dollar labltes, revenues from sales of ol and gas to the Unted States serve as a natural hedge aganst fluctuatons n the exchange rate. 26 When OWNDEBT was squared or nteracted wth FEDDEBT or PROVDEBT only observatons for postve values of OWNDEBT were used, whch reduced the sample sze to 188 observatons. 27 Statstcs Canada does not provde data on the guaranteed labltes of the federal government. 28 Note that, snce both the orgnal specfcaton and ths modfed specfcaton nclude provnce s debt to GDP rato as a separate varable, ths modfcaton mples only a very small change to the specfcaton. 18

20 6. Concludng Comments The results presented above provde evdence of a spllover effect on subcentral government bond yelds that arses from central government debt accumulaton n the Canadan federaton. Specfcally, a one percentage pont rse n the Canadan federal government debt to GDP rato leads to a statstcally sgnfcant ncrease n provncal government bond yelds of approxmately 4.3 bass ponts. Of ths ncrease, approxmately 3.0 bass ponts are due to a rse n the expected rate of currency deprecaton and 1.3 bass ponts are due to an ncrease n the probablty of provncal government default. The sgnfcant postve effect of the federal debt on provncal bond yelds s consstent wth the vew that, as the federal government ncreases ts debt, t s n a weaker poston to provde assstance to the provnces and, n addton, s more lkely to monetze ts debt and precptate a currency deprecaton. The estmated spllover effects are not only statstcally sgnfcant, but may also be mportant n terms of magntude as the Canadan federal government s debt to GDP rato ncreased by almost 35 percentage ponts through the 1980s and early 1990s. The estmates appear reasonable as they explan a large proporton of the varaton n provncal government bond yelds, pass standard specfcaton tests, and the estmated federal government spllover coeffcent s robust to several generalzatons of the estmatng equatons. Whle the analyss fnds a sgnfcant effect of central government debt on provncal government bond yelds, the spllover effect assocated wth debt accumulaton by the provnces s not statstcally sgnfcant. One reason for ths may be that an mportant channel for potental spllover effects, monetary accommodaton, s avalable only to the federal government n Canada. The nsgnfcant coeffcent assocated wth aggregate provncal government debt also suggests that the other provnces are not vewed as a potental source of balout funds for a provnce n fnancal dffculty. An mplcaton of the absence of evdence of provncal government debt spllovers s that spllovers cannot be used to ratonalze the mposton of lmts on provncal government debt accumulaton. 19

21 One polcy mplcaton of the fndng that federal government debt has a sgnfcant postve mpact on provncal government bond yelds s that the mposton of restrctons on central government borrowng may be n the nterest of subcentral governments. To our knowledge, the welfare effects of restrctons on central government debt accumulaton n a federaton have not been explored, although a number of studes have suggested the need for lmts on subcentral government debt accumulaton. It s perhaps noteworthy that the European Unon mposes a strct celng on European Unon expendtures (European Commsson, 2005). The results presented above for Canada show that much of the spllover effect on yelds from government debt accumulaton comes through a rse n the expected rate of currency deprecaton, rather than through an ncrease n the rsk of outrght default. Ths mples that a major source of debt spllovers s the perceved rsk of monetary accommodaton, as ths s lkely to be the chef cause of changes n the expected rate of currency deprecaton. In a monetary unon, f the market beleves that the central bank s ndependent and wll not be pressured by governments wth large debts to nflate, yelds wll be less lkely to rse wth the sze of government debt. Another mplcaton of the fndngs of ths study s that the control of monetary polcy n a currency unon s lkely to be mportant to the magntude of debt spllovers. Ths result may be of nterest to countres or regons that are consderng the formaton of a monetary unon, partcularly f preferences for ant-nflatonary polces dffer across potental members, or f the proposed unon s central bank s lkely to be subject to regonal poltcal pressures, as n the model of Debrun, Masson and Pattllo (2005) Ths s n contrast to a case where monetary polcy control s delegated to a supranatonal central bank, say through an external guarantee of a peg to a hard currency, as dscussed n Alesna, Angelon and Etro (2001). 20

22 Table 1: Government Debt Spllover Effects ρ equaton ε equaton Explanatory Varables (Equaton (5)) (Equaton (6)) Spllover Varables: Z FEDDEBT.0126*.0305* (3.42) (5.62) PROVDEBT (1.54) (1.32) Non-Spllover Varables: ρ ε X, W and W OWNDEBT.0065 (1.11) Y * (2.47) TRANSFERS.0237* (3.80) r BankR r FFR.0753*.4452* (4.97) (17.35) π π Can π π US Can US 1 1 Can Y.1168* (3.55).0392 (0.94) (1.71) MATURITY.0013* * (2.93) (4.89) Notes: Estmates of the ndvdual bond fxed effects are not reported to conserve space. There are 203 observatons. The numbers n brackets under each coeffcent estmate are the absolute values of the heteroskedastcty corrected t- statstcs. * The estmated coeffcent s sgnfcant usng a 95 percent confdence nterval. Equaton (7) Equaton (8) R AR1 test (t-statstc) 1.63 a 1.07 a Reset Test (t-statstc) 1.83 b.66 b AR1 Test methodology: Estmate equatons (7) and (8) separately usng OLS. Regress the resduals from each of these equatons on the lagged resduals and the explanatory varables of the model. The test statstc s the t- statstc on the lagged resdual. (See Davdson and MacKnnon (1993, 358).) Reset Test methodology: Estmate equatons (7) and (8) jontly. Add the squared predcted values for each dependent 21

23 varable to the model as explanatory varables and re-estmate. Test f ether squared predcted value s ndvdually sgnfcant (t-statstc). a Cannot reject the hypothess of no seral correlaton at 5 percent. b Cannot reject the hypothess of no msspecfcaton at 5 percent. 22