EXCHANGE RATE REGIMES AND THE EXPECTATIONS HYPOTHESIS OF THE TERM STRUCTURE

Transcription

1 BIS WORKING PAPERS No. 43 EXCHANGE RATE REGIMES AND THE EXPECTATIONS HYPOTHESIS OF THE TERM STRUCTURE by Sefan Gerlach and Frank Smes July 1997 BANK FOR INTERNATIONAL SETTLEMENTS Moneary and Economic Deparmen BASLE

2 BIS Working Papers are wrien by members of he Moneary and Economic Deparmen of he Bank for Inernaional Selemens, and from ime o ime by ouside economiss, and are published by he Bank. The papers are on subjecs of opical ineres and are echnical in characer. The views expressed in hem are hose of heir auhors and no necessarily he views of he BIS. Bank for Inernaional Selemens 1997 CH-4 Basle, Swizerland Also available on he BIS World Wide Web sie (hp:// All righs reserved. Brief excerps may be reproduced or ranslaed provided he source is saed. ISSN 1-959

3 EXCHANGE RATE REGIMES AND THE EXPECTATIONS HYPOTHESIS OF THE TERM STRUCTURE * by Sefan Gerlach and Frank Smes July 1997 Absrac This paper uses weekly daa on shor-erm euroraes for en counries for he period o documen ha he abiliy of he expecaions hypohesis (EH) o accoun for movemens in he erm srucure is greaer, and ha shor-erm ineres raes are more predicable, under fixed han under floaing exchange raes. The paper also shows ha he higher predicabiliy does no arise solely because of moneary policy responses o speculaive pressures in he foreign exchange markes: while i is more difficul o rejec he EH in periods of exchange marke urmoil, he EH is no rejeced in ranquil periods. Key words: Term srucure of ineres raes, expecaions hypohesis, speculaive aacks. JEL classificaion: E4. * The views expressed in his paper are solely our own and do no necessarily reflec hose of he BIS. This paper was wrien in response o he commens we received on he paper we presened a he conference (Gerlach and Smes, 1997). We are graeful o our discussan, Gikas Hardouvelis and conference paricipans for discussions, and o Ignazio Angeloni, Gabriele Galai, Greg Suon and Kosas Tsasaronis for commens on his and our earlier paper. Florence Béranger provided excellen research assisance.

4

5 Conens Inroducion Theory Economeric resuls Full-sample resuls The EH, exchange rae regimes and he predicabiliy of fuure shor raes The EH and exchange marke urmoil Change in ineres rae spreads Oulier analysis Realignmen daes Conclusions References... 19

6

7 Inroducion The erm srucure of ineres raes plays an imporan role in much of modern financial economics and in he design of moneary policy. Unforunaely, he sandard conclusion in he exising empirical lieraure on he erm srucure is ha curren heoreical models do no explain movemens in shor and long ineres raes. Shiller (199, p. 7), for insance, in surveying he erm srucure argues ha: "Empirical work on he erm srucure has produced consensus on lile more han ha he raional expecaions model... can be rejeced." More recenly, however, several auhors have noed ha he explanaory power of he expecaions hypohesis (EH) of he erm srucure - which saes ha longer ineres raes are averages of expeced fuure shor ineres raes plus, perhaps, a consan erm premium - ends o be greaer ouside he Unied Saes. For example, Hardouvelis (1994), sudying he behaviour of hree-monh and en-year raes in he Group of Seven (G-7) counries, finds ha he EH appears o do a paricularly poor job of accouning for he behaviour of ineres raes in he Unied Saes, bu ha i works quie well in oher counries. 1 Furher evidence suggesing ha i is more difficul o rejec he EH using non-us daa is provided by Gerlach and Smes (1997), who use 1, 3, and -monh euroraes in 17 counries o show ha, while hey are unable o rejec he EH in a majoriy of cases, for he Unied Saes he EH can be rejeced for all mauriies. Kugler (1988, 199) also finds ha he EH does a much beer job of explaining he shor end of he erm srucure of ineres raes in Swizerland and Germany han in he Unied Saes. One possible reason why i is more difficul o rejec he EH ouside he Unied Saes is ha shor-erm ineres raes may be easier o predic in oher counries. Mankiw and Miron (198) show ha in he presence of a ime-varying erm premium, differences in he predicabiliy of shor-erm ineres raes can have a large impac on ess of he EH. Gerlach and Smes (1997) use a simple forecasing model o esimae he predicabiliy of shor-erm raes for he 17 currencies hey sudy, and demonsrae ha ess of he EH fare beer in counries in which shor-erm raes are more forecasable, as hypohesised by Mankiw and Miron (198). This raises he quesion of why he predicabiliy varies beween counries. The poin of deparure for his paper is he suggesion in Gerlach and Smes (1997) ha he higher predicabiliy of shor-erm raes in some European counries may be due o he fac ha hese counries operaed wih adjusable peg regimes during a large par of he sample periods used 1 In relaed work, Gerlach (199) sudies shor and long ineres raes in 14 counries, and finds large differences beween counries in he abiliy of he expecaions hypohesis o accoun for movemens in he erm srucure. 1

8 and ha hey experienced episodes of speculaive pressures in he foreign exchange marke. Since cenral banks ypically respond o such pressures by raising shor-erm ineres raes o very high levels for a brief period, afer which hey reduce hem o more normal levels, he combinaion of such pressures and sysemaic policy responses renders shor-erm raes relaively predicable. Tha speculaive aacks could be imporan for ess of he EH is suggesed by he work of Jondeau and Ricar (199), who, using daa on French franc ineres raes, show ha ess of he EH are sensiive o he inclusion of a few observaions ha are relaed o episodes of exchange marke urmoil. While he EH is acceped for he pairs of mauriies hey sudy when he whole sample is used, only for five of hese do hey accep he EH when observaions associaed wih speculaive aacks in 1981 and 1983 are removed. Considerable insabiliy of erm srucure regressions is also documened by Dahlquis and Jonsson (1995), who sudy he shor end of he erm srucure of Swedish Treasury bills. In paricular, i appears ha he expecaions hypohesis can be rejeced for periods in which foreign exchange markes were calm. The purpose of his paper is wofold: firs, o see wheher he EH is rejeced less frequenly in counries in which cenral banks conduc policy using inermediae exchange rae arges han in counries where no such arges are employed; and secondly, if his is he case, o assess wheher his is due o occasional episodes of exchange marke urmoil. To do so, we use weekly daa on 1, 3, and -monh euro-ineres raes from en counries. So as o clearly disinguish beween fixed and flexible exchange rae regimes he daa sar afer he esablishmen of he ERM in 1979 and end in In order o assess he imporance of he exchange rae regime, we informally disinguish beween counries operaing wih and wihou inermediae exchange rae arges. The group of counries wihou inermediae exchange rae objecives consiss of he Unied Saes, Japan, Germany, Canada and he Unied Kingdom. The comparison group of counries wih inermediae exchange rae arges comprises Belgium, France, Ialy, he Neherlands and Sweden. This selecion of counries warrans several commens. Alhough Germany has been an ERM member since he sysem was founded in 1979, i has no faced he same exchange rae consrain as he oher members by virue of he fac ha i has been he de faco anchor counry. Moreover, some counries changed his exchange rae regime during he esimaion period. The Unied Kingdom was a member of he ERM beween Ocober 199 and Sepember 199. Ialy belonged o he ERM unil Sepember 199. While Sweden has never paricipaed in he ERM, beween 1977 and 1991 Swedish moneary policy was geared o an inermediae exchange rae expressed in erms of a rade-weighed baske in which he core ERM currencies played an imporan role, and in a unilaeral ECU peg was used. Since Kugler (1988, 199) argues ha shor-erm ineres raes have been more predicable in Swizerland and Germany because he auhoriies have conduced moneary policy using money sock argeing, while he Federal Reserve has ended o smooh ineres raes. 3 Gerlach and Smes (1997) used all available daa o es he expecaions hypohesis, and hus disregarded he fac ha he exchange rae regime may have changed in he esimaion period.

9 November 199, however, he krona has been floaing. Despie hese changes in he operaing framework in he Unied Kingdom, Ialy and Sweden, we use he same sample period for all counries since preliminary empirical work indicaes ha shorening he sample period o for he Unied Kingdom and for Ialy and Sweden has virually no effec on he esimaed parameers. To disinguish beween periods of ranquilliy and urbulence in he foreign exchange marke we use hree differen crieria. The firs of hese is based on he behaviour of he spread of domesic ineres raes over foreign ineres raes. Speculaive pressures on he domesic currency are ypically associaed wih sharp bu emporary increases in he spread beween domesic and foreign ineres raes. We hus define a speculaive aack episode as he week in which he spread beween domesic and foreign ineres raes increases subsanially plus he immediaely following weeks. The second crierion is purely saisical: we drop observaions for which he residual is more han wo sandard deviaions large in absolue value. Finally, we define as urbulen episodes he periods around ERM realignmens involving he French franc and Ialian lira. 4 Overall, he resuls are no very sensiive o he crierion we use o divide he sample ino calm and urbulen periods. In general, we find ha he EH fis he daa beer in periods of speculaive pressures in he exchange marke. However, we are ypically no able o rejec he EH in he calmer periods. Thus, we do no believe ha speculaive aacks are he sole reason why i ends o be more difficul o rejec he EH in daa for counries where moneary policy is conduced using inermediae exchange rae objecives. The remainder of his paper is organised as follows. In Secion 1 we use he expecaions heory o derive he equaion ha we esimae o es he EH. The heory implies ha he realised fuure excess reurn on rolling over a one-period invesmen over j periods should be relaed on a oneo-one basis o he curren spread beween j-period and one-period ineres raes. Secion repors he esimaes of he equaion, and shows ha he EH does indeed fare beer in counries conducing moneary policy using fixed exchange raes. Secion 3 goes on o see wheher his is due o he occurrence of occasional speculaive aacks, bu finds ha his is no he case. Finally, he las secion conains some conclusions. 1. Theory In order o es he EH we follow Hardouvelis (1994) and Gerlach and Smes (1997), among ohers, and perform muliperiod regressions. The heoreical underpinning for hese regressions is sraighforward. Le R denoe he j-monh ineres rae a ime, θ a possible non-zero bu 4 The reason for focusing on France and Ialy is ha, among he counries sudied here, hey have been involved in a relaively large number of pariy changes. 3

10 consan j-monh erm premium and E r + k he expeced value of he one-monh ineres rae a ime +k, formed on he basis of informaion available a ime. All ineres raes are expressed a annualised raes. In wha follows, we use 3, and -monh raes, so ha j = 3, and. In he economeric work below we use weekly daa. Thus, a j-monh eurodeposi will maure a ime +4j, assuming, for noaional simpliciy, ha here are exacly four weeks per monh. 5 The expecaions heory posis ha he reurn on he j-monh invesmen should equal he reurn on a one-monh invesmen, rolled over j imes, plus he erm premium: j 1 j (1) ( 1+ R ) = θ + ( 1+ Er+ 4i/ j). i= Linearising, we have (approximaely) () R + ( E r + / j) or j 1 θ 4i i= (3) ( Er+ 4( 1) + Er+ 4( ) r)/ j θ + R. j j Equaion (3) spells ou one implicaion of he EH ha is he focus of he presen paper: he expeced reurn on rolling over he one-monh posiion (on he LHS) should equal he reurn on holding he j-monh invesmen unil mauriy, adjused for he erm premium (on he RHS). This implicaion of he EH can be esed by esimaing (4) rs = α+ β( R r ) + v j 1 4 i= where rs ( r+ i r)/ j. Under he EH, we have α = θ, v ( r + 4i Er + 4i )/ j, and β =1. Thus, he erm spread should predic a weighed average of he fuure change in one-monh ineres raes during he holding period. In wha follows we refer o he LHS of equaion (4) as he rollover spread, rs. 7 Noe ha alhough we, as ohers, assume ha expecaion errors, r E r j 1 i= + i + i, are 5 In esimaion, we assume ha here are 13, and 5 weeks when j equals 3, and. See also foonoe 8. Anoher esable implicaion of he EH is ha an upward-sloping yield curve should predic rising long ineres raes (see Hardouvelis (1994)). This implicaion, however, is ypically no esed in research focusing on he shor end of he yield curve. 7 In wha follows, when we discuss he predicabiliy of shor-erm ineres raes, we mean, sricly speaking, he predicabiliy of he rollover spread. 4

11 uncorrelaed over ime, v is no whie noise. If he observaions are weekly, he errors obey an MA(4j-1) process. 8. Economeric resuls Nex we esimae equaion (4) using weekly euromarke daa for en currencies from 14h March 1979 o 7h Augus The sar of he esimaion period reflecs he adopion of he exchange rae mechanism (ERM) wihin he EMS. We use euroraes for wo reasons. Firs, euroraes are less affeced han onshore raes by capial conrols, ax consideraions or legal regulaions, which could drive observed raes away from equilibrium levels. Secondly, he raes for differen currencies are direcly comparable and do no depend on facors such as defaul risk, differences in duraion, he calculaion of yields, ec..1 Full-sample resuls Before urning o he economeric resuls, i is informaive o consider Figure 1, which conains scaer plos of he rollover spread, rs, agains he erm spread, (R r ). To enhance comparabiliy, he scale on boh axes is ± 1% for all scaer plos. 1 For space reasons we only consider he case of j =. To see wha he EH implies for he scaer plos, assume ha here are no expecaion errors and ha he erm premium is consan. If ha is he case, he daa poins should lie on a line ha has a slope of uniy and cus he verical axis a θ. Expecaion errors would cause he daa poins o scaer around his line, bu should no lead us o rejec he hypohesis of a slope of uniy. As discussed below, a ime-varying erm premium could bias he esimaed slope of he regression line away from uniy. Consider firs he scaer plo for Germany. The daa poins all fall relaively close o he origin, and here is lile visual evidence ha a fied line would have a slope of uniy. However, inspecion of he scaer plos for he oher counries suggess ha he daa for Germany are somewha unusual, as he oher counries show much more variaion in boh he erm and rollover spreads. Furhermore, and more imporanly, in France, Ialy and Sweden a considerable fracion of he daa 8 We herefore follow Newey and Wes (1987) and use sandard errors ha are heeroscedasiciy-consisen and robus o MA errors. Since here are no exacly four weeks per monh, we allow for MA errors of order /5/51 when he long rae is 3// monhs. 9 The daa sem from he BIS daabase, and are bid raes recorded a abou 1 a.m. on he Wednesday of each week. 1 For France, Ialy and he Unied Kingdom, in a number of cases eiher he erm spread or he rollover spread is larger han 1% in absolue value. These observaions have been dropped from he scaer plos. In Secion 3. below we explore wheher ess of he EH are sensiive o he presence of ouliers. 5

12 Figure 1 Rollover spreads versus erm spreads (j = monhs) BELGIUM Term spread CANADA Term spread FRANCE Term spread GERMANY Term spread ITALY Term spread JAPAN Term spread NETHERLANDS Term spread SWEDEN Term spread UNITED KINGDOM Term spread UNITED STATES Term spread

13 poins ends o fall in he lower lef-hand and he upper righ-hand quadrans, leading o a visual impression of a 45 degree line. These daa poins are likely o reflec exchange marke urmoil. To see his, suppose ha he domesic exchange rae is subjec o pressures in he foreign exchange marke. To sem he ouflows, he cenral bank raises domesic shor-erm ineres raes. Since marke paricipans realise ha his measure is likely o be emporary, longer-erm ineres raes rise by less han shor-erm raes, ha is, R r <. Since he cenral bank will reduce shor-erm ineres raes afer some ime - eiher because he coss of defending he currency grow oo high or because he exchange marke pressures subside - fuure shor-erm ineres raes will be falling, ha is, Table 1 Full-sample esimaes Counries Mauriy in monhs (j) Observaions 1 Parameer esimaes (p-values for es of α= and β=, in %) p-values for es of b = 1 (in %) a b Floaing exchange raes Canada 3 Germany 3 Japan 3 Unied Kingdom 3 Unied Saes (1.1) -. (13.5) -.33 (7.9) -. (.9) -.11 (13.4) -.18 (35.) -.7 (.) -.9 (14.8) -.14 (38.) -.5 (13.9) -. (57.) -.11 (.3) -.9 (5.9) -. (3.8) -.3 (.).8.7 (.1) (.1) (1.).4 (1.).55 (.3)

14 Table 1 (con.) Counries Mauriy in monhs (j) Observaions 1 Parameer esimaes (p-values for es of α= and β=, in %) p-values for es of b = 1 (in %) a b Pegged exchange raes Belgium 3 France 3 Ialy 3 Neherlands 3 Sweden (5.1) -.14 (18.8) -.1 (9.8) -.18 (8.5) -.3 (9.) -.3 (8.) -.5 (1.) -.4 (.5) -.54 (3.) -.8 (.4) -.15 (.7) -.9 (7.4) -.7 (3.3).1 (9.4).1 (3.4).84 (.) Esimaion periods are 1979:3:-199:7:15 for he hree-monh mauriy, 1979:3:-199:4:15 for he six-monh mauriy and 1979:3:-1995:1:1 for he -monh mauriy. Newey and Wes (1987) sandard errors, which are heeroscedasiciy-consisen and robus o MA errors of order, 5, or 51 depending on he mauriy of he long rae, j. rs ( r r )/ j <. Thus, speculaive aacks will generae observaions in he lower lef-hand j + 4( i 1) i= 1 quadran in he scaer plos in Figure 1. Table 1 provides esimaes of he parameers in equaion (4) ogeher wih he esimaed sandard errors and he p-values for he hypoheses ha β = and β =1. Noe, firs, ha in all cases b is esimaed o be posiive and significanly differen from zero a he 5% confidence level (excep he case in which j = 3 for he Unied Saes, for which he p-value is 1%). Thus, here is evidence ha 8

15 he erm spread conains informaion abou he fuure pah of one-monh ineres raes. 11 The esimaes vary considerably across counries however. The smalles esimaes are obained for he Unied Saes (.45,.4 and.55 for j = 3, and ), followed by Germany, he Unied Kingdom and Canada. For his group of counries we can rejec he hypohesis ha β =1 a he 1% confidence level for a leas one mauriy. The larges esimaes are obained for France (1.7, 1.4 and.97), Ialy (.9, 1.7 and 1.13) and Sweden (1.4, 1.1 and 1.3). While he esimaes for Japan for he case j = 3 are moderae and he p-value for a es of a parameer of uniy is abou 7%, he esimaes for j = and are high (1.4 and 1.4). The resuls in Table 1 sugges wo conclusions. Firs, as discussed in Gerlach and Smes (1997), i is more difficul o rejec he EH in counries ha have conduced moneary policy using fixed exchange raes han in counries in which policy is conduced under floaing raes. Indeed, a simple OLS regression of all he β esimaes (repored in Table 1) on a consan and a dummy variable equal o one if he counry operaes under fixed exchange raes and zero oherwise confirms ha he average β esimae of.99 in he fixed exchange rae counries is significanly higher han he average esimae of.74 in he floaing exchange rae counries (see regression 1 in Table 3). Secondly, while he esimaes of b are high in counries which have been exposed o a number of speculaive aacks, he esimaes for Belgium and he Neherlands, where speculaive aacks have been less frequen, are also close o uniy. This suggess ha he occurrence of speculaive aacks may be less imporan han he choice of exchange rae regime.. The EH, exchange rae regimes and he predicabiliy of fuure shor raes In his secion we explore wheher he fac ha i is more difficul o rejec he EH in counries ha conduc moneary policy wih inermediae exchange rae arges can be explained by a higher predicabiliy of shor-erm ineres raes in hose counries. To do so, we follow Fama (1984), Mankiw and Miron (198) and many ohers in noing ha variaions in he erm premium in violaion of he EH - can bias he coefficien on he spread in equaion (4). As a firs sep, recall from equaion (4) ha R r = E rs +θ. If he correlaion beween he erm premium θ and he expeced j rollover spread, E rs (), is denoed by r, he plim β $ is given by: (5) plim $ σ ( Ers ) + ρσ( Ers ) σ( θ ) β= σ ( Ers ) + σ ( θ ) + ρσ( Ers ) σ( θ ) where σ () denoes variance and σ( ) sandard deviaion. From equaion (5) i is clear ha he esimae of b converges o uniy when he variance of he erm premium is zero. However, if here is a ime-varying erm premium, he esimae will be biased and he size of he bias will depend, iner alia, 11 The fac ha he slope parameer is significan in all regressions is somewha surprising in he ligh of he exising lieraure on he EH. However, since we use weekly daa, here are abou 9 observaions in our sample. 9

16 on he variance of he expeced rollover spread. As his variance goes o zero, he coefficien in (5) converges o zero (assuming ha σ ( θ ) > ). In order o clarify he imporance of he predicabiliy of changes in he shor-erm ineres rae for he esimae of he slope parameer, we divide he numeraor and denominaor of he rs righ-hand side of equaion (5) by σ ( ), resuling in: () β $ R + ρrθ = R + ρrθ+ Θ, where R is he raio of σ () ( Ers ) o σ ( ), and Θ is he raio of σ ( θ ) o σ ( ). The firs of j rs hese raios can be inerpreed as a measure of he predicabiliy of he rollover spread, and he second as a measure of he imporance of he erm spread. To assess wheher differences in he predicabiliy of shor-erm raes accoun for he cross-counry differences in he esimaes of β i, we follow Mankiw and Miron (198) and Gerlach and Smes (1997) and esimae a simple univariae forecasing equaion comprising he curren and lagged shor and longer-erm raes as explanaory variables. More precisely, we regress he rollover spread for each of he mauriies on curren and hree lags of he shor-erm ineres rae and he relevan erm spread, ha is rs e j. 3 1i i i i i i= (7) rs = γ + γ r + γ ( R r ) + ε As suggesed above, he R from his regression, presened in Table, is a naural measure of he degree of predicabiliy of changes in he shor-erm ineres rae. I is noeworhy ha he R :s are smalles in he Unied Saes, for which we obain he lowes b esimaes. Furhermore, he R :s are larges in France, Ialy and Sweden, counries for which we obain b esimaes close o uniy. Following Gerlach and Smes (1997), i is possible o invesigae more formally wheher here is a relaionship beween he esimaes of β and he predicabiliy of he shor rae in he counry in quesion. Leing α = 1 ρ Θ, α =Θ, and adding an error erm allowing for esimaion error in he β esimaes, we obain from equaion (): (8) $β i Ri + α1ri = + ξ i Ri + α1ri + α, where he subscrip i refers o he i:h esimae of he slope parameer. Since we have daa on hree erm spreads for en counries, i = 1,...,3. Noe ha he bias also depends on r and ha for sufficienly negaive values of r he β esimae could be biased above one. See Mankiw and Miron (198, p. 19). 1

17 Table Predicabiliy of he rollover spread Mauriy in monhs (j) Counry 3 Floaing exchange raes Canada Germany Japan Unied Kingdom...4 Unied Saes Pegged exchange raes Belgium France Ialy Neherlands Sweden Noe: The predicabiliy of he rollover spread is given by he R² from equaion (7). The sample periods are he same as in Table 1. Equaion (8) involves he β i :s esimaed in Secion.1 (abulaed in Table 1), and he R i :s in he regressions used o calculae he variance of he expeced rollover spread (provided in Table ). Thus, by fiing his equaion we can use he resuling $α 1 and $α o compue $ρ. Of course, in inerpreing hese esimaes, i should be recalled ha he underlying assumpion ha Θ and ρ are consan across counries and mauriies may no be rue. Furhermore, since boh he dependen and independen variables in his regression sem from firs-sep regressions and may be subjec o esimaion error, our esimaes of equaion (8) are subjec o generaed regressor bias of unknown magniude. For hese reasons, he empirical resuls discussed below mus be reaed wih cauion. Table 3 repors esimaes of equaion (8) using non-linear leas squares (regression ). The resuls are quie sriking: more han 8% of he cross-counry and cross-mauriy variaion in β esimaes can be explained by differences in he predicabiliy of he shor rae. The implied esimaes of he correlaion beween he erm premium and he expeced rollover spread is -.51, while he esimae of he raio of he variance of he erm premium o he variance of he rollover spread is.11. These resuls are very similar o hose repored in Gerlach and Smes (1997), who use monhly daa from 17 counries and differen sample periods. The resuls from he regression are illusraed in Figure, which conains a scaer plo of he 3 esimaes of β i agains he predicabiliy of he rollover spread, ogeher wih our esimaes of he non-linear relaionship. Regression 3 in Table 3 also shows ha once one conrols for differences in predicabiliy, he dummy variable which capures he exchange rae regime is no longer significan. 11

18 ^ βi i Table 3 Esimaes of: Ri + α1ri = R + α R + α 1 i + α D + α + ξ 3 i 4 Regression a 1 a a 3 a 4 Adjused R (4.59) (19.53) (-1.45) (15.4) (-3.47) (3.97) (.99) (.83) Noe: D is a dummy variable which is one for counries conducing moneary policy wih inermediae exchange rae arges. In regression 1 only he dummy and he consan are included among he regressors; in regression only he firs erm of he equaion is included. The -saisics are in parenheses. i This confirms ha he higher β esimaes in counries wih pegged exchange raes can be explained by he higher predicabiliy of shor-erm ineres raes in hese counries. Indeed, Figure shows ha, wih he excepion of Japan, he R :s for counries wih floaing exchange raes are almos uniformly lower han hose for counries wih fixed exchange raes. We view Table 3 and Figure as presening compelling evidence ha inernaional differences in he predicabiliy of shor-erm ineres raes, coupled wih ime-varying risk premia, accoun for differences beween counries in he esimaes of β i. Furhermore, he inabiliy o rejec he EH by his es in counries wih pegged exchange raes is due o he higher predicabiliy of shor-erm raes in hese counries. Figure Slope parameers versus he predicabiliy of he shor rae 1.4 Two observaions (one fixed, one floaing) Floaing rae counries Fixed rae counries Esimaed Predicabiliy of shor rae

19 The laer finding raises he quesion of wheher his higher predicabiliy is aribuable o occasional episodes of speculaive aacks in he foreign exchange marke. In he nex secion we invesigae his issue. 3. The EH and exchange marke urmoil In his secion we examine he sabiliy of he resuls repored in he secion above by considering wo subsamples corresponding o periods of exchange marke urmoil and periods of ranquilliy. The goal of he analysis is o examine wheher he reason why he EH works beer in counries operaing under fixed exchange raes is o be found in occasional periods of speculaive pressures in he foreign exchange marke. To his end, we esimae he following modified version of equaion (4): NA NA SA SA (9) rs = ( 1 D ) α + β ( R r ) + D α + β ( R r ) + v o o, where D is a dummy variable which akes on he value of one in periods of exchange marke urmoil, SA denoes "speculaive aack" and "NA" denoes "no aack". Noe ha i seems plausible ha he variance of he error erm is larger during speculaive aack episodes. However, since we calculae heeroscedasiciy-consisen sandard errors, we do no adjus for his. I is no self-eviden how o discriminae beween speculaive aack and calmer periods, ha is, how o define D. Since he resuls may depend on he exac way in which his is done, we use hree differen crieria - one economic, one saisical and one insiuional - which are described in he hree subsecions below. For each of hese crieria, Tables 4 o give our esimaes of β NA and β SA, ogeher wih he p-values for ess of he hypoheses ha hey are equal o zero or one, and ha hey are equal o each oher. 3.1 Changes in ineres rae spreads The firs crierion we use o divide he sample in urbulen and calm periods is based on he fac ha periods of exchange marke urmoil are usually associaed wih large increases in domesic ineres raes relaive o hose in he reference counry, as invesors need o be compensaed for he expeced depreciaion of he domesic currency. Since speculaive aacks ypically las for several weeks, we define a period of exchange marke urmoil as he week in which he ineres rae spread increases by 1 basis poins, plus he following seven weeks. 13 We sress ha since his crierion has clear economic underpinnings, he resuls repored in his secion are more easily inerpreable han hose repored in he subsequen wo secions. 13 The resuls are no very differen if, for example, he bigger poin is changed from 1 basis poins o 5 or he duraion from eigh weeks o four. 13

20 We use German ineres raes as reference raes for each of he five counries wih pegged exchange raes. While, sricly speaking, no counry pegged is exchange rae o he DM in he sample period, Germany played a cenral role for he counries paricipaing in he ERM, as well as for Sweden. Table 4 Esimaes of equaion (5), using increases in spreads beween domesic and foreign one-monh ineres raes o idenify speculaive aacks Parameer esimaes 1 (p-values for es of β=, in %) p-values for hypohesis ess (in %) Counries Mauriy in monhs (j) Observaions NA/SA β NA β SA β NA = 1 β SA = 1 β NA = β SA β NA = β = 1 Belgium 3 France 3 Ialy 3 Neherlands 3 Sweden 3 84/1 791/1 75/1 77/ / / /187 7/187 8/187 89/37 85/37 83/37 74/14 751/14 75/ (.1).9 (1.).59 (3.).5 (1.) (1.) (.8) Noe: The dummy variable equals one in he week in which he spread beween domesic and foreign one-monh ineres raes rises by 11 basis poins. 1 Newey and Wes (1987) sandard errors, which are heeroscedasiciy-consisen and robus o MA errors of order, 5 or 51, depending on he mauriy of he long rae, j. Esimaion periods are 1979:3:-199:7:15 for he hree-monh mauriy, 1979:3:-199:4:15 for he six-monh mauriy and 1979:3:-1995:1:1 for he -monh mauriy. 14

21 In Table 4 we presen he resuls from he esimaion of equaion (9). Firs, wih he excepion of Belgium, he β esimaes are higher in periods of exchange marke urmoil han in calm periods. Leaving aside he case of Belgium, we find ha he average β esimae is.79 in calm periods - which is very close o he average β esimae of.74 we found above for he counries wih floaing exchange raes - and 1.17 in urbulen episodes. In half of hese cases we can rejec he hypohesis ha β NA SA = β. Furhermore, in half of he cases we can also rejec ha boh are equal o one. Somewha surprisingly, in he case of France we canno rejec hese wo hypoheses in five ou of six cases, in spie of he fac ha he differences in β esimaes are comparable o hose in some of he oher counries. Overall, his evidence suggess ha including episodes of exchange marke urmoil in he esimaion period generally leads o higher β esimaes and can hus make i more difficul o rejec he EH in counries which have undergone speculaive aacks. In essence, speculaive aacks give rise o large, bu emporary, increases in he one-monh ineres rae which reduce he relaive imporance of ime-varying risk premia. However, Table 4 also shows ha only in one of he 15 cases can we rejec he EH (β =1) in calm periods. This again underlines he general finding in Gerlach and Smes (1997) and in Table 1 of his paper ha i is surprisingly difficul o rejec he EH in counries oher han he Unied Saes. Moreover, hese resuls show ha his finding canno simply be explained by he fac ha mos European economies have operaed moneary policy under fixed exchange raes and have experienced occasional speculaive aacks. 3. Oulier analysis In order o assess wheher he resuls repored above depend on he crierion used o disinguish beween urbulen and calm periods in he foreign exchange marke, we nex re-esimae he regressions repored in Table 1 and es wheher he slope parameers are larger for observaions ha are associaed wih ouliers. Thus, we se D = 1 for hose observaions for which he residuals from he regressions repored in Table 1 are more han wo sandard deviaions in absolue value. The resuls repored in Table 5 are very similar o hose repored in Table 4, and do no warran much furher commen. Again we find ha in 14 of he 15 cases he β esimaes are higher in periods of exchange marke urmoil han in calm periods, alhough in many cases hese differences are no saisically significan. In addiion we find ha, as in Secion 3.1, dropping he ouliers does no invalidae our previous finding ha i is more difficul o rejec he EH in hese counries han in he Unied Saes, as evidenced by he fac ha we rejec he EH only in one ou of 15 cases. 3.3 Realignmen daes Finally, we use informaion abou ERM realignmens o idenify periods of exchange marke urmoil. In his secion we define as a speculaive aack episode he eigh weeks before and 15

22 Table 5 Esimaes of equaion (5), using ouliers o idenify speculaive aacks Counries Mauriy in monhs (j) Belgium 3 France 3 Ialy 3 Neherlands 3 Sweden 3 Observaions NA/SA 857/49 87/ 84/43 81/45 83/3 8/5 844/ 838/55 8/45 859/47 831/ 79/71 874/3 848/45 817/5 Parameer esimaes 1 (p-values for es of β=, in %) p-values for hypohesis ess (in %) β NA β SA β NA = 1 β SA = 1 β NA = β SA β NA = β = (.7) 1. (.1) (1.1) 1. (5.5) (1.) 1.7 (11.) 1.8 (3.) Noe: The dummy variable equals one for hose observaions in he regressions in Table 1 for which he residuals are larger han wo sandard deviaions in absolue value. 1 Newey and Wes (1987) sandard errors, which are heeroscedasiciy-consisen and robus o MA errors of order, 5 or 51, depending on he mauriy of he long rae. Esimaion periods are 1979:3:-199:7:15 for he hree-monh mauriy, 1979:3:-199:4:15 for he six-monh mauriy and 1979:3:-1995:1:1 for he -monh mauriy. wo weeks afer an ERM realignmen. 14 While his crierion suffers from he shorcoming ha i will no capure periods of exchange marke urmoil which did no give rise o a realignmen, i provides a useful way of esing he robusness of our resuls. 14 The realignmen daes are from Svensson (1993). We used realignmen daes from he BIS daabase o exend he sample period beyond Svensson's sample. The resuls are no very sensiive o alernaive assumpions regarding he duraion of he episode. 1

23 Table Esimaes of equaion (5), using dummies corresponding o ERM evens o idenify speculaive aacks Parameer esimaes 1 (p-values for es of β=, in %) p-values for hypohesis ess (in %) Counries Mauriy in monhs (j) Observaions NA/SA β NA β SA β NA = 1 β SA = 1 β NA = β SA β NA = β = 1 French and Ialian ERM evens 3 France 3 Ialy 3 84/1 791/1 75/1 73/17 73/17 97/17.84 (.9).74 (.1) France 3 Ialy 3 31/75 18/75 59/75 31/75 18/75 59/75.8 (7.).49 (5.4).5 (.).5 (.5).79.9 General ERM evens Newey and Wes (1987) sandard errors, which are heeroscedasiciy-consisen and robus o MA errors of order, 5 or 51, depending on he mauriy of he long rae, j. Esimaion periods are 1979:3:-199:7:15 for he hree-monh mauriy, 1979:3:-199:4:15 for he six-monh mauriy and 1979:3:-1995:1:1 for he -monh mauriy. 3 The dummies perain o he week in which an ERM realignmen of he French franc vis-à-vis he Ialian lira occurred (plus he eigh weeks before and afer), he abandonmen of ERM pariy of he lira in Sepember 199 and he broadening of he ERM bands in Augus The dummies perain o he week in which an ERM realignmen occurred (plus he eigh weeks before and afer), he abandonmen of he ERM pariy of he lira in Sepember 199 and he broadening of he ERM bands in Augus Before urning o he resuls, i should be noed ha he fac ha his crierion only classifies as speculaive aacks episodes ha led o changes in he pariy has implicaions for he likely size of he slope parameer. Since cenral banks are likely o cu ineres raes faser afer 17

24 speculaive aacks ha resul in realignmens han afer hose ha do no, he esimaes of β SA may well be larger han uniy. The resuls in Table are quie ineresing. Noe ha he esimaes of β NA are less han uniy (bu ypically no significanly so), while he esimaes of β SA end o be larger han uniy, and in mos cases significanly so. Thus, hese resuls also sugges ha he EH works "beer" (in he sense ha he esimae of he slope parameer is larger) if episodes of speculaive aacks are included in he sample. However, since we are unable o rejec he hypohesis ha β=1 in periods wihou such aacks, i is clear ha moneary policy reacions o speculaive pressures on he exchange rae are no he sole reason why he EH fares beer in European daa. Conclusions The main findings in he paper can be summarised as follows. Firs, here is considerable evidence ha i is more difficul o rejec he EH of he erm srucure in counries in which moneary policy is conduced using inermediae exchange rae arges han in counries where no such arges are used. Secondly, shor-erm ineres raes are more predicable in counries wih such exchange rae objecives. Thirdly, he occurrence of occasional speculaive aacks is one reason why i may be more difficul o rejec he EH under fixed exchange raes, bu i is no he only one. Even resricing he sample period o calm periods, we sill fail o rejec he EH in counries wih exchange rae objecives. The fac ha differences in he predicabiliy of shor-erm ineres raes maer for he es of he EH considered here suggess ha ime-varying erm premia may be presen. In floaing exchange rae counries where shor-erm raes are relaively difficul o forecas, he ime variaion of he erm premium drives b sufficienly far away from uniy for us o rejec he EH. By conras, in fixed exchange rae counries where shor-erm raes are easier o predic, esimaes of b are dominaed by expeced changes of he shor rae and we are herefore unable o rejec he EH. 18

25 References Dahlquis, Magnus and Gunnar Jonsson (1995): "The informaion in Swedish shor-mauriy forward raes". European Economic Review, June, 39: Gerlach, Sefan (199): "Moneary policy and he behaviour of ineres raes: Are long raes excessively volaile?" BIS Working Paper, No. 34. Gerlach, Sefan and Frank Smes (1997): "The erm srucure of euro-raes: some evidence in suppor of he EH". Journal of Inernaional Money and Finance, April, 1: Hardouvelis, Gikas A. (1988): "The predicive power of he erm srucure during recen moneary regimes". Journal of Finance, June, 43: Hardouvelis, Gikas A. (1994): "The erm srucure spread and fuure changes in long and shor raes in he G-7 counries: Is here a puzzle?" Journal of Moneary Economics, April, 33: Kugler, Peer (1988): "An empirical noe on he erm srucure and he ineres rae sabilizaion policies", Quarerly Journal of Economics, November, 13: Kugler, Peer, "The erm srucure of Euro ineres raes and raional expecaions". Journal of Inernaional Money and Finance, June 199, 9: Jondeau, Eric and Roland Ricar (199): "The expecaions heory: ess on French, German and American euro-raes", in The deerminaion of long-erm ineres raes and exchange raes and he role of expecaions, BIS Conference papers Vol., Basle, Swizerland. Mankiw, N. Gregory and Jeffrey A. Miron (198): "The changing behavior of he erm srucure of ineres raes". Quarerly Journal of Economics, May, 11: Newey, Whiney K. and Kenneh D. Wes (1987): "A simple, posiive semi-definie, heeroscedasiciy and auocorrelaion consisen covariance marix". Economerica, May, 55: Shiller, Rober J. (199): "The erm srucure of ineres raes", in Benjamin M. Friedman and Frank H. Hahn, eds., Handbook of Moneary Economics, Vol. 1, Amserdam: Norh-Holland. Shiller, Rober J., John Y. Campbell, and Kermi I. Schoenholz (1983): "Forward raes and fuure policy: Inerpreing he erm srucure of ineres raes". Brookings Papers on Economic Aciviy, No. 1: Svensson, Lars E.O. (1993): "Assessing arge zone credibiliy: Mean reversion and devaluaion expecaions in he ERM, ", European Economic Review, May, 37:

26

27 Recen BIS Working Papers No. Tile Auhor 8 Augus Sepember November December December January January July Augus Sepember Ocober January March May June 1997 The erm srucure of Euro-raes: some evidence in suppor of he expecaions hypohesis The informaion conen of he erm srucure: evidence for Germany Money demand sabiliy and currency subsiuion in six European counries ( ) Tesing he quaniy heory using long-run averaged cross-counry daa The anaomy of he bond marke urbulence of 1994 Derivaives and asse price volailiy: a es using variance raios Moneary policy and he behaviour of ineres raes: are long raes excessively volaile? Varieies of moneary policy operaing procedures: balancing moneary objecives wih marke efficiency Esimaion of speculaive aack models: Mexico ye again Does he erm srucure predic recessions? The inernaional evidence Inernaional agreemens in he area of banking and finance: accomplishmens and ousanding issues Banking sysem failures in developing and ransiion counries: diagnosis and predicion Moneary policy operaing procedures indusrial counries The euro and European financial markes Measuring moneary policy shocks in France, Germany and Ialy: he role of he exchange rae Sefan Gerlach and Frank Smes Sefan Gerlach Renao Filosa Sefan Gerlach Claudio E.V. Borio Rober N. McCauley Benjamin H. Cohen Sefan Gerlach Joseph Bisignano William R. Melick Henri Bernard and Sefan Gerlach William R. Whie Parick Honohan Claudio E. V. Borio Rober N. McCauley and William R. Whie Frank Smes

28

29

30 ISSN 1-959