Was there monetary autonomy in Europe on the eve of EMU? The German Dominance hypothesis re-examined

Oscar Bajo-Rubio M. Dolores Montávez-Garcés

EEE 52

January, 2002

FEDEA Fundación de Estudios de Economía Aplicada

http://www.fedea.es/hojas/publicado.html

Oscar Bajo-Rubi o M. Dolores Mont ávez- Gar cés

(Uni ver sidad Públi ca de Navarra)

AB ST RAC T

In this paper we re- examine the Germ an dom inance hypothesi s, as a way to assess whet her the loss of monetary autonom y in Eur ope associated with EMU had been si gni fi cant. We use Granger- causalit y tests bet ween the interest rat es of Germany and all the countr ies part ici pat ing at any ti me in the E ur opean Monet ary S yst em, with the sam ple peri od runni ng until December 1998. Our result s would support a weak ver sion of the hypothesis, wit h Ger many pl aying a certain “l eadershi p” or speci al role in the EMS, although she woul d not had been st ri ctl y t he “domi nant” pl ayer.

Key wor ds: European m onetar y uni on, Germ an dominance hypothesis, Gr anger -causali ty JE L codes: F33, F36, E50

1. Introdu ct ion

Begi nni ng on January 1st 1999, and foll owi ng the adopti on of a com mon curr ency (the eur o) and the st ar ting of the European Central Bank’s oper ati ons, 11 Eur opean countri es now form a monetar y uni on (the Econom ic and Monetary Union, EMU). As it becom es obvious, EMU means the loss of monet ary independence of the par ti cipati ng count ri es, which might be seen as a cost, at l east at a f irst sight.

Things are not so si mpl e, however. As it is wel l known, accordi ng to the so- cal led “i nconsist ent t rinit y” pri ncipl e, a fixed exchange r ate, f ul l capi tal m obi li ty, and the independence of monetar y pol icy, are not mut ual ly compati ble. And this si tuation roughl y applied to the European econom ies before EMU, whi ch shared a quasi- fixed exchange rate syst em (the European Monetary System, EMS), and especi al ly fol lowing the el imi nation of capital controls af ter the Si ngl e E ur opean Act i n 1990-92. This fact led to t he count ries par ticipati ng in the E MS to real ize that they were gr aduall y losing the contr ol of their monetar y pol ici es in favor of the Bundesbank, the cent ral bank of Germ any, i. e., the countr y presum ed to act as a leader in the EMS. Hence, EMU coul d emer ge as an economi c response to that si tuati on, on allowing those countri es to regai n som e control over monetary pol icy thanks to the creati on of an European Cent ral Bank replaci ng the Bundesbank, in which they could have a vote (Wypl osz, 1997).

In fact , a general consensus had emerged in Eur ope which would justi fy the previous ar gument, i. e. , that the EMS had worked in an asymm etr ic way, wit h Ger many assumi ng the leading role and t he remai ni ng count ries passively adjusti ng to Germ an monet ary poli cy act ions. In its tur n, these countri es would have benefit ed fr om behaving in such a way, since they would have taken advantage of the fir mly establi shed ant i- inf lat ion credibili ty of the Bundesbank [see, e. g., Giavazzi and Pagano (1988) or Mélit z (1988) ]. This di scussi on ul tim at ely li es in the socall ed n-1 problem f aced by fixed exchange r ate syst ems, since there ar e onl y n-1 exchange rat es am ong the n countr ies part ici pating in an exchange rate agreement. Ther efore, in such a si tuati on, either one countr y becomes the leader and sets monet ary poli cy independently (wit h the other count ries fol lowing it), or all count ries are al lowed to deci de joint ly over the im pl ementati on of monet ary poli cy (De Grauwe, 1997).

The fir st em pir ical studies on the subj ect seem ed to conf irm the hypothesis of German domi nance into the EMS [see, e. g. , Giavazzi and Giovannini (1987, 1989) or Karf aki s and Moschos (1990)] . However, these conclusions wer e not confi rm ed in further research, most of it consist ing of test s for Gr anger -causali ty between German and ot her countri es’ interest rat es at a mont hly or quar ter ly fr equency [see, am ong others, Cohen and Wyplosz (1989), von Hagen and Fr at ianni (1990), Koedi jk and Kool ( 1992), Katsimbri s and Mi ller ( 1993) , or Hassapis, P itt is and Pr odrom idi s (1999) ]. In this way, a mil der support for the hypothesi s was found in the above quot ed paper s; nam el y, that the ot her countr ies’ int erest rates depended on the Germ an ones, but al so conversely, even though in a lower extent in terms of both si ze and per sistence. Fi nally, results al ong these lines were also report ed in some st udi es using high fr equency (i . e., daily) data on interest r at es [see Gar dner and Perr audin (1993), Henry and Wei dmann (1995) and Bajo, Sosvill a and Fernández (2001)], so t hat it m ight seem t hat Germ any woul d have played a speci al role in the EMS , alt hough calli ng it “domi nance” would be too strong.

In this paper we re- examine the Germ an dom inance hypothesi s, as a way to assess whet her the loss of monetary autonom y in Eur ope associated with EMU had been si gni fi cant (whi ch, in its tur n, could be taken as an ar gum ent in favor of EMU itself) . The em pi rical methodology makes use of Granger-causal ity test s bet ween the monthly inter est rates of Germ any and all the countr ies part icipating at any time in the exchange rate mechani sm (ERM) of the EMS , wit h the sampl e per iod running unti l December 1998. This paper cont ribut es to the existing l it eratur e in the f oll owi ng respect s:

a) The sam ple peri od cover s unt il just the eve of EMU, i. e., December 1998. This al lows us to include the m ost r ecent event s in Eur opean monetar y histor y, such as the Germ an reuni ficati on, t he monet ary t urm oil at t he end of 1992, the broadening of t he EMS fluctuat ion bands in August 1993, and the rather quiet peri od leadi ng to the bi rt h of EMU. Regarding pr evious studies on the subj ect , those wit h a more recent sample per iod are Hassapis, Pit tis and Prodr omi dis (1999), who use quar ter ly data unti l the end of 1994, and Bajo, Sosvil la and Fernández (2001), who use dail y data unti l F ebr uary 1997.

b) The analysis is extended to all the countr ies part icipating at any time in the ERM of the EMS . So, unl ike previous studies (wi th the only excepti on of Bajo, Sosvil la and Fernández (2001)), that consider onl y the founding members of the EMS (i. e. , Germ any, France, Italy, Belgium , the Netherl ands, Denmark, and Ireland) , our anal ysi s also incl udes those count ri es whi ch later joined the ERM of the EMS (i . e., Spai n, the UK, Por tugal , and Austr ia).

c) Gr anger -causali ty in a coi nt egr ati on setti ng is pr operl y tested. That is, an er ror - corr ect ion mechani sm (E CM) is incl uded int o every equat ion to be est imated when cointegrat ion is found, which allows us to dist ingui sh bet ween short -run and longrun Granger- causal it y¹. Al so, and fol lowing Katsim bri s and Mi ller’ s (1993) suggest ion, Granger- causal it y r elati onships bet ween Ger man and the other countr ies’ interest rat es have been invest igated both in a bi variate and a tr ivari ate sett ing, in or der to avoid possi ble spur ious result s due to the omi ssi on of some relevant vari abl e. As usual , the US interest rat es is the additi onal var iable added to the anal ysi s.

d) Fi nally, and gi ven the impor tance of the choice of lag lengt hs in Granger -causali ty test s, these have been sel ected by means of an appropri ate method. In part icular, we have used Hsiao’s (1981) sequenti al approach, specif icall y designed to avoid im posing oft en fal se or spur ious restri cti ons on the model . Not ice that , unl ike the VAR approach perform ed in ot her st udies [as in, e. g., Hassapi s, Pi tti s and Pr odrom idi s (1999) ], our procedur e implies that, for any pai r of var iables tested for Gr anger -causali ty between them, the num ber of lags of the ri ght -hand si de variables is not const rai ned t o be t he same.

The rest of the paper i s str uct ured as fol lows. The econom et ric methodology of the paper is discussed in Sect ion 2, and the empi rical resul ts ar e shown in Secti on 3. The m ai n conclusions ar e presented i n S ection 4.

Kats imb ris and Miller ( 199 3) were th e f irs t to notice this p oin t, us ually ov erlook ed in th e available em pirical studies on th is su bject.

2. Econ ometric met hodol ogy

As stat ed befor e, the econom etr ic methodol ogy used in this paper is based on Gr anger -causali ty test s (Granger, 1969). As it is well known, the results fr om these test s are hi ghl y sensit ive to the or der of lags in t he autor egressive process. An inadequate choi ce of the lag lengt h would lead to inconsi stent model esti mat es, so that the infer ences dr awn from them would be likely to be mi sl eading. In thi s paper, we will identif y the or der of lags for each var iable by means of Hsiao’s (1981) sequenti al approach, whi ch is based on Granger’s concept of causali ty and Akai ke’ s f inal predi cti on er ror cr it eri on.

Suppose two stationary var iables, and on which we woul d like to test for Gr anger - causali ty. Consider the models:

\[X _ {t} = \alpha + \sum_ {i = 1} ^ {m} \beta_ {i} X _ {t - i} + u _ {t}\tag{1}\]

\[X _ {t} = \alpha + \sum_ {i = 1} ^ {m} \beta_ {i} X _ {t - i} + \sum_ {j = 1} ^ {n} \gamma_ {j} Y _ {t - j} + v _ {t}\tag{2}\]

and then t he followi ng steps ar e used t o apply Hsi ao’s procedur e:

(i) Take to be a uni var iat e aut oregr essive process as in (1) , and compute it s final pr edict ion error cri ter ion (FPE hereaft er) with the order of lags i var ying from 1 to M. Choose the lag that yiel ds the smallest say and denote the corr esponding F PE as

(ii) Tr eat as a cont rolled var iable wi th m lags, add lags of to (1) as in (2), and compute the FPE s wit h the or der of lags j var ying from 1 to N. Choose the lag that yi elds the sm al lest FPE , say n, and denote the cor respondi ng FPE as

(iii) Compare wit h . If , then is sai d to Gr anger -cause whereas if , then would not be Gr anger -caused by

Fi nally, by repeat ing steps (i) to (i ii ) with as the dependent variable, whether or not Gr anger -causes can be establi shed.

Recall that bef ore it was assum ed that and wer e stat ionar y var iables. However, if they ar e int egr ated of order one (i. e. , fir st- dif ference st ati onary) and ar e cointegrated, equati ons (1) and (2) need t o be amended to:

\[\Delta X _ {t} = \alpha + \sum_ {i = 1} ^ {m} \beta_ {i} \Delta X _ {t - i} + \delta z _ {t - 1} + u _ {t}\tag{3}\]

\[\Delta X _ {t} = \alpha + \sum_ {i = 1} ^ {m} \beta_ {i} \Delta X _ {t - i} + \sum_ {j = 1} ^ {n} \gamma_ {j} \Delta Y _ {t - j} + \delta z _ {t - 1} + v _ {t}\tag{4}\]

wher e is the ECM (Engle and Granger, 1987). Not ice that if and are I( 1) but are not cointegrat ed, t he coeff ici ent i n equati ons ( 3) and ( 4) would be equal t o zer o.

Now, the previous definiti ons of Granger-causal ity for stati onary variables can be appl ied to the case of I(1) var iables from equations (3) and (4). In parti cular , if , Y_t is sai d to Granger- cause in the shor t run; and if is signif icant ly di ff erent fr om zer o, is sai d to Granger- cause in the long run. Conversely, if , X_t would not be Granger-caused by in the shor t run; and if is not si gnifi cantl y diff er ent fr om zero, would not be Granger-caused by in the long run. As before, by repeati ng the procedure with as the dependent variable, the hypothesis of shor trun and long-run Granger-causal ity f rom t o could be tested.

To conclude, notice that the above procedure corresponds to the bi variate case. Test ing for Granger- causal it y i n t he tr ivari ate case requi res amendi ng the previous equati ons t o:

\[X _ {t} = \alpha + \sum_ {i = 1} ^ {m} \beta_ {i} X _ {t - i} + \sum_ {k = 1} ^ {p} \theta_ {k} W _ {t - k} + u _ {t}\tag{1'}\]

\[X _ {t} = \alpha + \sum_ {i = 1} ^ {m} \beta_ {i} X _ {t - i} + \sum_ {j = 1} ^ {n} \gamma_ {j} Y _ {t - j} + \sum_ {k = 1} ^ {p} \theta_ {k} W _ {t - k} + v _ {t}\tag{2'}\]

wher e denotes the third vari abl e, for the case in which and are st ati onary; and, for the case i n whi ch the t hree var iables are I( 1) and cointegrated:

\[\Delta X _ {t} = \alpha + \sum_ {i = 1} ^ {m} \beta_ {i} \Delta X _ {t - i} + \sum_ {k = 1} ^ {p} \theta_ {k} \Delta W _ {t - k} + \delta z _ {t - 1} + u _ {t}\tag{3'}\]

\[\Delta X _ {t} = \alpha + \sum_ {i = 1} ^ {m} \beta_ {i} \Delta X _ {t - i} + \sum_ {j = 1} ^ {n} \gamma_ {j} \Delta Y _ {t - j} + \sum_ {k = 1} ^ {p} \theta_ {k} \Delta W _ {t - k} + \delta z _ {t - 1} + v _ {t}\tag{4'}\]

so that the rel evant compari son is now bet ween and , and bet ween and , r espect ively; where and ar e t he combi nat ions of lags leadi ng to the smallest FP E i n each case.

3. Empi rical resul ts

The dat a used in thi s paper are the thr ee- month interbank onshore inter est rates, at a monthly fr equency, of Germ any, France, Italy, Belgium, the Netherl ands, Denm ark, Ireland, Spain, the UK, Por tugal , Aust ri a, and the US. The previ ous li st includes all the European count ries part ici pat ing at any ti me in the ERM of the EMS , and coincides wit h that of the countri es joining EMU from the outset, which the exceptions of Denmark and the UK, and the inclusion of Luxembour g and Fi nland². The begi nning of t he sam pl e peri od is March 1979 ( i. e., when the ERM started to operate) for the founding mem ber s of the EMS (Fr ance, It aly, Bel gium, the Netherl ands, Denmark, and Ir eland) , and the month of accessi on to the E RM for t he newcomer s: June 1989 for Spai n, October 1990 for the UK, Apri l 1992 for Portugal, and January 1995 for Aust ria, wit h the data for Germ any and the US adjust ing accordi ngl y in each case. The end of the sam ple is in all cases December 1998 (i. e. , the last month before the star ting of EMU), and al l the data come fr om the St at ist ic Bulletin of the Bank of Spai n.

As a fi rst step of the analysis, we tested for the order of int egr at ion of the var iables by means of the Di ckey- Ful ler and Phi ll ips-Perr on tests. According to the resul ts from bot h tests, shown i n T able 1, the null hypothesi s of a unit root was not rejected i n all cases, at the same ti me that the nul l of a second unit root was al ways rej ected.

Next , we tested for coi ntegr ati on between the Germ an inter est rate and the interest rat es of the other European countr ies in our sam pl e, bot h in a bivari ate and tri variate setti ng, in the latt er case including t he US inter est r ate as an additi onal var iable. T wo tests were perform ed: the (coi ntegrati ng regressi on) augm ent ed Di ckey- Ful ler and Phi ll ips-Ouli ari s tests. Both tests were computed usi ng the resi duals fr om the (bivar iat e or tri var iate) cointegrat ing regr essions esti mat ed by the met hod pr oposed by Phi lli ps and Hansen (1990), robust to the presence of seri al cor relat ion and endogeneity bias.

The result s of the cointegration tests appear in Table 2. As can be seen, the only interest rates appear ing to be cointegrated with the Ger man ones in the bivar iat e case woul d be those of

No tice that Lux emb ou rg, a fo und ing m emb er of th e EMS , is n ot in clu ded in the sample sin ce sh e already fo rm ed a m on etary un ion with Belgium befor e EMU . A ls o, Fin land, wh ich p articipated in th e ERM of the EMS s ince O ctober 1 99 6, is no t includ ed giv en th e s mall num ber o f o bs erv ation s av ailab le.

Aust ria and the Netherl ands; whereas, in the tr ivari ate case (i . e., when the US int erest rates ar e incl uded int o the coint egr at ion equation), cointegration is also found in the cases of Bel gi um, Denm ark and Ireland. These results should not be too surpr ising si nce, as noticed by Capor al e and Pit tis ( 1995), t he int egrat ion of t he fi nancial mar ket s of the E MS count ries would had been a gr adual pr ocess, leading to a slow convergence process of inter est rates towards the German levels. Hence, coi nt egr ati on shoul d be expected only when full convergence had been achieved³.

Now, we ar e abl e to per for m Granger- causal it y test s in a coi ntegrati on framewor k, and the result s for the bivari at e case are shown in Tabl e 3. Ger man interest rat es appear to Grangercause all the other EMS interest rat es, the opposi te being also tr ue in al l cases but those of Aust ria, Ireland, and the UK; bilateral causali ty is al so found between Germ an and US interest rates. Not ice, however, that , when bilater al Gr anger -causali ty is found, the decrease in FPE s is gr eater when German int erest rates are added to the equati ons expl ai ning the ot her interest rat es than in the opposi te case (t he exception bei ng the US case). On the other hand, bi later al long- run causali ty would appear in the case of the Netherlands, whereas Ger man interest rat es would cause those of Austr ia in the long run, but not the other way round. These results woul d suggest that , although there would have been some degree of sym met ry in the EMS , the influence of Germ any on t he other EMS countr ies woul d have been greater t han the other way r ound.

Next , we t ur n t o t he tr ivari ate case in Tabl e 4. Beginning with causali ty between German interest rat es and those of the ot her EMS count ries, the result s, shown in part A) of Tabl e 4 are quit e simi lar to those in Table 3. The onl y except ion woul d be the bilater al Gr anger -causali ty now found for Aust ri a; also, Danish int erest rates do not appear to Granger- cause the Germ an ones, even though the diff er ence bet ween FPE s in thi s case woul d be ver y small. Agai n, the Germ an int er est rates add more explanat ory power to the equations explaini ng the other int er est rates than in the opposite case. Regarding long-run Granger- causal it y, it would ar ise in a bi later al way for Belgi um, Denm ark, and the Net her lands; also, Ger man interest rat es would Gr anger -cause those of Austr ia, and, more surpr isi ngly, Ir ish interest rat es would Granger -cause those of Ger many, al though only at a 10 per cent signif icance l evel.

So me ev idence alon g these lines fo r the Sp an ish case can b e fou nd in Camar er o, Estev e and Tamarit (1 99 7).

We have al so tested for Gr anger -causali ty between the US int erest rates and those of the EMS countr ies other than Ger many, as well as between German and US interest rat es, with the results appeari ng in parts B) and C) of Tabl e 4, respectivel y. As can be seen, the US interest rates woul d Granger- cause those of Spai n, Fr ance, It aly, and Portugal, in the shor t run; Denm ark, the Netherl ands, and Irel and, in the long run; and Austri a and Belgium , bot h in the shor t run and the long run. In its turn, the inter est rates of Bel gi um, Denm ark, France, the Netherl ands, Ir eland, Ital y, Portugal, and the UK would Gr anger -cause the US inter est rates in the short run. On the other hand, bi lat eral short- run Granger-causal ity is found bet ween Ger man and US int er est rates in most cases (the except ions bei ng when the interest rat es of Spain, Port ugal, and the UK ar e included in the regressions), but no clear long-r un Gr anger -causali ty is detected (ot her than that found fr om the US to Ger many when the interest rat es of Denmark ar e used).

Fi nally, we have also test ed for str uct ural change in all the esti mated equations shown in tabl es 3 and 4, by means of the Chow test. The dat es chosen are: November 1990 (the Ger man reunifi cat ion), Sept ember 1992 (the beginning of the turbulent per iod affect ing the EMS ), and August 1993 (the broadening of the fluctuati on bands in the EMS ), and the tests ar e onl y perf orm ed for the inter est rates of the EMS foundi ng members, given the reduced number of observations avail able for the newcomer s. As can be seen in Table 5, most of the tests fai l to reject the null hypothesis of stabil ity. The most relevant excepti on would be the Fr ench case, wher e a st ructural change in Gr anger -causali ty from Ger many would be detected foll owing the Germ an reuni ficati on and the monet ar y turm oi l at the end of 1992, both in the bivari ate and tr ivari ate m odels.

To summ ari ze, bilateral Gr anger -causali ty has been found between the int erest rates of Germ any and the ot her countr ies part ici pat ing at any ti me in the ERM of the EMS , the main excepti ons being Ireland and the . However, when bi later al Gr anger- causalit y is found, the incr ease in explanat ory power is greater when Germ an inter est rates are added to the equat ions expl aining t he other inter est r ates than t he ot her way round. T her ef ore, our resul ts would point t o a certain “l eadershi p” or speci al role of Germany wi thi n the EMS, al though we coul d not talk of “dom inance” in a str ict sense. In parti cul ar , and accor ding to the term inology int roduced by Hassapi s, Pi tti s and Pr odr om idi s ( 1999) , we could establish the foll owi ng typol ogy:

In f act, A rtis and Zhan g ( 19 97) fo un d that I reland and the U K wo uld h ave f ollowed in recen t year s a diff erent cy clical evolution as co mp ared to the other Euro pean cou ntries.

a) St rong Ger man domi nance: t he UK.

b) Weak German dom inance of t ype 1: S pain and P ort ugal.

c) Weak German dom inance of t ype 2: Austri a, Belgi um, F rance, and Italy.

d) Semi st rong Ger man domi nance: Denm ar k, Ireland, and the Netherl ands.

4. Conclusions

In this paper we have re-examined the Germ an dominance hypot hesis, extendi ng pr evi ous fi ndings by other authors to al l the count ri es par ti cipati ng at any tim e in the ERM of the EMS, wi th the sam ple peri od cover ing unti l just the eve of EMU, i. e., Decem ber 1998. The em pir ical methodology makes use of Granger-causal ity test s bet ween the inter est rates of Ger many and the other EMS countr ies, in a proper coint egrat ion fram ework where the lag lengths of the vari abl es have been chosen by means of Hsi ao’s sequenti al approach in order to avoid mi sl eading inferences arising from inconsi st ent model esti mates. The tests have been perform ed in both a bi var iat e and a tr ivariate setti ng, in thi s case incl udi ng the US int erest rate as the addi tional vari abl e.

Summ ari zing, our result s poi nt to a mut ual but asymm etr ical rel ati onshi p bet ween Germ any and the ot her countr ies part ici pat ing at any ti me in the ERM of the EMS , since bi later al Gr anger- causalit y was found between the inter est rates of Ger many and those of the ot her countr ies (wit h the exceptions of Ir el and and the UK), al though the German int erest rates added more to the explanat ion of the ot her interest rat es than in the opposi te case. Al so, we did not find evi dence of si gni fi cant str uct ural changes in the esti mat ed relat ionships foll owi ng the Germ an reuni ficati on, the monet ary turm oil at the end of 1992, and the broadeni ng of the fl uctuation bands in the E MS .

Ther efore, our resul ts would support a weak ver sion of the hypothesi s of Ger man domi nance during the worki ng of the EMS , since there would have pr evail ed a mut ual relationship am ong the monet ary poli cies of all the countr ies involved (wi th the except ions of Ir el and and the UK), even though that relati onship woul d have been stronger from Ger many to the other count ries than in the opposit e way. Then, Ger many would have played a cert ain “l eader shi p” or special role in the EMS , alt hough she woul d not had been str ict ly the “dom inant” player.

Regardi ng the poli cy im pli cations of the paper, these woul d provide som e mil d suppor t to the hypot hesis about EMU as an economic response to the loss of monetar y aut onomy in Europe in favor of Ger many, especially af ter the achievem ent of ful l capi tal mobi li ty in the first ni neties (Wyplosz, 1997). Al so, the positi on of the Mediterr anean count ries (It aly, Spain, and

Port ugal) faced to EMU does not seem to be quit e dif fer ent to that of the “core” Eur opean countri es, at least in ter ms of the aut onomy of thei r monetary pol icies. The same can be sai d for Denm ark and the UK, two countri es currentl y not in EMU; in fact , according to our resul ts, the UK woul d have been (together wi th Ir eland) the count ry the most “dom inated” by Ger man monetar y pol icy acti ons.

Referen ces

Ar ti s, M. J. and Zhang, W. (1997): “Int ernat ional business cycl es and the ERM: Is there a European business cycle?”, International Journal of F inance and Economi cs 2, 1-16.

Bajo, O., Sosvi lla, S. and Fernández, F. (2001) : “Asymm etr y in the EMS: New evi dence based on non- linear f orecasts”, European E conomic Review, 45, 451- 473.

Camarer o, M. , Esteve, V. and Tamar it , C. (1997) : “Convergencia en ti pos de interés de la economí a española ante la unión monetar ia europea”, Revi sta de A nál isi s Económico 12, 71-99.

Caporal e, G. M. and Pit tis, N. (1995): “Interest rat e linkages wit hi n the European Monetar y Syst em: An alternati ve int er pretat ion”, Appl ied Economi cs Letters 2, 45- 47.

Cohen, D. and Wypl osz, C. (1989): “T he Eur opean Monetar y Uni on: An agnosti c evaluati on”, en R. Bryant et al. (eds.) : Macroeconomi c poli ci es in an interdependent world, Washi ngt on, DC: Int ernat ional Monet ary F und, 311-337.

De Grauwe, P. (1997) : The economics of monetary integrat ion (3r d edit ion), Oxford: Oxford University P ress.

Engl e, R. F. and Granger, C. W. J. (1987): “Co-integration and er ror corr ection: Representation, esti mat ion and testi ng”, Econometri ca 55, 251-276.

Gardner , E. and Perr audin, W. (1993) : “Asymm etr y in the ERM: A case study of Fr ench and Germ an int er est rates before and aft er Ger man unif icati on”, International Monetary Fund Staff Papers 40, 427-450.

Gi avazzi, F. and Giovannini, A. (1987): “Models of the EMS : Is Eur ope a gr eater Deut schmar k ar ea?”, in R. Bryant and R. Por tes (eds.): Gl obal macroeconomics. Pol icy conf li ct and cooperation, London: Macmi llan, 237-265.

Gi avazzi, F. and Giovannini, A. (1989): Li mi ting exchange rate flexi bil ity. The European Monetary System, Cambr idge, MA: T he MI T P ress.

Gi avazzi, F. and Pagano, M. (1988) : “The advant age of tying one's hands. EMS di sci pl ine and Cent ral Bank cr edi bi lit y”, European E conomic Review 32, 1055- 1075.

Gr anger , C. W. J. (1969): “I nvesti gating causal relations by econometri c models and crossspectral m et hods”, Econometri ca 37, 424-438.

Hassapi s, C. , Pitt is, N. and Pr odr om idi s, K. (1999): “Unit root s and Gr anger causali ty in the EMS int erest rates: The Germ an Dom inance Hypothesi s revisi ted”, Journal of Internati onal Money and Fi nance 18, 47-73.

Henr y, J. and Weidmann, J. (1995): “Asymmetr y in the EMS revisi ted: Evi dence fr om the causali ty analysis of dail y Eur orates”, Annales d' Économie et de Statistique 40, 125-160.

Hsiao, C. (1981): “Autoregressi ve model ling and money-i ncome causali ty det ection”, Journal of Monetary E conomics 7, 85- 106.

Karf aki s, C. and Moschos, D. (1990): “I nterest rat e linkages wi thi n the European Monetary Syst em: A ti me ser ies anal ysis”, Journal of M oney, Credi t, and B anking 22, 388-394.

Katsimbris, G. and Mill er, S. (1993) : “Int er est rate li nkages within the Eur opean Monet ary Syst em: Furt her anal ysi s”, Journal of M oney, Credi t, and B anking 25, 771-779.

Koedijk, K. and Kool , C. (1992) : “Dominant interest and infl ati on di fferenti als wi thin the EMS”, European E conomic Review 36, 925-943.

MacKinnon, J. G. (1991) : “Cr iti cal values for cointegration tests”, in R. F. Engle and C. W. J. Gr anger (eds.): Long-run economic relat ionships: Readings in coint egrat ion, Oxfor d: Oxfor d University P ress, 267-276.

References

1. Méli tz, J. (1988): “Monetary di sci pl ine and cooper at ion in the Eur opean Monetar y System : A synt hesis”, in F. Gi avazzi , S. Micossi and M. Mill er (eds. ): The European Monet ary Syst em, Cambridge: Cambridge University Pr ess, 51- 79.

References

1. Phil lips, P. C. B. and Hansen, B. E. (1990): “S tat istical infer ence in instr umental var iables regr ession with I( 1) pr ocesses”, Revi ew of Economic Studies 57, 99-125.

References

1. Phil lips, P. C. B. and Oul iaris, S. (1990) : “Asymptotic pr opert ies of resi dual based tests for cointegrat ion”, Econometri ca 58, 165-193.

References

1. von Hagen, J. and Fr ati anni, M. (1990): “Ger man domi nance in the EMS : Evidence from interest r at es”, Journal of I nternati onal M oney and F inance 9, 358-375.

References

1. Wypl osz, C. (1997) : “EMU: Why and how it might happen”, Journal of Economi c Perspect ives 11, 3-21.

A) DICK EY- FULLER TES T

No tes: (i) an d τ are th e D ickey -Fu ller statistics with drift an d trend, with dr ift, and with out dr if t, res pectively. (ii) (a), (b ), an d ( c) denote s ig nif icance at the 1%, 5 %, an d 1 0% levels, respectively. The critical valu es are taken f rom MacK inn on (1 991 ).

B) P HILLIP S- PERRON TEST

No tes: (i) an d are th e P hillips- Perro n s tatis tics with d rift and tren d, with drift, and witho ut dr ift, resp ectively. (ii) (a), (b ), an d ( c) d eno te significan ce at th e 1 %, 5%, and 10 % levels , r esp ectiv ely . The cr itical v alues ar e taken f rom MacK inn on (1 991 ).

CO INTEG RATIO N TESTS

A) DICK EY- FULLER TEST

No tes: (i) Th e tes t r ef ers to the cointegr ating -regress ion au gm ented Dickey-F uller statistics o n the Ph illips -H ans en residuals. (ii) (a), (b ), an d ( c) denote s ig nif icance at the 1%, 5 %, an d 1 0% levels, respectively. The critical valu es are taken f rom MacK inn on (1 991 ).

B) P HILLIP S- OULIARIS TEST

No tes:

(i) Th e tes t r ef ers to the cointegr ating -regress ion s tatis tic o n the Ph illips -H ans en resid uals.

(ii) (a), (b ), an d ( c) denote s ig nif icance at the 1%, 5 %, an d 1 0% levels, respectively. The critical valu es are taken f rom P hillip s and Ou liaris ( 19 90).

GRANGER-CAUS ALITY TESTS : BIVARIATE MODELS

(i) m and n d en ote th e lag s f or th e d ep end ent v ariable and th e add ition al reg resso r, respectively, lead ing to the sm alles t F PE in each case; the max im um num ber o f lag s tried has been 12 . X and G d en ote ev er y coun tr y in the fir st co lum n of the table, an d G erm an y, res pectively. No tes: (ii) (a), (b ), an d ( c) denote s ig nif icance at the 1%, 5 %, an d 1 0% levels , r esp ectiv ely , for th e t-s tatis tics of the ECMs (in p ar enthes es ).

GRANGER-CAUS ALITY TESTS : TRIVARIATE MODELS

A) Causality between German and EMS interest rates

No tes: (i) m, n and p d en ote th e lag s f or th e d ep end ent v ariable, th e add itional reg res so r, and the US in ter est r ate, resp ectively , lead in g to the sm alles t F PE in each case; th e max imu m num ber o f lags tried h as been 12 . X an d G d en ote ev er y coun tr y in the first colu mn of th e tab le, an d Ger man y, respectively. (ii) (a), (b ), an d ( c) denote s ig nif icance at the 1%, 5 %, an d 1 0% levels, respectively, f or the t-s tatis tics of the ECMs (in p ar enthes es ).

TABLE 4 (con tin ued )

B) Causality between US and EMS interest rates

No tes: (i) m, n and p d en ote th e lag s f or th e d ep end ent v ariable, th e add itional reg res so r, and the Ger man interest rate, resp ectively , lead in g to the sm alles t F PE in each case; th e max imu m num ber o f lags tried h as been 12 . X an d US d en ote ev er y coun tr y in the first colu mn of th e tab le, an d the US , res pectively. (ii) (a), (b ), an d ( c) denote s ig nif icance at the 1%, 5 %, an d 1 0% levels, respectively, f or the t-s tatis tics of the ECMs (in p ar enthes es ).

TABLE 4 (con tin ued )

C) Causality between US and German interest rates

No tes: (i) m, n and p d en ote th e lag s f or th e d ep end ent v ariable, th e add itional reg res so r ( Ger many or th e U S), an d every co untry in th e f ir st colum n o f the table, res pectively, leading to th e s mallest FP E in each case; the maxim um nu mber o f lag s tried has b een 12 . G and US d en ote Germ any an d the US , resp ectively. (ii) (a), (b ), an d ( c) denote s ig nif icance at the 1%, 5 %, an d 1 0% levels, respectively, f or the t-s tatis tics of th e ECMs ( in paren th eses).

TABLE 5 TESTS O F S TRUCTURAL CHANGE

A) Causality between German, EMS, and US interest rates (bivariate models)

B) Causali ty between German and EMS i nt erest rates (t ri variat e mod els)

C) Causali ty between US an d EMS in terest rates (trivari ate models)

D) Causali ty between US an d G erman in terest rat es (trivari ate m odels)