Testing for Monetary Policy Convergence
in European Countries
Stilianos Fount as2 3
Working Paper No. 19
Department of Economics
National University of Ireland, Galway
1 Department of Economics, University of Newcastle, upon Tyn.P. United King-
2Department of Economics- National University of Ii-elaud, Gahvay, email:
3This paper is based on the MA Thesis of the first author submitted to the
National University of Ireland, Galway. We thank participants in the Galway Eco-
nomics Workshop for very helpful comments and suggestions. All remaining errors
and omissions are our own responsibility.
The paper tests for long-run monetary policy convergence and short-run
policy interactions in seven ERM countries over the 1979-1992 period using
the approach of multivariate cointegration and Granger-causality tests. We
provide evidence for very little monetary policy convergence, even during
the more stable 1987-1992 period. Our tests for short-run monetary policy
interactions show that, in agreement with some other studies, Germany is
not the leader country in the system as it appears to accommodate shocks
in other member countries. Our tests show also that full monetary pol-
icy convergence applied among Germany, Belgium and Netherlands in the
1987-1992 period implying that these countries could be the first to join a
European monetary union should a two-speed approach to monetary union
become a reality.
Keywords: Policy Convergence, German Dominance Hypothesis, Com-
mon Stochastic Trends.
JEL Classification: F33, F42
The European Monetary System (EMS) was established to allow member
countries to reduce their inflation rates and nominal exchange rate volatil-
ity. Empirical evidence has shown this to be the case (see, for example,
Artis, 1987, Rogoff, 1985). Convergence to a lower inflation rate would
be consistent with long-run monetary policy convergence. However, opin-
ion is divided on the issue whether the EMS has accomplished its primary
objectives by functioning as a symmetric or asymmetric system. In other
words, the necessary monetary policy convergence could have been achieved
through symmetric policy adjustments or in an asymmetric fashion. Under
an asymmetric system, Germany is the leader or dominant country in the
EMS and other countries follow monetary policies similar to Germany's by
pegging their currencies to the DM. By tying their hands, these countries
manage to earn counterinflation reputation (Giavazzi and Pagano, 1988).
It would be expected that the degree of monetary policy convergence would
change during the period since the set up of the EMS. Several authors have
divided the pre-1992 period into at least two subperiods: the volatile March
1979 to January 1987 period and the stable February 1987 to August 1992
period. During the first subperiod, several exchange rate realignments took
place and one would expect to observe lack of monetary policy convergence.
However, the second period was one of considerable exchange rate stabil-
ity and, therefore, one where some degree of monetary policy convergence
The issue of monetary policy convergence is very important for the creation
of monetary union. Significant progress in terms of monetary policy conver-
gence would be necessary for the establishment of monetary union. Hence,
one of the objectives of this paper is to test for the change in the degree
of monetary policy convergence during the pre-August 1992 period in seven
EU countries. In this regard, we make use of the multivariate cointegration
approach and test for the number of common stochastic trends. A single
common stochastic trend would be consistent with full long-run monetary
policy convergence while multiple common trends would imply less than full
convergence. Lack of full long-run convergence would allow some member
countries to pursue monetary policies independent of those of Germany and
hence allow for more than one degree of freedom in monetary policy making
in the EMS. This would allow these countries to achieve certain objectives
such as domestic output growth or a competitive real exchange rate.
A finding of full convergence would imply that in the long run there is
a common monetary policy applied by the member countries. However,
short-run deviations from this long-run common policy would be possible
and one could test for policy asymmetry, i.e., whether German monetary
policy changes are independent of monetary policy changes in the system or
they accommodate shocks to the other member countries. Evidence in favour
of symmetric policy effects would reject the German Dominance Hypothesis
Our objective is to test for both short-run interactions and long-run rela-
tionships in the monetary policies of seven EMS countries. We also look
at smaller groups of countries that could be considered to be in the core
of a European monetary union. The paper is structured as follows: section
2 presents a short literature survey and highlights differences between our
approach and other studies in the area. Section 3 outlines our empirical
methodology, section 4 presents and interprets our results for a group of
seven countries, and section 5 looks at the issue of convergence in the core
countries. Finally, section 6 summarises our results and derives some policy
2 A review of the literature
The issue of monetary policy interdependence in the EMS has been the
subject of extensive research since the late 1980s. The majority of this
research deals with tests for short run and long-run relationships among
proxies of national monetary policies using modern econometric techniques
of nonstationary time series. The long- run relationship among national
monetary policies is examined with the application of cointegration tech-
niques whereas the short-run monetary policy interactions are examined
using causality tests. Studies along this line of research can be classified
into two groups:
(a) Bilateral approaches look at pairs of countries that include Germany.
Examples include Giavazzi and Giovannini (1989), de Grauwe (1989), Kar-
fakis and Moschos (1990) and Katsimbris and Miller (1993).
(b) Multilateral approaches look at groups of three countries or more. For
example, Henry and Weidmann (1994) consider Germany, France and the
U.S., MacDonald and Taylor (1991) look at Germany, France and Italy
and Hafer and Kutan (1994) include in their analysis a group of five ERM
countries. Katsimbris and Miller (1993) look at three-country groups that
include the U.S. when performing causality tests.
Most of the above studies proxy monetary policy by some short-term interest
rate. Notable exceptions are MacDonald and Taylor (1991) who use money
supply and Hafer and Kutan (1994) that consider monetary base in addi-
tion to interest rates. The evidence in favour of monetary policy convergence
and German dominance is mixed. MacDonald and Taylor (1991), Hafer and
Kutan (1994) and Henry and Weidmann (1994) find some evidence support-
ing convergence. The GDH is supported by evidence provided by Karfakis
and Moschos (1990), MacDonald and Taylor (1991), Henry and Weidmann
(1994) and Hafer and Kutan (1994) when monetary policy is proxied by
monetary base. Henry and Weidmann (1994) find also that following the
German unification, the asymmetric nature of the ERM appears stronger.
Evidence against the GDH is supplied by de Grauwe (1989), Pratianni and
von Hagen (1990a, 1990b) and Katsimbris and Miller (1993).
For our purposes the most relevant of the above mentioned studies are Mac-
Donald and Taylor (1991) and Hafer and Kutan (1994). MacDonald and
Taylor (1991) use monthly money supply data for the period March 1979 to
December 1988 and test for convergence in an EMS group of countries and a
non-EMS group. The EMS group includes Germany, Prance and Italy. The
authors find only partial convergence of monetary policies but significant
evidence for the GDH.
Hafer and Kutan (1994) measure monetary policy using the proxies
of overnight interest rates and monetary base. Their sample includes monthly
data for the period March 1979 to December 1990 for Germany, France,
Belgium, Netherlands and Italy. The authors find evidence of less than
full convergence among the monetary policies of the five ERM countries for
both monetary policy proxies. They interpret this evidence to imply that
the GDH does not hold in the long run and "that monetary policies have not
been set totally independent from others' actions" (Hafer and Kutan, 1994,
p. 690). These authors test also for short-run monetary policy interactions.
Their Granger-causality test results performed in a VAR in levels imply that
the ERM has functioned as a symmetric system when monetary policy ac-
tions are captured by interest rates. However, significant evidence in favour
of asymmetry is provided when the tests are conducted using monetary base
which is perhaps a more accurate measurement of monetary policy actions.
The present paper improves upon Hafer and Kutan (1994) in several ways.
First, we extend the sample period to August 1992, second, we increase
the number of countries in our sample (by adding Denmark and Ireland)
and, third and more important, we split the sample into two periods. The
latter allows us to test for increasing convergence following the transition
from an unstable ERM period highlighted by several realignments to a more
stable period that lasted until August 1992. An important innovation of
the paper is also the consideration of small groups of countries in order to
identify which countries have experienced full monetary policy convergence
and hence could be part of the core in a likely two-speed monetary union.
3 Econometric Methodology
It is well known that although economic time series may wander through
time, i.e., be non stationary in their level, there may exist some linear com-
binations of these variables that will converge to a long-run relationship over
time (Engle and Granger, 1987). Therefore, if a linear combination of time
series that are stationary only after differencing is stationary, these series
are considered to be cointegrated. The concept of cointegration is used be-
cause if there has been convergence of monetary policy, then interest rates or
monetary base of EMS-member countries should move together over time.
Therefore, assuming that the monetary authorities try to control the move-
ment in short-term interest rates and that EMS countries are price takers
relative to Germany, the uncovered interest parity (UIP) is1 ,
where il is the domestic (non-German) interest rate, it is the German rate,
E is the expectations operator, A is the difference operator and S is the spot
exchange rate between the particular country and Germany. If the expected
exchange rate between the particular country and Germany is fixed, then the
expected change in the exchange rate is zero and the domestic interest rate
equals that of Germany. However, if the expected change in the exchange
rate is stationary under the adjustable peg system of ERM, then the above
result will also hold. Therefore, the two interest rates do not have to be
identical, but if they move together over time they are cointegrated.
The Engle and Granger (1987) tests for cointegration use bivariate relation-
ships, i.e., test for the existence of a single cointegrating vector between two
variables. However, problems arise with Engle and Granger (1987), when we
consider more than two variables, as it is no longer possible to demonstrate
the uniqueness of the cointegrating vector. Johansen (1988) and Johansen
and Juselius (1990) provide a method to investigate cointegration in a mul-
Tests for cointegration require nonstationary time series of the same order
of integration. Therefore, we first test for the presence of a unit root in
This part draws on Hafer and Kutan (1994, p. 686).
both the levels and the first difference of the interest rates and monetary
base using the Dickey-Puller (DF) and the Augmented Dickey-Fuller (ADF)
tests (Fuller, 1976 and Dickey and Fuller, 1979).
The method used to test for cointegration is the Johansen procedure intro-
duced in Johansen (1988) and extended in Johansen and Juselius (1990). If
there are p — 1 cointegrating vectors among p policy measures, then there
is only one common policy shared by all countries and so there is complete
long-run convergence of policies. If, however, the number of cointegrating
vectors is less than p— 1, but greater than one, then this implies that there is
some partial convergence of policies. Finally, if the number of cointegrating
vectors is zero, then this means that there are several (p) common trends
and so no long-run convergence of policies2. As can be seen from the above
procedure, cointegration does not automatically imply convergence.
Granger-causality tests (Granger, 1986; Engle and Granger, 1987) can be
used to detect if short-run changes in German policy alone are transmitted to
the other EMS countries, which is evident from unidirectional causation from
Germany to the other EMS countries, or whether there is general feedback
among policy actions. If the GDH is correct, then Granger-causality should
run from Germany to the other EMS countries but not vice versa. The
definition of Granger-causality is that if X causes Y, then using past values
of X will give improved predictions for Y (Harvey, 1981).
4 Data and Empirical Results
We use monthly data for the period March 1979 to August 1992. Our
proxies for monetary policy are short-term (overnight) interest rates and
monetary base3 . Since countries in the EMS express their short-run actions
through the resulting changes in the domestic money market interest rates,
this measurement is appropriate when looking at changes in monetary policy
in the short term. If, however, measures were to be used beyond this point,
2 A lack of cointegration should not necessarily imply that convergence did not take
place at the end of the period. In other words, if monetary policies become closely linked
at the end of the sample period, rather than during the period, our cointegration tests
will not capture such a change and hence policy convergence will be rejected.
3Three-month interest rates were used for Denmark. It was planned that money supply
(Ml) would also be included in the tests. However, due to autocorrelation even after
allowing for twelve lagged differences, and since unit root tests assume white noise errors
it became necessary to exclude Ml.
then the change in interest rates would incorporate the endogenous response
resulting from changes in economic activity and inflation rates. Interest
rates are taken from Eurostats, except for the Danish interest rates which
is taken from the Central Bank of Denmark. We choose monetary base
as a proxy for monetary policy because it reflects the actions being taken
by the Central Bank to affect reserves in the banking system and so some
monetary aggregates. The data for monetary base is taken from line 14
of the International Financial Statistics (IPS) data tape of the IMF. As
mentioned in the introduction, in order to capture the varying degree of
monetary policy coordination in the ERM, we split our sample into two
subsamples, the first one including data from March 1979 to January 1987
and the second covering the February 1987 to August 1992 period.
We use the natural logs for all series. As has been discussed, the first step
is to establish the order of integration. This is done using the Dickey Fuller
(DF) tests and the Augmented Dickey Fuller (ADF) tests with up to twelve
lagged differences. A deterministic trend was included in the test regression
whenever it was statistically significant4. The unit root tests for monetary
base are reported in Table I5 . We include results of the ADF(2) tests since
two is the lowest lag where serial correlation in the error term can be ruled
out6 . It is clear that all series are nonstationary. Having first differenced
the variables, all were universally stationary.
The results for the unit root tests for interest rates for period one and two
are shown in Tables 2 (a) and 2(b), respectively. A deterministic trend was
included in the test regression whenever it was statistically significant. It
is evident that all variables are 1(1). LM(12) autocorrelation tests (not
reported) imply white errors.
Having established that all variables are 1(1), we are now free to test for
multivariate cointegration. We have applied three alternative lag length-
selection criteria in order to choose the appropriate lag length in the VAR.
These include Sims's modified Likelihood ratio (LR) test, the Akaike Infor-
mation Criterion (AIC), and the Schwartz Bayesian Criterion (SBC). The
The majority of the monetary base variables are seen to have a trend while the ma-
jority of the interest rate variables do not have a trend.
The tests using monetary base cover the full sample period due to persisting autocor-
relation in each of the two sub samples when running the unit root tests.
The only exception is Belgium and Netherlands where we had to add more than twelve
lags in the ADF regression to eliminate serial correlation in the error term. The t-statistics
(not reported) imply a unit root.
application of all three criteria for interest rates (period one and two) and
monetary base (full period) led to a choice of 2 lags, i.e., k = 2.
Table 3 shows the cointegration test results for interest rates and monetary
base. Our cointegration tests assume trended variables. Let us look first at
interest rates for period one. The results for the maximum eigenvalue tests
indicate no cointegrating relationship. Since there are seven countries in the
sample, a finding of six cointegrating vectors would indicate one common
policy shared by all countries; i.e., full convergence of monetary policy. How-
ever, a finding of no cointegrating relationship, means that there has been
zero convergence of policy. This is a result that we would have expected in
this first period, which was characterised by lack of monetary policy coordi-
nation in the member countries. The trace test implies three cointegrating
vectors, that is, there are four common stochastic trends. This result would
suggest that there has been some partial convergence of monetary policy.
However, since the trace test lacks power relative to the maximum eigen-
value test (Johansen and Juselius, 1990), one would place more emphasis on
Table 3 also shows the result for interest rates in period two. Looking first
at the results of the maximum eigenvalue tests, we find that there is one
cointegrating relationship, that is, there are six common stochastic trends.
As had been expected, there was an improvement in the degree of policy
convergence in this second period, although it was very small. The trace
test found three cointegrating vectors, i.e., four common stochastic trends.
The final cointegration test uses monetary base as the proxy for monetary
policy and covers the full time period 1979.3-1992.8. Again the appropriate
lag length is set at k = 2. Looking first at the results for the maximum
eigenvalue tests, one can see that there are three cointegrating vectors and
hence four common stochastic trends. The difference in the results when
monetary base is used may be due to the fact that it is a more accurate
measure of monetary policy. However, it is unusual to find greater conver-
gence over the full time period than was found using interest rates for period
two (which was a more economically stable time). Three cointegrating vec-
tors were also found when the trace test was used to test for cointegration.
That is, there were four common stochastic trends.
Following the cointegration tests, each of the estimated differenced VARs
was checked for autocorrelation. The results for these tests are shown in
table 4. The P-Value of the LM(12) statistic is reported in each case. Auto-
correlation was found not to be a problem for interest rates for periods one
and two when k is set equal to 2. Finally, the VAR when using monetary
base for the full period was tested, and autocorrelation was found to be a
problem not only at lag 2, but also for lags up to 67. Hence, the cointe-
gration results of the VAR using monetary base should be considered with
We now turn to Granger-causality tests in order to check for short-run in-
teractions. Granger-causality tests will be used to detect if changes in Ger-
man monetary policy alone are transmitted to other EMS countries, which
is evident from unidirectional causation from Germany to the other EMS
countries or whether there is general feedback among policy actions. These
tests are performed in the VAR in first differences that includes an error-
correction term, when cointegration exists. Tables 5(a), 5(b) and 6 report
the F-statistics from the causality tests. The number of lags in all cases is
equal to 1 since by construction the number of lags in the differenced VAR
is one less than the number of lags in the VAR in levels.
Let us look first at interest rates for period one. The results of Table 5 (a)
show that Belgium is the only country which is statistically influenced in
the short term by changes in the German interest rates. They also show
that the Netherlands is a significant influence on German rates and Danish
rates cause Belgian and Dutch rates.
Table 5(b) shows the Granger-causality tests for interest rates for period two.
In this case not only does Germany cause Belgium, but also the Netherlands.
Both France and the Netherlands have a statistical influence on German
rates. A check on the estimated coefficients shows that even though bidi-
rectional causality applies between Germany and Netherlands, the effect of
Germany is much larger than the effect on Germany. Again it was found
that Italy's rates Granger caused Danish rates at the 10% level of signifi-
cance. Also included in table 5(b) are the t-values and significance levels for
the error-correction term (ECT). As can be seen, the only country which has
been significantly influenced by all the other countries (through the ECT)
in the sample has been Belgium.
Finally, the results for monetary base for the full period are shown in table
6. For monetary base, France, Italy, Netherlands, and Ireland and Denmark
(through the ECT), are the countries whose interest rates are Granger-
caused by Germany. French, German, Dutch, Danish and Irish interest
rates are influenced by all other interest rates through the ECT. However,
these results should be interpreted with caution since autocorrelation exists
in the VAR.
7Even though not reported in Table 4, the residuals in the regressions for Germany,
Ireland and Italy are also serially correlated.
Let us focus first on interest rates for period one. The results from the
maximum eigenvalue tests found no cointegrating relationship, a result con-
sistent with the exchange rate instability of that period. In their maximum
eigenvalue tests, Hafer and Kutan (1994) found two cointegrating vectors,
which meant there were three common trends. However, their sample ran
from March 1979 to December 1990, and included only five countries (Bel-
gium, France, Germany, Italy, and the Netherlands). Their results may show
greater convergence because of there being fewer countries and also because
their sample period is larger and captures better the long-run relationship
between the monetary policy proxies. The aim of our paper is to compare
the progress made on the convergence of monetary policy in two time peri-
ods which differ in the degree of exchange rate stability. The results from
our paper clearly show that convergence of interest rates did not take place
in this volatile first period.
The important question then is, was there an improvement in convergence
of monetary policy in the second period, which as has been mentioned was a
more stable period characterised by the complete absence of realignments?
According to the maximum eigenvalue test there was one cointegrating vec-
tor, i.e., six common stochastic trends. One would have expected greater
convergence in this second period, given the economic stability of the pe-
riod and also given Hafer and Kutan's (1994) results. The difference in the
results of Hafer and Kutan and our paper, may be due to the inclusion of
two extra countries (Ireland and Denmark) in our tests for cointegration.
Another reason for the difference in the results is that two relatively small
sample periods are used, and since cointegration is a long-run phenomenon,
this may have led to the finding of fewer cointegrating vectors.
Our tests for convergence include also monetary base for the full time period
March 1979 to August 1992. Both the maximum eigenvalue and trace tests
imply four cointegrating relationships. These results should be taken with
caution since autocorrelation was found to be a problem in the VAR.
The Granger-causality tests also give some interesting results. The only
country which was statistically influenced by Germany when using inter-
est rates for the first period is Belgium. Hafer and Kutan (1994) found
Belgium, France, and the Netherlands to be influenced by changes in the
German rate8 . Karfakis and Moschos (1990) found that the German rate
8Hafer and Kutan (1994) follow MacDonald and Taylor (1991) and use the VAR ex-
pressed in levels without an error-correction term to test for Granger-causality. This
approach is asymptotically equivalent to using first differences and an error-correction
term. Since our two sample periods are relatively small, we use first differences and the
influenced Belgium, Prance, Italy, and the Netherlands. Katsimbris and
Miller (1993) found that along with the above four countries, Germany also
Granger-caused the Irish rate. If we look at period two we see that not only
does Germany Granger-cause Belgium, but also the Netherlands. Given the
findings of previous papers, we would have expected to have found greater
causality running from Germany to the other countries9 . A result obtained
in this paper which complements the results of Hafer and Kutan (1994),
Karfakis and Moschos (1990), and Katsimbris and Miller (1993) is that the
Dutch rates Granger-cause the German rates. In our study this result applies
in both time periods. We also find that French rates Granger-cause German
rates in the second period. These results imply that the ERM has not func-
tioned as an asymmetric system and hence the GDH can be rejected. Our
results also show significant interaction among the countries in the sample.
Belgium, Denmark, and Italy Granger-caused the Dutch rate (as in Hafer
and Kutan, 1994), while Denmark and Germany cause the Belgian rate in
Our Granger-causality tests, when monetary base was used as a proxy for
monetary policy, show that Belgium was the only country which was not in-
fluenced by Germany. Hafer and Kutan (1994) found Germany to Granger-
cause France, Belgium and the Netherlands. Our results do show a high
degree of interdependence among monetary policies. This is so since mone-
tary base in each country (except Belgium and Italy) is Granger-caused by
monetary base in all other countries through the EOT.
5 Core countries and Monetary union
One of the issues under discussion in the agenda on European monetary
union is that of which countries will be part of the union at least at its
launch. Countries contemplating membership in a single currency should
have achieved full monetary policy convergence at a minimum10. Our re-
error-correction term to test for Granger-causality. Also, another possible reason for the
stronger evidence in favour of German influence provided by Hafer and Kutan (1994) is
the fact that they used more lags (four) in performing the Granger-causality tests.
To examine whether the difference in our results was in part due to much smaller
lags used in the VAR, we repeated our cointegration and Granger-causality tests using
two lags in the VAR (i.e., three lags in the levels VAR). It turns out that the number of
cointegrating vectors increases from one to two in the second period and the number of
countries Granger-caused by Germany increases from three to four.
Pull monetary policy convergence would be necessary for a common monetary policy
applied by the European Central Bank. The Maastricht convergence criteria specify ad-
ditional entry requirements that relate to fiscal policy, interest rates, inflation rates and
suits in the last section showed lack of full monetary policy convergence
among the seven countries of our sample. As has been discussed in the
literature, a two-speed monetary union is a likely solution to the issue of
incomplete degree of convergence among the member countries. Hence, it
would be interesting to test for the degree of convergence in a smaller group
of countries that could represent the core in a two-speed monetary union. In
this respect, we have applied the methodology employed earlier for smaller
groups of countries. Specifically, we used two 3-country groups (Germany-
Netherlands-Belgium and Germany-Netherlands-Prance), several 4-country
groups, and two groups of five countries that add Denmark and Italy, respec-
tively to the group including Germany, Netherlands, Belgium and France.
The only group of countries where full monetary policy convergence applies
is the three-country group (Germany-Netherlands-Belgium) for the second
period11 . Applying the AIC criterion, three lags were included in the VARs
in levels in both periods. Our cointegration results reported in Table 7 indi-
cate an increase from zero to two cointegrating vectors using both the trace
and maximum eigenvalue tests12 . The result of full monetary policy conver-
gence in the second period (i.e., a single common stochastic trend) implies
that Germany, Netherlands and Belgium could be considered as the first
group to join a monetary union under a two-speed approach. Somewhat
surprisingly, according to our results, France should not be considered one
of the core countries in a European monetary union, at least based on the
criterion of monetary policy convergence.
Table 8 reports the results of Granger causality tests in the three-country
VAR in first differences. In accordance with the cointegration tests, the
VAR in first differences includes 2 (= k — 1) lags for each variable. The
VAR for the second period includes also an EOT for the most significant
of the two cointegrating vectors (i.e., the one corresponding to the largest
eigenvalue). The cointegrating vectors are normalised such that a nega-
tive sign of the error-correction parameter implies adjustment to restore the
long-run equilibrium. The results of period one confirm policy asymmetry
as there is unidirectional causality from German interest rates to Belgian
11 In the other cases, even though the degree of convergence in the second period is
less than full, it is greater than that in the first period. Full convergence in the second
period also applies when our sample includes Germany, the Netherlands, Belgium and
Ireland. However, the estimated VAR suffers from serial correlation in the error term. The
cointegration results for these cases are not reported but are available from the authors
12The cointegration results in the three-variable and the seven-variable cases are almost
unchanged (in two cases the null can only be rejected at the 10% level) if the small-sample
adjustment to the critical values suggested by Cheung and Lai (1993) is made. According
to these authors, the Johansen asymptotic critical values have to be adjusted by a scale
factor given by T/(T-nk) where T is the effective sample size, n is the number of variables
and k is the number of lags in the VAR.
and Dutch rates. The system appears to be more symmetric in the second
period as German rates are also influenced by Belgian and Dutch rates (the
latter through the ECT). In other words, two-way causality applies between
Germany and each of Belgium and Netherlands in the second period. How-
ever, a look at the estimated coefficients shows that, as expected, the effect
of Germany exceeds the effect on Germany for each of these two countries.
The wrong sign of the statistically significant ECT in the equation for the
German rates implies that German interest rates do not adjust to restore
last period's deviation from the long-run equilibrium. Hence, the adjust-
ment to the long-run equilibrium takes place through changes in the Dutch
and Belgian interest rates.
6 Conclusions and suggestions for future research
This paper has used recent developments in the econometrics of nonstation-
ary time series to test for long-run interdependencies and short-run interac-
tions in the monetary policies of seven ERM countries. Our conclusions are
as follows: First, our results show that there has been very little progress
made on the issue of convergence of monetary policy within the EMS. Al-
though there was a slight improvement in the second period for interest
rates, our findings show that the member countries are still a long way off
full monetary policy convergence. This lack of monetary policy convergence
even during the second period of our study that ends in August 1992 may
have contributed to the breakdown of the system in September 1992. Second,
our Granger-causality tests do not support the GDH. Although Germany
does play an important role in the EMS, so do a number of other countries,
i.e., the Netherlands and France. As expected though, it is still the case that
whenever bi-directional causality applies, the influence of German interest
rates exceeds the influence on German interest rates.
Given the lack of full monetary policy convergence among the seven countries
of our sample, we have also tested for the degree of convergence in a smaller
group of countries that could represent the core in a two-speed monetary
union. Our results show that full convergence applies for the second period of
our study when the group of countries includes only Germany, Netherlands
and Belgium. This finding has important policy implications as it supports
those arguing in favour of monetary union which at the first stage would
include only countries in the core of the ERM. Further evidence is required
to determine whether extending the sample period to include the currency
crisis of 1992-93 would significantly alter the present results.
Unit Root Test Results
ADF (2) Statistics*
Monetary Base (Full Period)
Country Levels Differences
* The critical value for ADF(2) is -3.45 (Fuller, 1976).
**For Denmark a time trend is not included. The critical value is -2.89 (Fuller, 1976).