Journal Issue
Share
Article

Asymmetry in the ERM: A Case Study of French and German Interest Rates Before and After German Unification

Author(s):
International Monetary Fund. Research Dept.
Published Date:
January 1993
Share
  • ShareShare
Show Summary Details

THE ISSUE of German leadership in the European exchange rate mechanism has returned to the forefront of policy discussions following the recent European currency crisis, which saw the British pound and the Italian lira suspended as full members of the ERM. Some have suggested that the crisis was the product of German insensitivity to the policy requirements of other ERM members. Others have argued that the ERM is predicated on German monetary leadership and that it is unrealistic to expect credibility gains from a target zone system if the leader country is not allowed to adopt anti–inflationary policies appropriate to its own domestic situation. By examining daily interest rate data, we aim to shed light on what has actually happened to German leadership in the ERM over the past five years, focusing especially on the role that German unification may have played. Our main conclusion is that, in the year following unification. Germany largely lost its leadership role within the ERM. Since then, however, Germany has increasingly reasserted its predominance.

Various authors have characterized the ERM as a currency block dominated by German monetary policy. Giavazzi and Giovannini (1987) and Giavazzi and Pagano (1988) argue that the ERM evolved from the cooperative system that was originally intended into a system centered on the deutsche mark because of the desire of countries with poor records of inflation to profit from the credibility of German monetary policy. An alternative explanation for the perceived asymmetry in the functioning of the ERM was proposed by Wyplosz (1989a), who explains the leadership role of Germany in terms of an inherent bias in the system rather than the result of self–imposed constraints by ERM member countries. In his model, any fixed exchange rate regime produces this sort of bias given the bilateral nature of intervention, essentially because the country with the more restrictive monetary stance—that is, the one accumulating reserves—has a greater capacity for sterilized intervention than the one losing reserves. Russo and Tullio (1988) and Ungerer and others (1990) make similar arguments, suggesting that the rules of the ERM confer a central role on German policy, since reserve losses caused by interventions within the exchange rate band have generally obliged the weaker–currency country to adjust more than the stronger–currency one.

The view that German monetary policy dominates the ERM has found surprisingly little support in the empirical literature, however. Most empirical studies suggest that Germany is an important player in the ERM, but that German monetary policy is also affected by innovations in other ERM member countries. For example, Cohen and Wyplosz (1989) and De Grauwe (1989) test for German leadership through simple Granger causality tests applied to changes in national interest rates and monetary aggregates in ERM countries. Both studies conclude that the asymmetry in the ERM has been much weaker than generally thought, although De Grauwe also finds strong evidence of German leadership in offshore markets, based on the response of Euromarket interest rates to changes in the forward exchange rate premium vis–à–vis Germany.

However, as Weber (1990) and De Grauwe (1989) point out, Granger causality tests are of limited significance when policy response is rapid, since they fail to capture contemporaneous “causality.” Moreover, Weber noted that causality tests applied to monetary aggregates are likely to be distorted by the sterilization of foreign exchange interventions: any European Monetary System (EMS) country that intervenes using deutsche marks to support its own currency might appear to “cause” changes in German monetary aggregates simply because of the time it takes for Germany to sterilize.

Another way to analyze the joint behavior of innovations in interest rates and monetary aggregates in the ERM, one which does not suffer from the problems with Granger causality tests, is to estimate a system of equations in which contemporaneous linkages between interest rates are identified on the basis of an underlying structural model, which is more or less explicitly derived from a central bank reaction function. Fratianni and von Hagen (1990) and von Hagen and Fratianni (1990) adopt this approach using money supply and interest rate innovations, respectively. Their results generally confirm the conclusions of the other studies: although changes in German policy have a strong impact on other ERM members, Germany itself is not immune to innovations in other countries. In particular, interest rate changes in the Netherlands, France, and Italy appear to have had a strong effect on German rates at different times during the 1979–88 period. However, the study based on changes in base money found that the effect on Germany of innovations in other ERM countries is only temporary.

Artus and others (1991) estimate a more general model of interest rate determination for France and Germany that allows for the effects of changes in long–term interest rates and in exchange rates. They find strong evidence of asymmetry, with German short–term interest rates reacting mostly to U.S. interest rates and the mark–dollar exchange rate and with French short–term rates reacting to German interest rates. the mark–franc exchange rate, and France’s current account. Long–term interest rates were found to he weakly related to short–term rates but strongly related to foreign long–term rates.

On balance, it appears that the hypothesis of complete German dominance of the ERM (that is, unidirectional causality) is easily rejected by the data, while weaker notions of German leadership are not. In some ways, this is not surprising since, as Wyplosz (1989b) points out, even if Germany is acting as the leader in some game theoretic ERM equilibrium, it could still be optimal for Germany to take other countries’ monetary actions into account. Thus, in many ways, it is more interesting to explore the magnitude of the asymmetry in the ERM and how this has evolved over time rather than to concentrate on some narrowly defined statistical question about whether or not dominance holds.

Finally, two other approaches to analyzing asymmetry in the ERM deserve mention. First, in an interesting study, Mastropasqua, Micossi. and Rinaldi (1988) use information on foreign exchange interventions and sterilizations to develop an alternative test of monetary independence. Under the hypothesis of monetary independence, changes in a member country’s net foreign assets position that are related to foreign exchange intervention using its currency should be fully sterilized. For the period 1979 to 1987, they find sterilization to be incomplete in three of the four ERM countries considered—the exception was Germany, for which the hypothesis of full sterilization after three months could not be rejected. They also observed that, although Germany was responsible for nearly all net sales of dollars over the 1979–87 period, it undertook almost no interventions in EMS currencies. This provides some support for German independence, since Germany appears to hold responsibility for the position of the ERM block relative to non–ERM currencies, but does not concern itself with the relative position of exchange rates within the band.

Second, in a recent study, Koedijk and Kool (1992) test whether the ERM has acted as a deutsche mark zone by applying a principal components analysis to interest rate and inflation differentials within the EMS, including the United Kingdom, from 1979 to 1989. Specifically, they investigate whether dominant movements in bilateral interest rate and inflation differentials could be attributed to a specific country or group of countries. Their study finds evidence of persistent independent interest rate and inflation differentials in the EMS originating from the independent movements in two currency blocks: Germany, the Netherlands, and the United Kingdom on one side, and France. Italy, and Belgium on the other. The independent movement of Irish interest rates was another important factor contributing to the overall variance. Although the authors conclude that the EMS has not functioned as a mark zone, it seems more appropriate to say that their analysis only rejects the restrictive hypothesis of unilateral German causality, not necessarily that of German leadership.

I. Data and Sample Period

We investigate the issue of German leadership using changes in one–month French and German onshore and offshore interest rates sampled daily from October 1987 to August 1992. We choose to restrict our sample to the co–movement of these two rates in order to present a fuller statistical analysis than would be possible in a broader system. We also believe that our choice of interest rates provides a sound basis for testing the hypothesis of German leadership. First, France, since 1987, has been one of the most vocal EMS members in calling for greater symmetry in the ERM and claiming a greater role for itself. Second, other ERM members either possess financial markets too small to influence those of Germany or France or have already openly accepted German monetary leadership. The obvious exception is the United Kingdom, which, however, only entered the ERM in October of 1990, and then only with a much wider fluctuation margin.1

The use of daily data allows us to look for the presence of regime shifts in the late 1980s and early 1990s. In particular, we look for a structural break around the time of German unification. Such regime shifts are possible because the budgetary and monetary strains caused by unification may have weakened Germany’s anti–inflationary resolve, and thus eroded the leadership role of Germany in the ERM. We do not identify a breaking point in the data with any particular historical or news event, since movement toward German unification gained strength over several months.2 Rather, we choose to break our sample at the end of December 1989, when the turmoil of unification brought the deutsche mark under pressure in exchange markets.

The nominal convergence toward German rates achieved by some ERM participants, notably France, over our sample period is viewed by some as another possible explanation for the erosion of German leadership after 1990. In our view, this development should not by itself lead to a regime shift. The improvement of inflation performance in France increased the credibility of French macroeconomic policy, and thus contributed to the reduction of interest rate differentials with Germany. However, increased credibility does not immediately confer greater independence; after all, the policy objectives of France did not change.

An important consideration in our selection of a sample period was to ensure that, unification apart, it included no obvious regime switches. Two points should he noted in this regard. First, our sample period begins after the ERM realignment of January 1987, in which the central parities of the deutsche mark and the Netherlands guilder were revalued by 3 percent and that of the Belgian franc by 2 percent. After the realignment, French authorities adopted what became known as the “competitive disinflation” strategy, by which the competitiveness of the economy was to be restored by lowering its inflation rate below that of its ERM partners, rather than resorting to further devaluations vis-à-vis the mark and the core of stronger currencies. Because it precluded further devaluations vis-à-vis the deutsche mark, this strategy could have constituted a regime shift.

Second, the chosen sample period coincides with the new ERM rules of intervention and policy coordination formalized in the Basle–Nyborg agreement of September 1987. The Basle–Nyborg agreement represented a significant change in the rules regulating intervention within the ERM. The agreement permitted intramarginal intervention to be financed for the first time through the very short–term financing facility (VSTFF), a network of mutual credit lines between participating central banks. The agreement also stressed that greater fluctuation of exchange rates should be allowed within the band and that interest rate differentials should be used more aggressively to defend exchange rate parities. Under the original rules of the ERM, access to the VSTFF was limited to interventions at the margin of the band. Since most interventions occurred intramarginally, the actions of the intervening country had no direct impact on the balance sheet of other central banks.3

Given the above discussion, we think it reasonable to regard the behavior of interest rates in our sample period as homogeneous, apart from the shock of German unification.4 The high frequency of the sample also distinguishes our study from the rest of the literature in that it permits us to detect dynamics in the data when policy response is very rapid. In fact, with the gradual dismantling of capital controls over our sample period, lags in the response of interest rates to foreign innovations are likely to have been reduced to a few days, at most. In this context, as mentioned above, statistical causality tests based on monthly observation are likely to lose too much information to be meaningful.

There is wide disagreement in the literature over the merits of using onshore versus offshore interest rates for empirical testing. In our study, we use both. The offshore rates employed consist of one–month Euromarket deposit rates (Figure 1), while the onshore rates are domestic one–month interbank rates.5 A potential disadvantage with the use of onshore money market rates is that they are likely to be contaminated by domestic developments related to reserve requirements and other institutional factors.6 The higher degree of autocorrelation present in onshore interest rate data is an indication of this problem (Table 1).

The problem with offshore rates, by contrast, is that they may not fully reflect domestic monetary policy actions if capital controls are present. In the case of France, capital controls were in effect from 1987 through 1989, although they do not appear to have insulated the domestic market from foreign innovations to any significant degree. As shown in Figure 2, the standard deviation of the differential between French offshore and onshore rates declined after the removal of capital controls, but had been relatively small even when controls existed (except for the end of 1987) if compared with the equivalent German differential. The offshore–onshore differential rarely rose above 20 basis points for France, whereas persistent deviations of that magnitude are observed for Germany. The lack of segmentation between the French onshore and offshore markets is also confirmed by Weber (1990), who finds that, over the 1983–89 period, Granger causality ran clearly from French offshore rates to French onshore rates.

The descriptive statistics reported in Table 1 reveal some more interesting information. First, the contemporaneous cross–correlation between France and Germany in the onshore markets is quite strong before January 1, 1990 (correlation coefficient of 0.57) but weakens substantially thereafter (0.09). By contrast, the correlation coefficient in the offshore market remains quite stable over the two periods (0.15 and 0.16, respectively). Second, the contemporaneous correlation between offshore and onshore rates is weaker for Germany than for France in both subperiods. Again, these observations raise doubts about the effectiveness of capital controls in insulating French monetary policy from external innovations before 1990. Admittedly, the correlation between onshore and offshore rates rises in France after 1990, but it does so to an even greater extent in Germany.

Figure 1.Offshore Interest Rates, One–Month Maturities, 1987–92

Table 1.Cross–Correlation Matrix of Interest Rate Changes
ΔrFΔrGΔrUSΔrFNatΔrGNat
Whole sample, October 1987–August 1992
ΔrF1.000.150.150.310.06
ΔrG0.151.000.350.100.18
ΔrUS0.150.351.000.070.17
ΔrFNat0.310.100.071.000.40
ΔrGNat0.060.180.170.401.00
Pre–unification, October 1987–December 1989
ΔrF1.000.150.210.270.07
ΔrG0.151.000.240.110.15
ΔrUS0.210.241.000.100.18
ΔrFNat0.270.110.101.000.57
ΔrGNat0.070.150.180.571.00
Post–unification, January 1990–August 1992
ΔrF1.000.160.080.380.05
ΔrG0.161.000.500.090.29
ΔrUS0.080.501.000.030.18
ΔrFNat0.380.090.031.000.09
ΔrGNat0.050.290.180.091.00

Note: The variables ΔrF, ΔrG, and ΔrUS are changes in offshore rates. The variables ΔrFNat and ΔrGNat are changes in onshore rates.

Finally, the cross–autocorrelations (Table 2) between onshore and offshore rates as well as between French and German rates do not reveal the presence of any obvious one–day temporal causality in the data, since the off–diagonal elements of the cross–autocorrelation matrix are all relatively similar (and also relatively small): in other words, the correlation between the lagged change in German rates and the current change in French rates is similar to the correlation between the lagged change in French rates and the current change in German rates.

II. Estimation and Identification of Vector Autoregression (VAR)

Consider a three–dimensional vector of short–term interest rate changes for France, Germany, and the United States. In the model we estimate, this vector is regressed on cross–country contemporaneous interest rate changes, five lags of the vector itself, and five lags of changes in comparable long–term interest rates for each country. In formulating a linear model of this kind, we ignore possible nonlinearities owing to “band effects” (see Krugman (1991) and Bertola and Svensson (1991)). Modeling interest rates with such effects explicitly accounted for is quite difficult. Recent empirical work (see, for example, Lindberg and Soderlind (1992)) suggests that large–scale intramarginal intervention within the band by central banks makes such nonlinearities relatively unimportant, and Svensson (1992) argues that it is therefore legitimate to approximate a target zone using a linear model of a managed float. The approach in this paper can be justified in a similar manner.7

Figure 2.Offshore Interest Rate Differentials, 1987–92

Table 2.Cross–Awocorrelation Matrix of Interest Rate Changes
ΔrFΔrGΔrUSΔrFNatΔrGNat
Whole sample, October 1987–August 1992
Lag ΔrF0.000.050.040.180.04
Lag ΔrG0.07–0.100.040.050.14
Lag ΔrUS0.030.09–0.020.040.07
Lag ΔrFNat0.180.04–0.00–0.12–0.12
Lag ΔrGNat0.110.10–0.01–0.16–0.27
Pre–unification, October 1987–December 1989
Lag ΔrF0.050.050.050.220.04
Lag ΔrG0.08–0.200.000.070.10
Lag ΔrUS0.070.110.050.030.07
Lag ΔrFNat0.190.020.00–0.12–0.20
Lag ΔrGNat0.120.090.01–0.26–0.33
Post–unification, January 1990–August 1992
Lag ΔrF–0.060.050.030.130.05
Lag ΔrG0.050.060.090.040.26
Lag ΔrUS–0.030.07–0.070.040.07
Lag ΔrFNat0.160.07–0.02–0.130.04
Lag ΔrGNat0.090.16–0.060.03–0.01

Note: The variables ΔrF, ΔrG, and ΔrUS are changes in offshore rates. The variables ΔrFNat and ΔrGNat are changes in onshore rates. Entries are covariances with lagged variables scaled by standard deviations of the two series.

To identify the model statistically, we assume (1) that German and French short–term interest rates are not directly affected by each others’ long–term interest rates (exclusion restrictions), (2) that U.S. interest rates are not affected by changes in French or German rates at any lag (exclusion restrictions), and (3) that the covariance matrix of innovations to the system are orthogonal instantaneously (covariance restrictions). These assumptions imply that the model is overidentified.8 Assumption 1 implies that short–term rates react to lagged changes in the other country’s long–term rates only through the induced change in that country’s short–term interest rate. It is reasonable to suppose that investors arbitrage along yield curves in each country or between similar–maturity bonds in different currencies. Our restriction amounts to assuming that links between the long and short ends of the yield curves in different currencies are weak. Assumption 2 means that U.S. interest rates are taken as predetermined and serves to simplify the system and reduce the number of estimated parameters. Assumption 3 implies that all instantaneous cross–correlation between short–term interest rates occurs through the matrix of coefficients on contemporaneous interest rate changes. Define Xt(XtF\XtG\XtUS)as a three–dimensional vector of changes in French, German, and U.S. short–term interest rates and Yt(XtF\YtG)as a two–dimensional vector of changes in French and German long rates. The model we estimate is then of the form

where Σ*cov(ϵt) is assumed to be diagonal.

Estimation is carried out using the generalized method of moments (GMM).9 The descriptive statistics suggest that interest rate changes are extremely leptokurtic, suggesting that maximum–likelihood estimation based on normal distributions is inadvisable and that a more robust estimation method, such as GMM, is to be preferred.10 The models are each estimated initially using an arbitrary weighting matrix. The resulting consistent parameter estimates are then used to construct an optimal weighting matrix based on the Newey–West approach to covariance matrix estimation.11The latter is then employed in a second iteration of GMM to obtain asymptotically efficient parameter estimates.

In our initial estimations, we also included three dummy variables that took the value of 1 if one of the markets in question had been closed for one, two, or three or more days, respectively, preceding a given observation; otherwise the dummies were zero. Most studies of financial market data ignore weekends and holidays on the presumption that what matters is some notion of “economic time.” Since these dummies proved insignificant, they were dropped from the version of the regressions actually reported.

Rather than looking at the statistical significance of the regression coefficients, either individually or in groups, we regard it as more interesting to focus on the significance, both economic and statistical, of the long–run multipliers implied by the regression equations. Such long–run multipliers take into account feedback both within a given equation and within the system as a whole. To calculate them, one must convert the system into a VAR of order 1 of the form

where I is a three–by–three identity matrix, 0 is a three–by–three matrix of zeros, and where ϵt(IA)1ϵt. If one denotes the coefficient matrix on the right side of the above equation by Φ, then the long–run multipliers for shocks to the different interest rate are given by the upper three rows of (I15Φ)1[(IA)1|0...|0], where I15 is a 15–dimensional identity matrix. We also calculate simpler equation–by–equation multipliers of the form (ajk+Σi=15bjki)/(1Σi=15bjji). The results based on these multipliers are not substantially different and are not reported here.12 Using the fact that these multipliers are complicated, nonlinear functions of the parameters, their standard errors, and t–statistics are calculated based on the covariance matrix of the parameters. The latter equals inv[(q/θ)Σa(q/θ)], where q/θ is the derivative of the sample–averaged moment conditions with respect to the parameters and Σq is a Newey–West kernel estimate of the covariance matrix of the moment conditions.

Table 3.System–Wide Long–Run Effects: Offshore Rates
MultipliersStandard errorsT–statistics
ΔrFΔrGΔrUSΔrFΔrGΔrUSΔrFΔrGΔrUS
Whole sample
 ΔrF1.070.210.050.060.080.1217.442.730.44
 ΔrG0.160.730.270.070.070.032.3010.958.43
 ΔrUS0.970.0333.51
Pre–unification
 ΔrF1.100.270.050.090.140.0711.771.920.74
 ΔrG0.110.820.310.090.100.041.177.857.74
 ΔrUS0.980.0334.96
Post–unification
 ΔrF0.950.210.020.080.100.2011.662.120.11
 ΔrG0.180.810.200.050.070.053.4912.264.12
 ΔrUS0.990.0424.59
Post–unification, 1990
 ΔrF0.930.150.040.070.110.0812.731.290.57
 ΔrG0.250.800.210.120.050.042.0615.575.26
 ΔrUS0.980.0812.99
Post–unification, 1991
 ΔrF0.990.22–0.060.080.160.0611.901.39–0.89
 ΔrG0.180.800.190.090.080.052.0210.454.25
 ΔrUS1.010.0812.00
Post–unification, January–August 1992
 ΔrF0.940.15–0.000.070.150.0612.551.01–0.08
 ΔrG0.180.800.190.080.080.042.1310.444.18
 ΔrUS1.000.0812.15

Note: The column stub indicates the origin of the unit shock and the row stub shows the interest rate affected.

III. VAR Estimates and Impulse Effects

Tables 3 and 4 and Figures 36 show the long–run effects of unit shocks to offshore and onshore French, German, and U.S. short–term interest rates, taking into account feedback effects across equations, as discussed above.13 To start, the long–run multipliers for offshore interest rates clearly reject the hypothesis of unidirectional causality for the whole sample (see Table 3). The effect of French innovations on Germany is significant, albeit smaller than the German effect on France. However, the results strongly suggest the presence of a structural break coinciding with German unification. In the pre–unification period (1987 through 1989), the German multiplier for France is 0.27 compared with a French multiplier for Germany of 0.11. After unification, the corresponding figures are 0.21 and 0.18. Examining the year–by–year estimation results, we find that the post–unification period is very uneven, with Germany losing its leadership role entirely during 1990 but recovering it thereafter.

Table 4.System–Wide Long–Rem Effects: Onshore Rates
MultipliersStandard errorsT–statistics
ΔrFΔrGΔrUSΔrFΔrGΔrUSΔrFΔrGΔrUS
Whole sample
ΔrF1.220.34–0.040.190.170.146.421.97–0.29
ΔrG0.291.220.160.150.210.061.905.712.53
ΔrUS0.920.0335.46
Pre–unification
ΔrF1.400.43–0.070.150.130.069.643.20–1.20
ΔrG0.321.160.170.150.180.062.096.362.98
ΔrUS0.930.0238.02
Post–unification
ΔrF1.020.280.200.340.150.183.001.851.10
ΔrG0.131.050.170.090.130.041.398.164.27
ΔrUS0.910.0425.25
Post–unification, 1990
ΔrF0.890.040.100.070.070.0613.480.591.69
ΔrG0.220.810.190.130.050.061.6415.802.96
ΔrUS1.000.1010.18
Post–unification, 1991
ΔrF0.970.200.200.190.214.841.030.03
ΔrG0.120.750.200.070.080.041.639.185.65
ΔrUS1.020.0521.16
Post–unification, January–August 1992
ΔrF0.980.29–0.070.160.220.206.241.33–0.33
ΔrG0.170.820.180.050.060.043.3414.424.73
ΔrUS0.980.0518.65

Note: The column stub indicates the origin of the unit shock and the row stub shows the interest rate affected.

Figure 3.Effect on Short–Term Interest Rates of a Unit Shock to Offshore Rates, Before and After German Unification

Figure 4.Effect on Short–Term Interest Rates of a Unit Shock to Offshore Rates, Whole Sample Versus 1990

Figure 5.Effect on Short–Term Interest Rates of a Unit Shock to Onshore Rates, Before and After German Unification

Figure 6.Effect on Short–Term Interest Rates of a Unit Shock to Onshore Rates, Whole Sample Versus 1990

Only innovations in U.S. interest rates affect German offshore rates to any significant degree. A unit shock to U.S. rates leads to a rise of 27 basis points in German offshore rates in the long run. The direct and feedback effects of changes in U.S. rates on France are extremely small and not statistically different from zero.

The multipliers based on onshore interest rates give similar results, although they display more symmetry in the pre–unification period and greater asymmetry in 1991–92 (Table 4). This difference could be related to the presence of capital controls in the pre–unification period, which somewhat insulated French domestic interest rates from foreign innovations. With their elimination at the beginning of 1990, national markets have become more integrated, and, possibly because national markets reflect more closely the monetary authorities’ actions, they may also have become more responsive than Euro markets to each other’s innovations.

The long–term response of interest rates to a unit shock to the same rate varies across equations and often differs from unity. For instance, German offshore rates show a clear mean–reverting tendency when compared with onshore German rates and French rates. Based on equation-by-equation multipliers, which do not take into account cross–equation feedback effects, a unit shock to German Euro–rates translates into a permanent increase in the rate of only 70 basis points.14

Monetary policy actions, however, cannot easily be described in terms of unit shocks. A more realistic description of a monetary policy change is a one–time permanent change in short–term interest rates. In Table 5, we therefore report the effects of unit permanent changes in French and German interest rates on each other, based on the estimated coefficients in Tables 3 and 4. This presentation also allows us to measure cross–effects based on the same monetary action, rather than on the same temporary shock in the two equations. The results generally confirm the conclusions drawn above: in the pre–unification period, German monetary policy actions have a stronger effect on France than French actions do on Germany, although to a significant degree only for offshore rates. In the first year of German unification, the roles are reversed and France assumes the leadership role, particularly for onshore rates. However, Germany appears to regain a stronger role in 1991–92.

Table 5.Impact of Unit Permanent Change in French and German Rates
Offshore ratesOnshore rates
SampleImpact of ΔrF on ΔrGImpact of ΔrG on ΔrFImpact of ΔrF on ΔrGImpact of ΔrF on ΔrG
Whole sample0.150.290.240.28
(0.07)(0.10)(0.12)(0.12)
Pre–unification0.100.330.230.37
(0.08)(0.16)(0.10)(0.10)
Post–unification0.190.260.130.26
(0.06)(0.12)(0.08)(0.14)
January–December 19900.270.180.240.05
(0.14)(0.14)(0.16)(0.08)
January–December 19910.180.270.120.26
(0.09)(0.19)(0.06)(0.24)
January–August 19920.190.190.170.35
(0.09)(0.18)(0.05)(0.25)

Notes: Entries equal ratios of cross to own long–run multipliers. Standard deviations are in parenthesis.

IV. Tests for Structural Instability

Table 6 gives Lagrange multiplier tests for structural breaks in the sample. The tests account for the fact that the timing of the break point is unknown. The statistics given represent the maximum of a series of Lagrange multiplier statistics calculated for various alternative break points. This approach to testing for breaks of unknown date has been in general use for nearly 30 years, but it is only recently that the true distribution of the test statistic has been established. Andrews (1990) and Hansen (1990) derive the distribution of such “Lagrange multiplier supremum” statistics.

Since our hypothesis is that German unification led to a major change in behavior followed by a gradual return to normality, we consider two samples—October 1987 to December 1990 and January 1990 to August 1992—and test for structural breaks of unknown date in each. We implement the tests by calculating the statistics for a range of break points at every 20 observations to each side of end–December 1989 and end–December 1990 for the two samples respectively. In each case, we extend the range of possible break points to cover most of the sample in question. We then take the maximum value of the calculated statistics and compare it with the significance levels given in Andrews (1990) and Hansen (1990).

The results suggest that a big change did occur at the end of 1989. The Lagrange multiplier tests for the October 1987 to December 1990 period are all significant at a 5 percent level except for one.15 All the break points suggested by the significance tests are within 20 observations of the start of January 1990.

Table 6.Lagrange Multiplier Tests for Structural Change
Period 1,

October 1987–

December 1990
Period 2,

January 1990–

August 1992


Test
Degree of

freedom


Statistic


Date


Statistic


Date
1. All parameters49152*3/10/89287*9/11/90
2. German–French effects2444*30/1/902911/10/90
3. Germany on France69*2/1/90412/9/90
4. France on Germany611*30/11/8949/11/90
5. Total U.S. effect25166*30/1/902811/10/90
6. U.S. effect on France and Germany101526/4/90512/9/90

Notes: An asterisk indicates significance at 5 percent significance level. Dates represent the sample breaks that give the maximum Lagrange multiplier statistics. Test 2 concerns all French and German parameters. Tests 3 and 4 concern parameters for the impact of German rates on French rates and vice versa.

Tests for the later sample of January 1990 to August 1992 show much less evidence of a regime break, although the test for a change in all the parameters is significant at a 5 percent level. The date around which this switch seems to have occurred is February 1991. This suggests that while there is some return to the situation prevailing before December 1989, the change itself is less pronounced and it is less clearly identified with a given moment in time.

REFERENCES

    Andrews, Donald W.K., “Tests for Parameter Instability and Structural Change with Unknown Change Point,”Cowles Foundation Discussion Paper No. 943 (Cowles Foundation, 1990).

    Artus, P., and others, “Transmission of U.S. Monetary Policy to Europe and Asymmetry in the European Monetary System,”European Economic Review,Vol. 35 (October1991), pp. 136984.

    Bertola, Guiseppe, and Lars E.O. Svensson, “Stochastic Devaluation Risk and the Empirical Fit of Target Zone Models,”NBER Working Paper No. 3576 (Cambridge, Mass,: National Bureau of Economic Research, 1991).

    Cohen, D. and C.Wyplosz, “The European Monetary Union: An Agnostic Evaluation,” inMacroeconomic Policies in an Interdependent World,ed. byRalph C.Bryant and others (Washington: International Monetary Fund, 1989).

    De Grauwe, Paul, “Is the European Monetary System a DM–Zone?”CEPR Discussion Paper No. 297 (London: Centre for Economic Policy Research, 1989).

    Fratianni, Michele, and Jurgenvon Hagen, “German Dominance in the EMS: The Empirical Evidence,”Open Economies Review,Vol. 1 (1990), pp. 6787.

    Gallant, A. Ronald, Nonlinear Statistical Models (New York: Wiley, 1987).

    Gardner, Edward H., and William R.M.Perraudin, “Asymmetry in the ERM: A Case Study of French and German Interest Rates Since Basle–Nyborg,”IMF Working Paper 92196 (Washington: International Monetary Fund, 1992).

    Giavazzi, Francesco, and AlbertoGiovannini, “Models of the EMS: Is Europe a Greater Deutsche Mark Area?” inGlobal Macroeconomics,ed. byRalph C.Bryant and RichardPortes(New York: St. Martin’s Press, 1987).

    Giavazzi, Francesco, and MarcoPagano, “The Advantage of Tying One’s Hands: EMS Discipline and Central Bank Credibility,”European Economic Review,Vol. 32 (June1988), pp. 105575.

    Hansen, B.E., “Lagrange Multiplier Tests for Parameter Instability in Non– Linear Models” (unpublished; University of Rochester, 1990).

    Hausman, Jerry A., and WilliamE. Taylor, “Identification in Linear Simultaneous Equations Models with Covariance Restrictions: An Instrumental Variables Interpretation,”Econornetrica,Vol. 51 (September1983), pp. 152749.

    Hausman, Jerry A., and others, “Efficient Estimation and Identification of Simultaneous Equations Models with Covariance Restrictions,”Econometrica,Vol. 55 (July1987), pp. 84974.

    Koedijk, K.G., and C.J.M.Kool, “Dominant Interest and Inflation Differentials within the EMS,”European Economic Review.Vol. 36 (1992), pp. 92543.

    Krugman, Paul R., “Target Zones and Exchange Rate Dynamics,”Quarterly Journal of Economics,Vol. 106 (August1991), pp. 66982.

    Lindberg, Hans, and PaulSoderlind, “Target Zone Models and Intervention Policy: The Swedish Case,”TIES Seminar PaperNo. 496 (Stockholm: Institute for International Economic Studies, 1992).

    Lipschitz, Leslie. and DonoghMcDonald (eds.), German Unification: Economic Issues,IMF Occasional PaperNo. 75 (Washington: International Monetary Fund, 1990).

    Mastropasqua, C., S. Micossi, and R.Rinaldi, “Intervention, Sterilization and Monetary Policy in EMS Countries, 1979–87,” inThe European Monetary System,ed. byFrancesco Giavazzi, S. Micossi, and M.Miller (Cambridge: Cambridge University Press, 1988).

    Newey, W.K., and K.D.West, “A Simple Positive Semi–Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix.”Econometrica,Vol. 55 (1987), pp. 70308.

    Russo, Massimo, and GuiseppeTullio, Monetary Coordination Within the European Monetary System: Is There a Rule?IMF Occasional Paper No. 61 (Washington: International Monetary Fund, 1988).

    Svensson, Lars E.G., “Monetary Independence and Optimal Intervention Policy in a Managed–Float Model of an Exchange Rate Target Zone” (unpublished; University of Stockholm, 1992).

    Ungerer, Hurst, and others, The European Monetary System: Developments and Perspectives,IMF Occasional Paper No. 73 (Washington: International Monetary Fund, 1990).

    Von Hagen, Jurgen, and MicheleFratianni, “German Dominance in the EMS: Evidence from Interest Rates,”Journal of International Money and Finance,Vol. 9 (1990), pp. 35875.

    Weber, Axel A., “EMU and Asymmetries and Adjustment Problems in the EMS: Some Empirical Evidence,”CEPR Discussion Paper Series No. 448 (London: Centre for European Policy Research, 1990).

    Wyplosz, Charles(1989a), “Asymmetry in the EMS: Intentional or Systemic?”European Economic Review,Vol. 33 (1989), pp. 31020.

    Wyplosz, Charles (1989b), “EMS Puzzles” (unpublished, 1989),

Edward H. Gardner is an Economist in the European I Department and holds a Ph.D. from the Massachusetts Institute of Technology. William R.M. Perraudin is a Fellow of Gonville and Caius College, Cambridge University, and of the Centre for Economic Policy Research and holds a Ph.D. from Harvard University. The paper was written, in part. while Mr. Perraudin was a Visiting Scholar in the European I Department. The authors thank Leonardo Bartolini and Michael Deppler for helpful comments.

The range of permitted fluctuation above and below the central rates over our sample period was 6 percent for the British pound and the Spanish peseta and 2.25 percent for the other participating currencies. The Spanish peseta only entered into the ERM in June 1989.

The first impulse to German unification came from the mass emigration from the German Democratic Republic (GDR) to West Germany that followed the opening of the border between Hungary and Austria in September 1989, With the opening of the border between the two Germanies in November of the same year, the volume of people moving from the GDR rose to massive proportions, bringing into question the viability of the GDR as a separate nation. A currency union between the two Germanies was proposed by Chancellor Helmut Kohl in February 1990. and the final terms of unification were negotiated in April and May 1990. For a further discussion of these events and their economic effects, see Lipschitz and McDonald (1990),

The Basle–Nyborg agreement caused some concern in Germany that the obligation to finance intramarginal interventions could lead to the creation of excessive liquidity by the Bundesbank, thereby undermining the anti–inflationary stance of German monetary policy. Several factors reduced this risk, however. First, intramarginal intervention using a partner’s currency still requires the prior approval of the central bank issuing the intervention currency, and, second, the amounts involved are small relative to the total monetary base. Moreover. with most central banks using interest rates as intermediate targets, the monetary effects of intervention tend to be automatically sterilized. Following the ERM crisis of September 1992, Germany. for the first time. intervened intramarginally in support of another ERM currency, the French franc. This is more likely to reflect the gravity of the strains within the ERM at the time than a change in Germany’s role in the ERM.

Of course, German unification occurred at the same time as Eastern Europe as a whole began its process of reform. Flows of net investment to Eastern Europe, except to the former East Germany. have remained fairly negligible, however, and it is hard to believe that the more general reform process has significantly affected monetary events in the Western economies.

Interbank rates have the advantage that their tax–exempt status insulates them from the effect of changes in taxation.

For example, regular movements in rates associated with the cycles of bank reserve accounting introduce negative autocorrelation that has nothing to do with the dynamics of genuine shifts in monetary policy, obscuring actual policy changes.

One might, nevertheless, argue that we should include lagged exchange rate changes in our regressions. Our initial estimations did include such lags, but they had no significant explanatory power so we felt justified in omitting them.

Assumptions 1 and 2 would be sufficient to identify all but one of the parameters. For discussions of identification in linear models with covariance restrictions, see Hausman and Taylor (1983) and Hausman, Newey, and Taylor (1987).

See, for example, Gallant (1987).

Onshore rate changes exhibit kurtosis ranging up to almost 200, compared with the kurtosis of any normal random variable of 3.

See Newey and West (1987).

Note that we ignore possible feedback effects through changes in long–term interest rates. According to Artus and others (1991), French and German longterm rates are only weakly affected by movements in short–term rates.

See Gardner and Perraudin (1992) for the equation–by–equation multipliers.

The one exception is for U.S. rate effects on French and German rates.

Other Resources Citing This Publication