Journal Issue
Share
Article

The Stabilizing Effect of the ERM on Exchange Rates and Interest Rates: Some Nonparametric Tests

Author(s):
International Monetary Fund. Research Dept.
Published Date:
January 1994
Share
  • ShareShare
Show Summary Details

Events since the so-called “Black Wednesday” of September 16, 1992, have caused some observers to doubt the future of the exchange rate mechanism (ERM) of the European Monetary System (EMS). The speculative spasm that seized the markets in 1992 precipitated the floating of both the lira and the pound sterling (the former “inside” and the latter “outside” the ERM) and caused the first devaluation of the Spanish peseta. Since that time several other currencies—the Portuguese escudo, the Irish punt, the Danish kroner, and the French franc—have come under intermittent, and at times intense, speculative pressure.

Whether and how the ERM might be reconstituted to insulate it from such pressures is still a controversial issue. Although worries concerning substitutability among member currencies and the long-term viability of the ERM had been voiced earlier (Artis and Taylor (1989)), few argue that the ERM has not succeeded at stabilizing volatility in member exchange rates. This paper provides empirical support for that presumption and examines some additional propositions where the presumption is less clear. Thus, we examine whether what is true of intra-ERM exchange rates is also true of exchange rates between ERM members and nonmembers, and we look at whether the stabilizing effect of the ERM holds for real exchange rates. In addition, because stable exchange rates might be bought at the cost of unstable interest rates, we explicitly examine whether this is true.

Further, the EMS, far from being a chronologically homogeneous regime, has evolved over time. In particular, Giavazzi and Spaventa (1990) observe that before the disturbances the system was exceptionally stable. We examine this proposition and confirm that the stabilizing effect of the ERM became more marked over this period of time. Given the recent intense pressures within the ERM, however, we also investigate the volatility in interest rates during the United Kingdom’s participation in the mechanism. The core of this paper is the application of a nonparametric method for testing whether the ERM had a stabilizing effect; this is in contrast to the large number of studies that employ parametric methods. There are good reasons, as we explain later, for our nonparametric approach.

I. Provisions of the EMS

The European Monetary System was instituted in March 1979, largely as a reaction to the volatility of exchange rates among European countries during the years following the breakdown of the Bretton Woods system. At the heart of the EMS is the exchange rate mechanism. Countries participating in the ERM agree to maintain their bilateral exchange rates within ±2.25 percent of a central rate. In an exception, Italy negotiated a temporarily wider band of ±6 percent at the outset of the system and moved to the narrow band in January 1990, an example subsequently followed by the United Kingdom, Spain, and Portugal when these countries entered the ERM. (Spain and Portugal are still working within this broader band while the United Kingdom now lets the pound float.) Spain entered the ERM in June 1989; the United Kingdom in October 1990; and Portugal in April 1992.

Maintenance of the bilateral rates within the bands is a symmetrical obligation: the strong-currency country is equally obliged to prevent its currency from piercing the ceiling as the weak-currency country is to prevent it from falling through the floor. Resources to support these obligations through market intervention are made available through credit lines, notably through the Very Short Term Financing Facility. The Basle-Nyborg agreements of 1987 strengthened the ERM by expanding these credit lines, lengthening the repayment period associated with them and establishing the “presumption” that credit would be available for intervention within the band (so-called “intramarginal intervention”), whereas previous provisions allowed only for intervention at the limits of the band (“marginal intervention”).1 The Basle-Nyborg agreement also called for a strengthening of cooperative monetary policy.

The central rates in each bilateral band can be changed in a realignment.Table 1 shows the timing and extent of such realignments before the speculative surge of September 1992.2 As can be seen, the use of realignments declined rapidly after the early 1980s.

Table 1.Changes in EMS Central Rates Prior to September 1992
Dates of realignments
9/2411/303/2210/52/226/143/217/214/78/41/121/8
197919791981198119821982198319851986198619871990
Percentage change in parity
Belgian franc0.00.00.00.0–8.50.0+1.5+2.0+1.00.0+2.00.0
Danish kroner–2.9–4.80.00.0–3.00.0+2.5+2.0+1.00.00.00.0
German mark+2.00.00.0+5.50.0+4.25+5.5+2.0+3.00.0+3.00.0
French franc0.00.00.0–3.00.0–5.75–2.5+2.0–3.00.00.00.0
Irish punt0.00.00.00.00.00.0–3.5+2.00.0–8.00.00.0
Italian lira0.00.0–6.0–3.00.0–2.75–2.5–6.00.00.00.0–3.7
Dutch guilder0.00.00.0+5.50.0+4.25+3.5+2.0+3.00.0+3.00.0

Although provisions concerning bilateral rates form the core of the ERM, the system is formally organized around a composite currency, the European currency unit (ECU); central rates for participating currencies are expressed in terms of ECU. While it is purely a formal construction, the ECU allowed for an interesting technical innovation in the EMS—the divergence indicator and threshold positions. According to these provisions, when the ECU value of a currency reaches 75 percent of its band limit relative to other currencies, a presumption is created that the country concerned should take corrective action involving monetary and fiscal policy or a realignment. This technical provision—also called the divergence threshold—provides an early warning of wide bilateral movements and, more important, isolates the errant currency.

There seems to have been little doubt in the minds of those who constructed these provisions that the errant currency would be the deutsche mark.3 It is one of the curiosities of the ERM that the mark has not often been at the higher end of its permitted range and that, for most of its operation, the ERM’s anti-inflationary stance has been so strong that the mark was not singled out (Padoa-Schioppa (1983)), and the mark naturally evolved as the anchor of the system.

It is important to note that the introduction of the ERM did not require the abolition of exchange controls nor were controls over capital movements initially removed, notably in France and Italy (although these countries did abolish remaining restrictions on capital movements in January and May 1990 respectively, ahead of the deadline of July 1 imposed by a European Community directive). Artis and Taylor (1988) provide evidence—the offshore-onshore interest rate differentials for the mark and lira after March 1979—that these controls were used substantially. The controls may have fostered stability by giving the authorities the whip hand in negotiating realignments and by avoiding the immediate convergence of monetary policy, which the freedom from capital controls, coupled with the obligation to defend central bilateral parities, would have implied.

The immediate objective, then, of the ERM has been the stabilization of bilateral nominal exchange rates among its members. Although the ERM in its early phase of operation was used to restrain exchange rate “overshooting”—preserving the competitiveness of participant countries through frequent realignment—the system also evolved into a counter-inflationary framework. Without convergent inflation rates, real exchange rate stability does not necessarily follow from stable nominal rates. Yet, it can be argued that, like a customs union, the EMS must have an “inner rationale” of maintaining broadly stable conditions of competitiveness. Otherwise, the purpose of reducing exchange rate protection will be questioned. It is just as important, therefore, to explore real exchange rate stabilization as it is to test for nominal rate stabilization.

A further issue involves the stability of EMS exchange rates vis-à-vis outside currencies. To the extent that the ERM encourages greater coherence among the partner currencies, the more inevitable it becomes that the currencies vis-à-vis third currencies will become more homogeneous. To take an example that is not entirely fanciful, if the dollar-mark rate is exceptionally volatile, some of the mark’s volatility against the dollar will be imparted to the franc. (To the extent to which this does not happen, tensions within the ERM will be caused by shifts of sentiment about the dollar, which would exert pressure on the mark-franc exchange rate.) Thus, we also explore whether the reduced volatility among intra-EMS parities has been offset by increased volatility in extra-EMS parities for some European currencies.

II. Volatility in the EMS

Before the events of September 1992, there had been 12 realignments in the EMS. This fact, together with the fact that wide variations are allowed by the margins and that less than full convergence of inflation has occurred among ERM members (Masson and Taylor (1992)), calls into question whether the EMS actually does induce greater stability in either the nominal or the real exchange rate.

The difference, stressed by John Williamson (1985), between the concepts of exchange rate volatility and misalignment is important here. Volatility is a “high frequency” concept referring to movements in the exchange rate over comparatively short periods of time. Misalignment, on the other hand, refers to the capacity of an exchange rate to depart from its fundamental equilibrium value (however defined) over a protracted period of time. In a world of risk-neutral producers and consumers, it can be shown that higher exchange volatility actually enhances overall welfare (see De Grauwe (1992), pp. 64–67), but this conclusion becomes more debatable when allowance is made for risk aversion and incomplete forward markets. A related question concerns the effect of ERM membership on interest rates. If the union is successful in generating a convergence of interest rates toward those of the low-inflation anchor, there may be positive welfare gains since higher interest rates may generate important principal-agent problems in domestic capital markets (Stiglitz and Weiss (1981)). Also, some authors (Baldwin (1989), European Commission (1990)) have argued—using an endogenous growth model framework—that lower interest rates may generate permanently higher growth for GDP.4

With respect to the effects of exchange rate volatility on trade flows, the evidence is mixed. A study of Akhtar and Hilton (1984) found a negative correlation between exchange rate volatility and U.S.-German trade flows, but a comparable study by the International Monetary Fund (1983) failed to confirm this finding for other trade flows, time periods, and volatility measures. Cushman (1986), however, does find volatility effects on trade when “third country” effects are controlled (for example, dollar-mark volatility may affect U.S.-U.K. trade).

Despite these caveats, a number of studies have concentrated on evidence that the EMS has reduced exchange rate volatility, most notably those by Ungerer, Evans, and Nyberg (1983), Ungerer and others (1986, 1990), the European Commission (1982), Padoa-Schioppa (1983), Rogoff (1985), and Artis and Taylor (1988). There are many possible approaches to this question—the choice of exchange rates (bilateral, effective, nominal, or real), data frequency (daily, weekly, monthly, or quarterly), the standard against which stability is to be judged (the level or change in exchange rates, conditional or unconditional), and the precise statistical measure chosen (standard deviation). Then, there is the question of the counterfactual—supplied in these studies and others like them by the behavior of a control group of non-EMS currencies. Without exception, the EMS in these studies has been judged as having improved the stability of intra-EMS bilateral exchange rates, although the improvement is less marked for effective rates.

III. Some Nonparametric Volatility Tests

Many of the studies that have tested for a post-March 1979 downward shift in exchange rate volatility have relied on purely descriptive statistics. As such, they can only suggest results, and it is difficult to assess the performance of the EMS from this evidence. The more straightforward approach to the problem, namely estimating a specific parameterization of the volatility and testing for a structural shift, is fraught with pitfalls. This is because economists are uncertain about the statistical distribution of exchange rate changes.

It is a stylized fact that percentage exchange rate changes tend to follow leptokurtic (fat-tailed, highly peaked) distributions. Westerfield (1977), for example, finds that the stable paretian distribution with characteristic exponent less than two fits the change in the logarithm of spot exchange rates better than the normal distribution. In a similar vein, Rogalski and Vinso (1977) suggest Student’s t-distribution as a good approxmiation. It may well be that the distribution of exchange rate changes is normal but that the variance shifts through time—perhaps according to the amount of “news.” This would give the appearance of a stable leptokurtic distribution. Some evidence for such behavior is provided by Boothe and Glassman (1987), who find that mixtures of normal distributions provide some of the best fits.

We wish to stress the importance of the distributional properties of exchange rate changes in any volatility study. Studies that rely on simple variance measures implicitly invoke a normality assumption, the legitimacy of which is being challenged by a growing number of studies (see Boothe and Glassman (1987) for additional references). For example, it is conceivable that exchange rate changes at a certain frequency have a Cauchy distribution, for which no finite moments of any order exist.

To circumvent some of these problems, we employ nonparametric tests for volatility shifts, which do not require estimation of the distributional parameters. Instead, exchange rate changes are ranked by size, and inferences are drawn with respect to the shape of the ranking. Intuitively, if a significant number of lower-ranked percentage changes are recorded in the latter half of the sample period, a reduction in volatility would be indicated. Note, however, that although the test procedure is nonparametric in the sense that no volatility measures are actually estimated, we still must choose an appropriate distribution for exogenous disturbances. This issue is discussed later in this section.

In the nonparametric technique we employ, let Δet be the change in the logarithm of the exchange rate at time t. Then, the hypothesis is

where μ, α, and β are unknown constant scalars, εt is independently and identically distributed with distribution function F and density function f, and zt is a binary variable reflecting the hypothesized shift in volatility at time N + 1. Thus,

Given equation (1), the null hypothesis of no shift in volatility is

Hajek and Sidak (1967) develop a number of nonparametric tests for dealing with this kind of framework, which under appropriate regularity conditions are locally most powerful (Hajek and Sidak (1967), pp. 70–71). The test statistics take the form

where z is the arithmetic mean of the zt sequence of T observations and ut is defined as follows. Let r() be the rank of Δei—that is, r(Δei) is the r(Δei)th smallest absolute change in the total sequence considered—then ut= r(Δet)/(T + 1).

Clearly, ut must lie in the closed interval [1/(T + 1), T/(T + 1)], for no ties in rank. The function α(⋅) in equation (4) is a score function defined by Hajek and Sidak that depends upon the assumed density of e,. Hajek and Sidak define a class of functions that can be used in place of the score function in large samples, since α(⋅) may be difficult to evaluate in practice. If F is the assumed distribution function of εt,

and if F-1(u) is the inverse of F,

then the asymptotic score function,ψ(⋅), is defined (Hajek and Sidak (1967), p. 19)

Under equation (1), the statistic

(α(⋅) in equation (4) is replaced by ψ(⋅))will be asymptotically normally distributed.

Under the null hypothesis (3), η will have a zero mean and variance ρ2 given by (Hajek and Sidak (1967), pp. 159–160)

where

Thus, for a given choice of f, the statistic (η/ρ) will be asymptotically standard normal under the null hypothesis of no shift in volatility. Significantly negative values of η reflect a negative value for β in equation (2)— an increase in volatility after the shift point—while significantly positive values of η imply a reduction in volatility.

As noted at the outset of this section, although the test procedure is nonparametric in the sense that no volatility measures are actually estimated, we cannot avoid choosing an appropriate distribution for εt in implementing the procedure. To minimize the risk of choosing an inappropriate distribution, we selected four well-known ones in the belief that the true distribution will be close to one of them. If qualitatively similar nonparametric results are obtained for a range of assumed distributions, then the results may be said to be robust to this uncertainty. The densities correspond to the normal, logistic, double exponential, and Cauchy distributions. The density and asymptotic score functions (as defined in equation (8)) for these distributions are given in the appendix. All of the chosen distributions are symmetric, and both the double exponential and Cauchy distributions have fat tails.5

IV. Data

The data we use are monthly data on bilateral U.S. dollar exchange rates and on nominal effective exchange rates, both taken from the International Financial Statistics data tape for the period January 1973 through October 1990, and on Eurocurrency interest rates (three-month maturity) for the period January 1975 through February 1993. The end point of the exchange rate sample coincides with Britain’s entry into the ERM: this choice of sample allows us to use sterling as a representative non-ERM currency. Bilateral rates against the German mark and the pound sterling were also constructed by assuming a triangular arbitrage condition. Real exchange rates were constructed by deflating by the wholesale price relatives (data also from the IFS tape).6 The currencies used included three ERM currencies—German mark, French franc, and Italian lira—and three non-ERM currencies—U.S. dollar, pound sterling, and Japanese yen. We also obtained monthly data on three-month Eurodeposit interest rates to test the hypothesis that exchange rate fixity may impart interest rate volatility. All results are for shifts in the volatility of monthly changes.

V. Overall ERM Effect

In the first set of tests, we look for a post-March 1979 shift in exchange rate volatility. In some sense, we expect such tests to be affected by the behavior of the dollar during the 1980s, which was highly volatile. Thus, if intra-ERM exchange rates displayed reduced volatility during a period when most exchange rates against the dollar were highly volatile, this would be strong evidence of an overall ERM effect.

Nominal Exchange Rates

As expected, the nonparametric tests reveal a significant reduction in the volatility of ERM currencies against the mark after March 1979 (Table 2), while the volatility of the mark against the pound, yen, and dollar appears unchanged.7Table 2 also shows that although the dollar-lira exchange rate became more volatile after March 1979, this is not the case for the dollar-franc and dollar-mark exchange rates. (The test statistics, however, are uniformly negative, suggesting a tendency toward increased volatility.) The volatilities of the dollar-pound and dollar-yen exchange rates have risen significantly since March 1979.

Table 2.Test Statistics for a Shift in Nominal Exchange Rate Volatility(1973:1–1979:3 vs. 1979:4–1990:10)
Exchange rateNormalLogisticDouble

exponential
Cauchy
Mark nominal exchange rate
Mark–franc6.405 (0.000)5.396 (0.000)5.272 (0.000)5.399 (0.000)
Mark–lira7.563 (0.000)6.118 (0.000)5.942 (0.000)5.093 (0.000)
Mark–pound0.222 (0.824)0.135 (0.893)0.103 (0.918)–0.159 (0.874)
Mark–yen–0.498 (0.619)–0.444 (0.657)–0.451 (0.652)–0.546 (0.585)
Mark–dollar–0.502 (0.616)–0.685 (0.493)–0.644 (0.520)–1.310 (0.190)
Dollar nominal exchange rate
Dollar-franc–0.935 (0.350)–1.098 (0.272)–1.323 (0.186)–2.869 (0.004)
Dollar-lira–1.835 (0.067)–1.929 (0.054)–2.058 (0.040)–3.770 (0.000)
Dollar-mark–0.604 (0.546)–0.755 (0.450)–0.720 (0.471)–1.361 (0.173)
Dollar-pound–2.625 (0.009)–2.171 (0.030)–2.189 (0.029)–2.830 (0.017)
Dollar-yen–1.956 (0.050)–2.088 (0.037)–2.179 (0.029)–4.286 (0.000)
Nominal effective exchange rate
Franc2.720 (0.007)2.099 (0.036)1.941 (0.052)1.148 (0.251)
Lira3.143 (0.002)2.225 (0.026)2.144 (0.032)0.637 (0.524)
Mark3.077 (0.022)2.356 (0.018)2.294 (0.022)1.419 (0.156)
Pound–1.694 (0.090)–1.635 (0.102)–1.717 (0.086)–2.634 (0.008)
Yen–1.531 (0.126)–1.120 (0.263)–1.101 (0.271)–0.458 (0.647)
Dollar–3.006 (0.003)–2.670 (0.008)–2.738 (0.006)–3.652 (0.000)
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.

Finally, Table 2 reveals an unequivocal reduction in the volatility of nominal effective ERM exchange rates after March 1979. Exactly the converse is true of the U.S. dollar nominal effective rate. The results for the effective pound and yen rates, although significant in only one case, are nevertheless uniformly negative and large in absolute size, indicating a tendency toward increased volatility.

Real Exchange Rates

Table 3 shows a marked and significant reduction in the volatility of the mark-lira rate after March 1979, and a similar—albeit statistically insignificant—reduction in the mark-franc rate (as indicated by the large, positive values of the test statistics). The real exchange rates of the mark against the non-ERM countries do not exhibit a significant shift (Table 3).

Table 3.Test Statistics for a Shift in Real Exchange Rate Volatility(1973:1–1979:3 vs. 1979:4–1990:10)
Exchange rateNormalLogisticDouble

exponential
Cauchy
Mark real exchange rate
Mark–franc1.652 (0.099)1.469 (0.142)1.492 (0.136)1.993 (0.046)
Mark–lira7.390 (0.000)6.159 (0.000)6.088 (0.000)6.170 (0.000)
Mark–pound0.502 (0.615)0.313 (0.754)0.287 (0.774)–0.154 (0.877)
Mark–yen0.006 (0.996)–0.023 (0.981)–0.017 (0.986)0.030 (0.976)
Mark–dollar0.314 (0.754)–0.041 (0.967)–0.005 (0.996)–0.914 (0.361)
Real effective exchange rate
Franc3.543 (0.000)2.916 (0.004)2.860 (0.004)2.563 (0.010)
Lira4.780 (0.000)3.872 (0.000)3.814 (0.000)3.573 (0.000)
Mark1.232 (0.218)1.019 (0.308)0.985 (0.325)1.001 (0.317)
Pound–0.808 (0.419)–0.773 (0.440)–0.830 (0.407)–1.200 (0.230)
Yen–1.390 (0.164)–1.017 (0.309)–0.948 (0.343)–0.365 (0.715)
Dollar–4.273 (0.000)–3.703 (0.000)–3.735 (0.000)–4.641 (0.000)
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.

Table 3 also shows the results of the tests applied to real effective exchange rates. Strong and significant reductions in volatility are indicated for the franc and the lira, and positive statistics were also recorded for the mark. For the pound and the yen, the test statistics are negative but insignificant, while the real effective rate of the dollar shows a strongly signficant rise in volatility after March 1979.

An Overall ERM Effect?

Overall, the results reported in this section indicate that the ERM has reduced exchange rate volatility—both real and monimal—since March 1979. This is particularly impressive in light of evidence that the volatility of non-ERM exchange rates—particularly the dollar—has risen over the same period.

VI. A New EMS?

A number of commentators (including Giavazzi and Spaventa (1990)) have noted a shift in the nature of the EMS toward less frequent realignments and more concerted action toward internal adjustment. It is not entirely clear when this shift began, and thus we tested for a shift using two different subperiods of ERM operation. We first tested for a shift in exchange rate volatility after the realignment of March 1983 (Tables 4 and 5). The results indicate a downward shift in the volatility of ERM exchange rates—real and nominal, bilateral and effective. For the dollar and the pound, however, there is little shift in volatility, although yen exchange rates appear to have become more stable over this period. The second subperiod begins after the realignment of April 1986. The results of testing for a shift in volatility after this realignment (Tables 6 and 7) are broadly comparable to those reported for the first subsample— although yen volatility appears lower.

Table 4.Test Statistics for a Shift in Nominal Exchange Rate Volatility(1979:4–1983:3 vs. 1983:4–1990:10)
Exchange rateNormalLogisticDouble

exponential
Cauchy
Mark nominal exchange rate
Mark–franc2.200 (0.028)1.694 (0.090)1.688 (0.091)1.139 (0.255)
Mark–lira2.326 (0.020)1.934 (0.053)1.969 (0.049)2.087 (0.037)
Mark–pound1.671 (0.095)1.339 (0.818)1.337 (0.818)1.203 (0.229)
Mark–yen3.224 (0.001)2.636 (0.008)2.648 (0.008)2.626 (0.009)
Mark–dollar0.564 (0.573)–0.500 (0.617)–0.481 (0.631)–0.528 (0.598)
Dollar nominal exchange rate
Dollar-franc0.755 (0.450)0.572 (0.567)0.623 (0.533)0.706 (0.480)
Dollar-lira0.165 (0.869)0.027 (0.978)0.107 (0.915)–0.005 (0.996)
Dollar-mark–0.228 (0.819)–0.292 (0.771)–0.255 (0.799)–0.467 (0.640)
Dollar-pound0.037 (0.970)–0.026 (0.979)–0.038 (0.970)–0.439 (0.661)
Dollar-yen1.899 (0.058)1.582 (0.114)1.594 (0.111)1.512 (0.131)
Nominal effective exchange rate
Franc0.951 (0.341)0.748 (0.455)0.797 (0.425)0.760 (0.448)
Lira1.340 (0.180)1.074 (0.283)1.025 (0.306)0.859 (0.390)
Mark1.740 (0.082)1.196 (0.232)1.152 (0.249)0.201 (0.841)
Pound0.696 (0.486)0.473 (0.636)0.407 (0.684)0.027 (0.978)
Yen Dollar1.460 (0.144)1.241 (0.215)1.269 (0.204)1.552 (0.121)
Dollar0.294 (0.769)0.158 (0.875)0.152 (0.879)–0.258 (0.796)
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.
Table 5.Test Statistics for a Shift in Real Exchange Rate Volatility(1979:4–1983:3 vs. 1983:4–1990:10)
Exchange rateNormalLogisticDouble

exponential
Cauchy
Mark real exchange rate
Mark–franc1.731 (0.083)1.337 (0.181)1.283 (0.200)0.658 (0.511)
Mark–lira3.976 (0.000)3.453 (0.001)3.388 (0.001)3.923 (0.000)
Mark–escudo3.751 (0.000)2.854 (0.004)2.699 (0.007)1.414 (0.157)
Mark–pound1.809 (0.070)1.426 (0.154)1.407 (0.160)1.209 (0.227)
Mark–yen3.397 (0.001)2.790 (0.005)2.820 (0.005)2.703 (0.007)
Mark–dollar–0.417 (0.679)–0.404 (0.687)–0.394 (0.693)–0.536 (0.592)
Real effective exchange rate
Franc2.636 (0.008)2.177 (0.029)2.169 (0.030)2.269 (0.023)
Lira3.985 (0.000)3.433 (0.001)3.313 (0.001)3.489 (0.001)
Mark3.185 (0.001)2.611 (0.009)2.569 (0.010)2.562 (0.010)
Pound1.143 (0.253)0.849 (0.396)0.823 (0.410)0.432 (0.666)
Yen2.350 (0.019)1.947 (0.052)1.977 (0.048)2.155 (0.031)
Dollar0.362 (0.717)0.261 (0.794)0.273 (0.784)0.028 (0.978)
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.
Table 6.Test Statistics for a Shift in Nominal Exchange Rate Volatility(1979:4–1986:5 vs. 1986:6–1990:10)
Exchange rateNormalLogisticDouble

exponential
Cauchy
Mark nominal exchange rate
Mark–franc0.248 (0.804)0.105 (0.917)0.083 (0.934)–0.362 (0.717)
Mark–lira1.575 (0.115)1.201 (0.230)1.312 (0.190)1.067 (0.286)
Mark–pound1.474 (0.140)1.179 (0.238)1.064 (0.288)0.691 (0.489)
Mark–yen1.431 (0.153)1.034 (0.301)1.035 (0.301)0.471 (0.638)
Mark–dollar–0.175 (0.861)–0.306 (0.759)–0.295 (0.768)–0.698 (0.485)
Dollar nominal exchange rate
Dollar-franc0.912 (0.362)0.647 (0.517)0.695 (0.487)0.466 (0.641)
Dollar-lira–0.156 (0.876)–0.266 (0.790)–0.252 (0.801)–0.559 (0.576)
Dollar-mark–0.177 (0.860)–0.277 (0.782)–0.276 (0.782)–0.587 (0.557)
Dollar-pound–0.119 (0.905)–0.119 (0.905)–0.096 (0.924)–0.209 (0.834)
Dollar-yen–0.456 (0.649)–0.467 (0.641)–0.428 (0.669)–0.927 (0.354)
Nominal effective exchange rate
Franc2.200 (0.028)1.829 (0.067)1.868 (0.062)2.104 (0.035)
Lira0.974 (0.330)0.798 (0.425)0.767 (0.443)0.721 (0.471)
Mark2.388 (0.017)2.025 (0.043)2.021 (0.043)2.349 (0.019)
Pound0.818 (0.413)0.569 (0.569)0.463 (0.643)–0.076 (0.939)
Yen–0.811 (0.417)–0.789 (0.430)–0.782 (0.434)–1.282 (0.200)
Dollar1.041 (0.298)0.777 (0.437)0.775 (0.438)0.433 (0.665)
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.
Table 7.Test Statistics for a Shift in Real Exchange Rate Volatility(1979:4–1986:5 vs. 1986:6–1990:10)
Exchange rateNormalLogisticDouble

exponential
Cauchy
Mark real exchange rate
Mark–franc1.290 (0.197)1.070 (0.285)1.102 (0.270)1.093 (0.274)
Mark–lira2.096 (0.036)1.728 (0.084)1.679 (0.093)1.612 (0.107)
Mark–pound1.442 (0.149)1.111 (0.267)1.032 (0.302)0.596 (0.551)
Mark–yen1.751 (0.080)1.359 (0.174)1.437 (0.151)1.144 (0.253)
Mark–dollar–0.400 (0.689)–0.532 (0.595)–0.521 (0.602)–1.177 (0.239)
Real effective exchange rate
Franc2.706 (0.007)2.231 (0.026)2.193 (0.028)2.176 (0.030)
Lira4.720 (0.000)3.975 (0.000)3.892 (0.000)3.857 (0.000)
Mark2.415 (0.016)1.996 (0.046)1.900 (0.057)1.767 (0.077)
Pound1.472 (0.141)1.157 (0.247)1.121 (0.262)0.889 (0.374)
Yen–0.016 (0.987)–0.182 (0.855)–0.236 (0.814)–0.968 (0.333)
Dollar0.826 (0.409)0.563 (0.574)0.588 (0.556)0.104 (0.918)
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.
Note: Figures in parentheses denote marginal, two-sided significance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.

We therefore conclude that the hypothesis that a “new” and harder EMS arose after 1982, in which greater emphasis was given to harmonizing the macroeconomic objectives of EMS members (Giavazzi and Spaventa (1990)).

VII. Volatility Transfer

It is sometimes argued that advanced macroeconomic systems naturally generate a “lump of uncertainty” that can be pushed out of the economy at one point only to reappear somewhere else (see Bachelor (1983, 1985)).8 In particular, this suggests that reducing exchange rate volatility inevitably causes interest rate volatility to rise. Such a conclusion might follow from inverting a standard exchange rate equation and noting that the interest rate is the only other major “jump variable” in the system. Such a phenomenon might be termed a “volatility transfer.” Insofar as the burden of increased interest rate volatility falls on the general public more squarely than that of exchange rate volatility, which presumably falls mainly on the tradable goods sector, then the welfare argument must hinge on which sector would find it easier to hedge the induced risk. Given the already well-developed forward foreign exchange markets, it is probable that such an argument would come down against membership in the ERM.

However, it is not at all clear that ERM membership is equivalent to “inverting the exchange rate equation.” Insofar as membership enhances the credibility of policy, there may be a significant reduction in speculative attacks on the exchange rate and hence a reduction in the volatility of short-term interest rates (if the authorities use interest rates as a short-term measure for “leaning into the wind”). Such credibility arguments rest crucially on the assumption that the costs to the authorities of revaluation outweigh the costs of internal adjustment—and, in particular, the costs of disinflation (see Giavazzi and Giovannini (1989)).

To shed some light on these arguments, we carried out the nonparametric tests for monthly changes in Eurocurrency short-term interest rates; the results are reported in Table 8.

Table 8.Test Statistics for a Shift in Eurocurency Interest Rate Volatility
Exchange rateNormalLogisticDouble

exponential
Cauchy
Period: 1975:1–1979:3 vs. 1979:4–1990:10
Franc1.53 (0.126)1.42 (0.155)1.37 (0.171)1.15 (0.250)
Lira2.04 (0.041)2.19 (0.028)2.13 (0.033)2.41 (0.016)
Mark1.18 (0.238)0.97 (0.332)0.84 (0.401)0.72 (0.472)
Pound1.14 (0.254)1.67 (0.095)1.43 (0.153)1.37 (0.171)
Yen1.13 (0.258)0.93 (0.352)1.13 (0.258)1.18 (0.238)
Dollar–2.18 (0.029)–2.13 (0.033)–2.26 (0.024)–2.19 (0.028)
Period: 1979:4–1983:3 vs. 1983:4–1990:10
Franc4.892 (0.000)4.231 (0.000)4.191 (0.000)4.779 (0.000)
Lira1.994 (0.046)1.585 (0.113)1.592 (0.111)1.349 (0.177)
Mark6.051 (0.000)5.001 (0.000)4.945 (0.000)4.826 (0.000)
Pound3.857 (0.000)3.266 (0.001)3.326 (0.001)3.892 (0.000)
Yen5.213 (0.000)4.293 (0.000)4.232 (0.000)4.068 (0.000)
Dollar7.642 (0.000)6.385 (0.000)6.204 (0.000)5.994 (0.000)
Period: 1979:4–1986:5 vs. 1986:6–1990:10
Franc4.542 (0.000)3.924 (0.000)3.915 (0.000)4.575 (0.000)
Lira–0.062 (0.950)–0.153 (0.879)–0.220 (0.826)–0.768 (0.442)
Mark3.639 (0.000)3.084 (0.002)3.150 (0.001)3.651 (0.000)
Pound3.009 (0.003)2.545 (0.011)2.585 (0.010)3.054 (0.002)
Yen3.553 (0.000)2.907 (0.004)2.829 (0.005)2.606 (0.009)
Dollar5.072 (0.999)4.301 (0.000)4.198 (0.000)4.347 (0.000)
Period: 1987:2–1990:10 vs. 1990:11–1992:9
Franc1.314 (0.189)0.913 (0.361)0.714 (0.475)0.398 (0.691)
Lira–0.331 (0.741)–0.183 (0.855)–0.043 (0.966)0.577 (0.564)
Mark0.699 (0.485)0.588 (0.556)0.674 (0.500)0.763 (0.445)
Pound1.223 (0.221)0.821 (0.411)0.638 (0.523)–0.532 (0.595)
Yen–0.307 (0.759)–0.235 (0.814)–0.225 (0.822)–0.191 (0.848)
Dollar–0.277 (0.821)–0.299 (0.765)–0.215 (0.830)–0.652 (0.515)
Note: Figures in parentneses denote marginal, two-sided sigificance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.
Note: Figures in parentneses denote marginal, two-sided sigificance levels. All test statistics are distributed as standard normal under the null hypothesis of no shift in volatility. Significantly positive test statistics indicate a reduction in volatility after the break point; significantly negative statistics indicate the converse.

Table 8 reveals that the overall effect of the ERM has not been to increase interest rate volatility—if anything, the test statistics show reduced interest rate volatility for ERM members, which is strongly significant for lira interest rates. By contrast, dollar interest rates have seen a significant rise in volatility during the operation of the EMS.

During operation of the EMS, there seem to have been further reductions in interest rate volatility (less marked for the lira), although a specific ERM effect cannot be separated from a global effect, since significant reductions in volatility are also indicated for dollar, yen, and sterling interest rates (Tables 8.B and 8.C).

Overall, the results of this section indicate that the stability of nominal and real ERM exchange rates was not bought at the expense of increased interest rate volatility.

VIII. British Participation in the ERM

Given the comparatively brief duration of the U.K.’s membership in the ERM (October 1990–September 1992) and the growing tensions inside the system during this period, it is interesting to examine whether our nonparametric test procedures can detect these tensions. While this period—less than two years—is arguably too short to yield reliable conclusions, it does seem reasonable that the reduced credibility of the system should be reflected in increased interest rate volatility among member countries. Thus, in Table 8 we report results of the nonparametric procedures applied to interest rate data for the period from the January 1987 realignment to the exit of the pound from the ERM in September 1992, with a hypothesized shift point in volatility after the pound’s entry in October 1990. The results reveal no significant shift in interest rate volatility after October 1990 either for the pound or for any other currency examined.

These findings may be explained by returning to the distinction between misalignment and volatility. It seems that some of the tensions within the system reflected cumulative currency misalignments, particularly in the case of the high-inflation currencies; in the case of the United Kingdom, some commentators have argued that the pound joined at too high a rate vis-à-vis the mark (see Wren-Lewis and others (1991)). The pound’s rapid devaluation since leaving the ERM gives this hypothesis some empirical support. Thus, if a primary cause of the tensions within the system was currency misalignment—a low-frequency concept—we should not necessarily expect to see this reflected in the volatility of interest rates—a high-frequency concept.

More generally, we have argued elsewhere (Artis and Taylor (1989)) that the ERM did not render member currencies perfect substitutes in international portfolios, as required of a fully credible exchange rate union (Canzoneri (1982)) and did not exhibit a convincing capacity for correcting cumulative misalignments over time. It seems plausible that market interest in these longer-run issues was heightened by the Danish rejection of the Maastricht Treaty in 1992 and the less than overwhelming (“petit oui”) support for the agreement in the French referendum a few months later.

IX. Conclusions

In this study we investigated the volatility of the exchange rates of the ERM countries up to October 1990, before the entry of the United Kingdom into the ERM, and the volatility of interest rates during subsamples of the period extending through the pound’s participation in the ERM. Because there are doubts about the true distribution of exchange rate and interest rate changes, a nonparametric statistical method was used. The volatility of ERM exchange rates and interest rates was compared with that of a control group of non-ERM currencies before and after the inception of the ERM. Their behavior through time was also examined.

The essential findings are very clear. During the operation of the ERM, the volatility of intra-ERM exchange rates (specifically other ERM rates against the mark) fell while the volatility of non-ERM currencies remained the same or increased. This effect was big enough to replicate itself for ERM countries’ overall effective exchange rates also. Similar conclusions, albeit not quite so striking, were obtained for real bilateral and real effective exchange rates. The general impression that the ERM evolved toward greater stability is also confirmed by this data set. The same technique was applied to the evolution of volatility in “offshore,” or “Eurocurrency,” interest rates: it also appears that volatility has been somewhat reduced for the ERM countries compared to the control group. This is inconsistent with the “volatility transfer” hypothesis, according to which reduced stability in exchange rates would imply added volatility in interest rates. There is no significant shift in interest rate volatility during the pound’s participation in the ERM.

Given the recent instability in the ERM, the conclusion that the mechanism has exerted an unequivocally stabilizing influence on its member currencies may seem surprising. We would again refer to the distinction made throughout this paper between the short-run (“high frequency”) concept of volatility and the longer-run (“low frequency”) concept of misalignment. In earlier work (Artis and Taylor (1989)), we showed that the ERM was not entirely successful either in correcting long-term misalignments of real exchange rates among member currencies or in rendering member currencies perfect substitutes in international portfolios, as would be expected in a fully credible exchange rate union (Canzoneri (1982)): “Both these findings are worrying since it is easy to imagine the stock of credibility which the EMS has earned being dissipated as sophisticated and forward-looking international capital markets begin to focus on the longer-run stability properties.…” (Artis and Taylor (1989), p. 305). Given the added instabilities that arise naturally in the transition to monetary union (Masson and Taylor (1992)), the recent abolition of exchange controls, which had previously been extensively used by France and Italy (Artis and Taylor (1988)), and the uncertainty occasioned by the sequence of national referenda on the Maastricht Treaty during 1992, the “events of ’92” are neither surprising nor inconsistent with the short-run stabilizing influence of the ERM documented in this paper.

APPENDIX Density and Asymptotic Score Function for the Nonparametric tests
DistributionDensity function, f(x)Asymptotic score function,ψ(u)
Normal(2π)–1/2 exp(–1/2 x2–1(u)}2–1
Logistice–x(1+e–x)–2(2 u–1)ln(u/(1–u)}–1
Double exponential1/2 exp(–|x|)2–In(1 –|2 u – 1|) – 1
Cauchyπ–1(1+x2)–12 tan2{π(u–1/2)}[1+tan2{π(u–1/2)}]–1–1
REFERENCES

    AkhtarM.A. and R.S.Hilton“Exchange Rate Uncertainty and International Trade: Some Conceptual Issues and New Estimates for Germany and the United States,”Research Paper No. 8403 Federal Reserve Bank of New York (New York: Federal Reserve Bank of New YorkMay1984).

    ArtisMichael J. and Mark P.Taylor“Exchange Rates Capital Controls and the European Monetary System: Assessing the Track Record” in The European Monetary Systemed.byGiavazziMicossi and Miller (Cambridge: Cambridge University Press1988).

    ArtisMichael J. and Mark P.Taylor“Some Issues Concerning the Long-Run Credibility of the European Monetary System” in Exchange Rates and Open Economy Macroeconomicsed.byR.MacDonald and M.P.Taylor (Oxford: Blackwell1989).

    BachelorRoy A.“European Monetary System, 13th Report from the Treasury and Civil Service Committee and Evidence” (London: House of Commons Select Committee1983).

    BaldwinRobert“On the Growth Effects of 1992,”NBER Working Paper No. 3119 (Cambridge, Mass.: National Bureau for Economic Research1989).

    BootheP. andD.Glassman“Statistical Distribution of Exchange Rates: Some Empirical Evidences,”Journal of International EconomicsVol. 22 (1987) pp.297319.

    CanzoneriMatthew B.“Exchange Intervention Policy in a Multiple Country World,”Journal of International EconomicsVol. 13 (1982) pp.26789.

    CushmanDavid O.“Has Exchange Risk Depressed International Trade? The Impact of Third Country Exchange Risk,”Journal of International Money and FinanceVol. 5 (1986) pp.36180.

    De GrauwePaulThe Economics of Monetary Integration (Oxford: Oxford University Press1992).

    European Commission“Documents Relating to the European Monetary System” European Economy No. 31 (Brussels: European Commission1982).

    European Commission“One Market One Money” European Economy No. 44 (Brussels: European Commission1990).

    GiavazziFrancesco and AlbertoGiovanniniLimiting Exchange Rate Flexibility: The European Monetary System (Cambridge, Mass.: MIT Press1989).

    GiavazziFrancesco and LuigiSpaventa“The ‘New’ EMS” in The European Monetary System in the 1990sed.byPaulDe Grauwe and L.Papademos (London: Longman1990).

    HajekJ. and Z.SidakTheory of Rank Tests (New York and Prague: Academic Press1967).

    International Monetary FundExchange Rate Volatility and World Trade Occasional Paper No. 28 (Washington: International Monetary Fund1983).

    LudlowP.The Making of the European Monetary System (London: Butter-worths1982).

    MassonPaul R. and Mark P.Taylor“Issues in the Operation of Monetary Unions and Common Currency Areas” in Policy Issues in the Evolving International Monetary System byMorrisGoldstein and othersOccasional Paper No. 96 (Washington: International Monetary Fund1992).

    Padoa-SchioppaTomaso“European Monetary System 5th Report and Evidence” (London: House of Lords Select Committee1983).

    PooleWilliam“Optimal Choice of Monetary Policy Instruments in a Simple Stochastic Macro Model,”Quarterly Journal of EconomicsVol. 85 (1970).

    RogalskiR.J. and J.D.Vinso“Price Level Variations as Predictors of Flexible Exchange Rates,”Journal of International Business StudiesVol. 2 (1977) pp.7181.

    RogoffKenneth“Can Exchange Rate Predictability Be Achieved Without Monetary Convergence,”European Economic ReviewVol. 28 (1985) pp.93115.

    StiglitzJoseph and A.Weiss“Credit Rationing in Markets with Imperfect Information,”American Economic ReviewVol. 71 (1981).

    UngererH.O.Evans and P.NybergThe European Monetary System: The Experience 1979–82 Occasional Paper No. 19 (Washington: International Monetary FundMay1983).

    UngererHurst and othersThe European Monetary System: Recent Developments Occasional Paper No. 48 (Washington: International Monetary Fund1986).

    UngererHurst and othersThe European Monetary System: Developments and Perspectives Occasional Paper No. 73 (Washington: International Monetary Fund1990).

    Van YperseleJ.The European Monetary System (Cambridge: Woodhead1985).

    WesterfieldJ.M.“Empirical Properties of Foreign Exchange Rates Under Fixed and Floating Rate Regimes,”Journal of International EconomicsVol. 7 (1977) pp.181200.

    WilliamsonJohn“The Exchange Rate System” (revised ed.) Policy Analyses in International EconomicsVol. 5 (Washington: Institute for International Economics1985).

    Wren-LewisSimon and others“Evaluating the United Kingdom’s Choice of Entry Rate into the ERM,”The Manchester SchoolVol. 49 (Supplement1991) pp.122.

Although this presumption was established, the financing central bank could still, in principle, object. Other special conditions were also attached to intramarginal interventions under the Basle-Nyborg agreements, relating to credit limits and currency of repayment.

The lira was formally devalued (by 7 percent) on September 12, 1992, but commenced a float soon afterwards on September 17, following the example of sterling which was floated from September 16. The peseta was devalued on September 17, by 5 percent and again, by a further 6 percent, on November 22, at which time the escudo was devalued by a similar amount. The punt was devalued by 10 percent on January 30, 1993; on May 13 the peseta was devalued by a further 8 percent with a parallel devaluation of the escudo of 6.5 percent.

Ludlow (1982) gives a detailed and informative account of the negotiations leading to the institution of the EMS. Van Ypersele (1985) provides a more detailed account of the institutional features of the System.

In fact, these authors apply these arguments to the prospect of a single European currency, but similar arguments apply in the present case.

Another relevant distribution would have been Student’s t. However, the score function (8) for this distribution would have been very difficult to compute. A possibility not considered is that there was a change in distribution of ERM exchange rate changes post-March 1979 (e.g., shifted from normal to Cauchy). Tests for this kind of behavior could conceivably be based on likelihood ratios, although one might suspect that the discriminatory power of such procedures would be low.

Wholesale prices were used as a proxy for tradable goods prices.

As derived above, the test statistic will be asymptotically standard normal (that is, with a mean of zero and variance of unity) under the null hypothesis of no shift in volatility. Significantly negative values of the test statistic (less than about -2) reflect a negative value for p in (2)—i.e., an increase in volatility after the shift point—while significantly positive values (bigger than about +2) imply a reduction in volatility.

Many of the arguments relating to systemic macroeconomic risk can be traced to Poole (1970).

Other Resources Citing This Publication