Assessing the credibility of a target zone: evidence from the EMS

Abstract

This paper provides some new evidence on the credibility of the Exchange Rate Mechanism (ERM) of the European Monetary System (EMS). The study differs from previous research in the literature in three main respects. First, the main contribution is the use of several credibility indicators, some of which have never been applied before to all of the currencies under study. This allows one to strengthen the results obtained in this paper. Second, a longer period than that of previous studies is analysed, covering the complete EMS history. Third, a comparison has been made of the prediction qualities of the different indicators, in order to explore their ability to capture the main ERM events (realignments, changes in the fluctuations bands and speculative pressures). Fourth, the indicators are applied to the experience of the new, modified ERM linking the currencies of non-euro area Member States to the euro, showing the relevance of this approach in the near future with the enlargement of the European Union.

Acknowledgements

The authors would like to thank Beatriz Sanz (Bank of Spain) and Mayte Ledo (BBVA) for kindly providing us with the data set used in this paper. Support from Fundación Centro de Estudios Andaluces (centr A) is also gratefully acknowledged.

Notes

¹ Svensson (1992) and Ayuso and Restoy (1992) have estimated risk premia that are insignificant for the currencies in the ERM and, hence, the expected rate of depreciation is closely related to the interest rate differential.

² When considering the practical implementation of the drift-adjustment method, the empirical studies that have computed this measure have used different econometric specifications for the expected rate of depreciation within the band. Lindberg et al. (1993), Svensson (1993), and Rose and Svensson (1994) have estimated a linear regression model where the exchange rate in t + τ depends on its value at moment t (and, in some cases, lagged exchange rates) and on the interest rate differential. On the other hand, Bertola and Svensson (1993) consider x _t as the only explanatory variable, assuming a mean-reverting model for the exchange rate within the band, as in Ayuso et al. (1994) and in Gómez and Montalvo (1997).

³ We have also taken into account the widening of the bands, since this event produced a major change in the ERM, as can be observed in a greater fluctuation of the exchange rates before August 1993.

⁴ Svensson (1993) eliminates from the sample the 65 observations corresponding to the three months before a realignment took place, given that he, like us, uses τ = 3 months. Given the important reduction in the number of observations implied by this strategy, similarly to Gómez and Montalvo (1997) we use Equation 3 to estimate the whole sample. In this way, we are estimating the expected depreciation rate within the band that includes possible jumps in each realignment. Therefore, we obtain the expected rate of realignment, but not the expected devaluation rate (which, in addition, includes the expected jump in the exchange rate within the band in the realignments).

⁵ Edin and Vredin (1993) employ a two-step procedure suggested by Heckman (1976) to calculate both the probability and the expected size of the devaluation. In the first step of the estimation procedure, the probability of devaluation occurring at time t + 1, based on information available at time t, is estimated. In the second step, the unconditional expectation of the rate of devaluation in period t is obtained.

⁶ Note that this measure, when formulated in this manner, assigns credibility to any period when the lower bound is negative and the upper bound is positive.

⁷ We employ an ARIMA model for its simplicity. On the one hand, the large variety of non-linear models makes it rather difficult to choose the best one and, on the other, previous experience indicates that although a non-linear model could improve on our predictions, the advantages are very limited (see Clements and Smith, 1999 or García and Gençay, 2000).

⁸ The fluctuation bands were built following Honohan (1979). We took into account the lack of symmetry between the two intervention limits due to the requirement that the upper intervention limit for currency X with respect to currency Y equals the lower intervention limit for currency Y with respect to currency X.

⁹ Plots of the results obtained using the exchange rate, the distance to the upper fluctuation band, and the distance to the central parity as the explanatory variable are available from the corresponding author upon request.

¹⁰ The choice of a subperiod prior to each one of the selected events tries to capture the predictive quality of the different measures. Nevertheless, it must be pointed out that we are not taking into account the number of events registered by each indicator; this last element could have been an alternative criteria in order to carry out the comparison.

¹¹ Svensson's simple test was eliminated due to its sensitivity to the size of the fluctuation band.

¹² We have also used Data Envelopment Analysis (DEA) to compare the credibility indicators by simultaneously examining the variation in mean and in standard deviation in all the realignments in the history of the ERM and in the broadening of the bands (August 1993). The results, not shown here but available from the authors upon request, support those obtained from the simple graphic comparison.

¹³ For example, Allen and Taylor (1990) and Taylor and Allen (1992) report that over 90% of participants in the London market rely on technical strategies when formulating short-term exchange-rate expectations.

¹⁴ As can be seen, the moving average rule is essentially a trend following system because when prices are rising (falling), the short-period average tends to have larger (lower) values than the long-period average, signalling an increase (a reduction) in the credibility indicator.

¹⁵ These MA rules are roughly the weekly equivalent to the daily rules examined by LeBaron (1992) and Levich and Thomas (1993) to show the statistical significance of the technical trading rules against several parametric null models of exchange rates.

¹⁶ We have also computed Fisher's exact test that gives the probability of observing a table with as much evidence of association as the table actually observed (under the null of no association). The results from this test (not shown here but available from the authors upon request) also favour the marginal credibility indicator, since in general it yields the probability that is closest to 0, suggesting that the classification of R derived from this indicator is perfectly associated with the observed realignments.

¹⁷ A comprehensive description of these methods can be found in Seiford and Thrall (1990). In this paper we have used the DEAP 2.1.