Contagion Phenomena in Financial Crises: Evidence from the Portuguese and Spanish Exchange Rate Crises in the Early 1990s

CONTAGION PHENOMENA IN FINANCIAL CRISES: EVIDENCE FROM THE PORTUGUESE AND SPANISH EXCHANGE RATE CRISES IN THE EARLY NINETIES* Margarida ABREU CISEP-ISEG Mabreu@iseg.utl.pt ABSTRACT Based on the experience of the Portuguese and Spanish financial crises in the early nineties, this paper suggests that the spillover of exchange rate crises may reveal a particular dimension of the financial contagion effect: the presumption of mimetic behavior by monetary authorities. This paper analyses the evolution of the credibility of the escudo and the peseta. We set out to test the existence of a contagion effect, that is, in what way does the polarization of exchange rate expectations in a scenario of devaluation of one currency explain the building up of a similar scenario for the other currency. We also examine the transmission mechanisms of such a scenario. Our results suggest the existence of a one-way contagion effect, of the escudo by the peseta. Speculative attacks against the peseta necessarily give rise to speculative attacks against the escudo, regardless of the evolution of the “fundamentals” of the escudo. In this case, the spillover of financial crises could be better understood by the anticipated mimetic behavior of monetary authorities, rather than by the geographical proximity of the countries in question or by the identical performance of the economies of both. JEL Classification: E4, E5, G1. Keywords: exchange rate crises, financial contagion. * I’m grateful to two anonymous referees and to V. Mendes for valuable comments on an earlier version of this paper. The usual disclaimer applies. 1 1. INTRODUCTION The Spanish Peseta joined the EMS exchange rate mechanism (ERM) in August 1989 and the Portuguese Escudo followed suit in April 1992. Both currencies participated in the +/-6% fluctuation band around the central parity until the widening of the band in August 1993. From the very beginning and until the general EMS crisis in September 1992, the external nominal value of both currencies remained stable. Both the peseta and the escudo were among the strongest currencies in the mechanism: they remained well within the superior half of their fluctuation band, appreciated relative to their central parities as well as to the deutsche mark. Despite this past behavior, both currencies were subject to numerous speculative attacks, far beyond the general EMS crisis, giving rise to 3 simultaneous currency devaluations of the respective central parities between September 1992 and March 1995. Prior research1 has attempted to find possible reasons for the Portuguese and Spanish exchange rate crises of the early nineties. However, to the best of our knowledge, none of these studies looked for contagion arising from links between events in the two countries. There are several explanations as to why crises tend to be bunched or clustered2. A crisis in one country can spread to others via trade links. Once one country has devalued, it is costly (in terms of a loss of competitiveness and output) for other countries, having strong trade links with the first country, to maintain their parities (Glick and Rose, 1999). An original disturbance can also spread across countries via financial links. Abreu and Scialom (2000) and Calvo (1996) stress the role of liquidity. A leveraged investor facing margin calls needs to sell his asset holdings. 1 See for instance Bartolini and Prati (1999), Campa and Chang (1998), Favero and Giavazzi (2000), Kempa and Nelles (1999), Martinez Peria (1999) or Gómez-Puig and Montalvo (1997). 2 This leads to other asset prices being depressed. Goldstein, Kaminsky and Reinhart (2000) highlight the role of common lenders: the need to rebalance the overall risk of a bank’s asset portfolio can lead to a credit crunch. Another family of contagion models has emphasized the role of the correlation of fundamentals in the presence of asymmetric information among investors (Kodres and Pritsker, 1998). The Portuguese and Spanish exchange rate events studied in this paper seem to correspond to another special case of contagion. In this case, spillover seems to be better explained by the anticipated mimetic behavior of the Portuguese monetary authorities vis a vis the Spanish monetary authorities, rather than focusing on the geographical proximity of the countries or on the strong direct trade or financial links existing between them. We view this phenomenon as a particular case of the escape clause approach to currency crisis (Eichengreen and Wyplosz, 1993; Isard, 1995; Sutherland, 1995). The escape clause approach views exchange rate arrangements as conditional commitment devices. A country which adheres to a fixed exchanged rate arrangement commits itself to maintaining the exchange rate within a particular band, in view of certain benefits. This commitment is, however, limited, in the sense that the country’s policymaker may always have recourse to the escape clause, i.e., if the costs associated with the fixed peg come to outweigh the benefits of the same. From this perspective, a currency crisis is a situation in which private agents perceive that monetary authorities are on the brink of abandoning the prevailing exchange rate policy. What would seem to be particular to the Portuguese case is that market expectations of the monetary authorities’ evaluation of benefits and costs depend mostly on the 2 See for instance Caramazza at al. (2000), Jeanne (1999) or Williamson (2000b). 3 likelihood of Spanish monetary authorities abandoning the prevailing exchange rate commitment in response to a speculative attack. If Spanish authorities respond to an attack with a devaluation of the peseta, this will sharply increase the cost of the prevailing exchange rate policy in Portugal. Investors revise their beliefs on the continuity of the Portuguese commitment to the exchange rate system insofar as mimetic behavior of Portuguese monetary authorities is expected. Speculative attacks against the peseta necessarily lead to speculative attacks against the escudo, regardless of the evolution of the “fundamentals” of the escudo. We evaluate this contagion effect via the econometric estimation of the expected rate of depreciation of each currency in the band, using information pertaining to the other country’s financial market as explanatory variables. Considering that among European ERM members, interest rates were, during this period, almost the only effective monetary instrument, we used the influence of interest rates on the expected rate of depreciation of the escudo within the band as a proxy for the effect of monetary policy on exchange rate expectations3. Section 2 briefly presents the parametric tests used to evaluate the credibility of the exchange rate policy in each country. Section 3 presents our own estimations: the expectation of a currency devaluation is measured here by the interest rate differential, adjusted by the estimated depreciation expectation of a currency within the band. Conclusions are presented in section 4. Our results suggest the existence of a one-way contagion effect, of the peseta on the escudo. The peseta does not suffer any impact as a result of fluctuations in the credibility of the escudo exchange rate. In this case, the spillover of financial crises can be better understood by the anticipated 3 Figure 1 shows exchange rates of the Portuguese escudo and the Spanish peseta against the deutsche mark and the one-month Portuguese and Spanish interest rates. 4 mimetic behavior of monetary authorities than by the geographical proximity of the countries or by an identical performance of the economies of both countries. 2. PARAMETRIC CREDIBILITY TESTS We estimated the anticipated rate of devaluation for the escudo and for the peseta using parametric credibility tests in the tradition of Svensson (1991a), Rose and Svensson (1991) and Bertola and Caballero (1990). It was assumed that the anticipated total rate of depreciation of a currency is equal to the interest rate differential, i.e. we accept the hypothesis of uncovered interest rate parity. In this case, the anticipated devaluation rate in a fluctuation band regime equals the difference between the interest rate differential and the currency’s depreciation within the band. This is a parametric method in the sense that the anticipated depreciation rate within the band is estimated rather than computed from the full width of the fluctuation band (as in the simple credibility tests – Svensson, 1991b). The expected devaluation rate can therefore be defined as follows: (1) gtm = δ tm − Et  ∆ xt + m | norealignment  / m.dt where gtm is the expected rate of devaluation (at period t) of the central parity m periods in advance (if it is positive a devaluation is expected); this is equal to the difference between the interest rate differential and the expected rate of depreciation within the band; δ tm = itm - i*tm , is the interest rate differential (domestic rate – foreign rate) at period t, on bonds with maturity m; 5 xt = st - ct , is the difference between the log of the market exchange rate s and the log of the bilateral central parity c; m is the maturity measured in periods and dt is the length of the period; Et  ∆ xt + m | norealignment  / m.dt , is the expected (at period t) depreciation rate within the band, between periods t and (t+m), conditional upon the non-existence of a realignment; Äxt+m = xt+m - xt. The expected devaluation rate of the domestic currency (escudo for Portugal and peseta for Spain) in relation to the deutschmark within a certain time horizon was calculated as the expected domestic rate of depreciation of the currency within the band, subtracted from the interest rate differential between each of the Iberian countries and Germany. The expected depreciation within the band was estimated by OLS using the following equation: (2) m + ε t +m ( xt + m − xt ) / m.dt = ∑ β 0 j d j + β1 xt + β 2∆ mpte + β3∆ esp where j dj : dummy variables, identifying each time-period between two realignments. We therefore allow for (and are able to test) different anticipated depreciation rates between realignments. If a coefficient is positive and significant, this means that on average the escudo (peseta) was a weak currency within the band during the period between the two realignments; xt : as above, is the exchange rate within the band, and is frequently used as the sole explanatory variable when explaining Äxt+m. The coefficient of xt, if negative and statistically significant, shows the existence of a force pulling the exchange rate 6 towards the central parity, i.e. the deviation of the exchange rate has a stabilizing effect; Ämpte and Ämesp represent the difference between the euro interest rates in Portugal and Germany (Ämpte) and in Spain and Germany (Ämesp). The use of Ämesp allows us to test the influence of the stability of the peseta market on the escudo market. We wished to estimate the expected depreciation rate within the band, conditional upon information available at time t for one, three and twelve-month horizons. Therefore, we consider m.dt equal to 1/12 year, ¼ year and 1 year, respectively (this corresponds to m equal to 21, 64 and 254 daily observations and dt equal to 1/254). As we wished to calculate the expected depreciation rate within the band conditional upon no realignment between dates t and t+m, therefore, for the 1-month horizon, we excluded from the sample the last 21 observations prior to each realignment. For the same reasons, we excluded from the sample 64 (254) observations prior to each realignment, in order to estimate the expected depreciation rate within the band in the 3-month (12-month) horizon. Exchange rate depreciation expectations were estimated by OLS, however we computed standard errors using a Newey-West covariance estimator, allowing for both serial correlation and heteroscedasticity, since the dates overlap. 3. DATA AND RESULTS 3.1. DATA The credibility of the escudo exchange rate band in the EMS was empirically checked through the use of data covering the period from 14 April 1992, one week before the escudo joined the European exchange rate mechanism, to 5 March 1995. During this 7 period, the central parity of the escudo against the mark was changed 3 times (involving three devaluations against the mark): on 23 November 1992 (-6%), 14 May 1993 (-6,5%) and 6 March 1995 (-3,5%), which was practically the closing date of the series available to us. Since the Spanish peseta joined the exchange rate mechanism before the escudo, we considered two series for this currency. One is longer and covers the whole period of the peseta’s participation in the ERM from 19 June 1989 to 5 March 1995; the other is shorter and is identical to that of the escudo, in order to facilitate comparisons. During this period, the central parity of the peseta against the mark was devalued 4 times: on 17 September 1992 (-5%), 23 November 1992 (-6%), 14 May 1993 (-8%) and 6 March 1995 (-7%). The credibility tests were carried out using the official central parities of these currencies in relation to the mark as the standard reference. Our choice presupposed the recognition of the role of the deutschmark as an anchor for the system. We used ECU spot exchange rates, which were recorded daily at a mid-morning central bank telephone conference4. We therefore deduced the exchange rates of the Portuguese escudo and the peseta in relation to the deutschmark. These exchange rates were compared with the official central parities of these currencies vis-à-vis the mark, and with a variation range of +/-6%, in spite of the widening of the official margins agreed to on 2 August 1993. Such a choice was justified by the fact that after this date, the escudo was kept within an implicit band, which corresponded to the earlier band of 12% (Abreu, 1999a). As far as the peseta was concerned, a similar phenomenon was to be noted until December 1994 (Cassola, 1995; Ayuso 1992). After the crisis which occurred in the summer of 1993, the DEM-ESP exchange rate 8 also seems to have been kept within an implicit band, corresponding to the ranges of variation authorized before the widening of the bands. However, from the beginning of December 1994 until the last realignment in March 1995, this exchange rate policy seems to have been abandoned, with the DEM-ESP parity exceeding the ceiling of 84 pesetas per mark5 on a continuous basis, despite the general increase in interest rates on the europeseta. We used daily Portuguese, Spanish and Germany interest rates, for the period from 14 April 1992 to 5 March 1995. The interest rates used are annualized bid rates for 1month, 3-month and 12-month euro-market deposits, recorded at 10 a.m. Swiss time. 3.2. ESTIMATIONS6 OF THE EXPECTED RATES OF DEPRECIATION WITHIN THE BAND • THE INFLUENCE OF MONETARY POLICY ON EXPECTATIONS AS REGARDS THE RATE OF DEPRECIATION OF THE ESCUDO The assessment of the influence of interest rates on the expected rate of depreciation of the escudo within the band can be considered as being a proxy for the effect of monetary policy on exchange rate expectations (Svensson 1990a, Moutot and Namur 1991)7. An effective Portuguese interest rate policy will be revealed by a positive â2 coefficient. Likewise, the influence of Spanish monetary policy on the behavior of the 4 In comparison with the “official fixing” dollar rates, the ecu-rates have the advantage of being simultaneously recorded for the relevant currencies. 5 The ceiling corresponds to the band of variation of +/-6%. 6 We use LIMDEP in our calculations. 7 In truth, it is not in fact monetary policy that is at stake here, but rather certain variables which are directly influenced by this policy and the precise effects on which we do not know. The use of the euromarket interest rates is not disapproved whenever we assume, as we do in this paper, that the monetary link takes place through interest. In fact, after the complete liberalisation of capital movements in December 1992, domestic and euromarket interest rates don’t differ significantly. 9 Portuguese exchange rate within the band will be revealed by a significant â3 coefficient. As far as the influence of interest rate differentials on the expected rate of depreciation of the escudo is concerned, results in Table 1 seem rather disappointing. In fact, â2 and â3 are not significant for the one and three month regressions, and only â3 is significant as regards the one-year regression. For such a maturity, â3 is in fact both negative and statistically significant, whilst the escudo-mark interest differential is negligible. The peseta interest rate therefore seems to play a decisive guiding role, whilst the influence of the escudo interest rate on the formation of expectations about the depreciation of the escudo would seem to be negligible, both for this time-period and for the 1 and 3 month horizons. The fact that â3 is negative suggests that an increase in the interest rate differential between the peseta and the mark led to a reduction of the depreciation rate of the escudo within the band. To put it more simply, an increase in Spanish rates (given German rates) reinforced the escudo. How can we interpret this set of results? Two possible explanations can be given: 1. Portuguese authorities were following the behavior of their Spanish counterparts, who were pursuing an effective policy in terms of the exchange rate objective. Whilst the rise in interest rates in Spain strengthened the peseta (and this is confirmed by the negative sign of the coefficient â3 estimated for the expected rate of depreciation of the peseta – table 2), it also provoked an immediate increase in Portuguese interest rates, since the Portuguese authorities knew that otherwise their currency would be subject to an attack. Thus, an increase in Spanish interest rates would have the effect of restoring the equilibrium of the escudo exchange rate within the band. An analysis of the correlation between the two series of interest rates, as well as a study of the 10 possible relations of causality between the variations in interest rates, would help to clarify whether mimetic behavior is likely to have occurred. 2. Statistical problems plague the results. It is possible to admit that the effectiveness of the Portuguese monetary policy as well as the Spanish influence may not always have had the same intensity, nor the same sign throughout the short period analyzed, resulting in the inaccuracy of the â2 and â3 estimates8. On the other hand, it is also possible that we have strongly correlated regressors due to the interdependent Iberian interest rates. An understanding of the influence of interest rate differentials on the behavior of the rate of depreciation of the escudo within the band therefore requires the continuation of this study along two other tracks. On the one hand, possible changes in the sign and size of the coefficients â2 and â3 need to be looked for throughout the period of participation in the EMS. From this perspective, new estimations were made of the rates of depreciation of the escudo and peseta, either by allowing the coefficients â2 and â3 to vary between sub-periods, or by rearranging the sub-periods and causing them to coincide with the different phases of credibility of the escudo. On the other hand, possible relationships of interdependence need to be highlighted between the evolution of Spanish and Portuguese interest rates. In this sense, we need to analyze the correlation between the two series of interest rates and the variation of this correlation throughout the period, as well as the possible causality relation (in Granger’s sense) between the interest rate variations in the two countries. 8 As seen in section 2, in order to estimate the 1-year regression only the observations from 13 May 1993 (the date of the second realignment of the escudo) to 5 March 1994 (one year before the third realignment of the escudo) were kept. This shorter period of time corresponds to the longest sub period 11 • IN SEARCH OF DIFFERENT CHARACTERISTIC PERIODS MARKING THE INFLUENCE OF INTEREST RATE DIFFERENTIALS ON THE BEHAVIOR OF THE ESCUDO WITHIN THE BAND In order to test the influence of interest rate differentials on the behavior of the escudo within the band, the rate of depreciation of the escudo was estimated for a one-month horizon, for each sub-period between realignments, using equation (3) below: (3) m + ε t + 21 ( xt + 21 − xt ) / ( 21/ 254 ) = β 0 + β1 xt + β 2∆ mpte + β 3∆ esp Results are presented in table 3. Two initial statistical conclusions emerge from these results. Firstly, the instability of the â2 coefficient is evident. Secondly, the low model performance for the 14/04/1992 – 22/11/1992 sub-period: if one uses the conventional 5% significance level, the explanatory variables are useless. However, during the other two regimes, that is, between the first and the third escudo realignments, the coefficients â2 and â3 were both negative, suggesting that the increase in Spanish and Portuguese interest rates strengthened the escudo within the band. As far as the Portuguese rates were concerned, the differential with Germany (â2) was negative, as expected9. This minus sign does in fact confirm the relationship between the interest rate differential and the exchange rate predicted by the model of target zones, whenever the exchange rate band is credible: in this case, an increase in the of Iberian exchange rate stability. This could explain the non-significant â2 and â3 in the one-month and three-month regressions and the significant â3 in the one-year regression. 9 Given that we have eliminated the observations preceding a realignment over a period that was shorter than the horizon considered for the estimated depreciation rate, the information relating to the periods when attacks against the escudo were not successfully counteracted by the monetary authorities was also eliminated (otherwise the realignment shouldn’t have happened). This means that, very probably, we have underestimated the possible existence of a positive relationship between the interest rate differential in relation to Germany and the expected escudo depreciation rate. In other words, we have over-estimated the efficiency of the interest rate policy. It is therefore normal to obtain a negative â2 coefficient for the full period, which expresses the fact that a rise in interest rates strengthens the currency, except in periods of particular tension in the foreign exchange market that precedes a change in regime. 12 differential (i.e. an increase in domestic interest rates) strengthens the currency within the band. In our opinion, the minus sign confirms that the management of interest rates in Portugal has been largely governed by the objective of exchange rate stabilization and that, most of the time, this policy has proved to be effective. As far as the influence of Spanish interest rates on the escudo is concerned, the hypothesis that the peseta has an influence on the escudo cannot be rejected, at least for the period from November 1992 to May 1993 and the 10% significance level. Thus, when taken together, the coefficients â2 and â3 seem to support, for the onemonth time horizon, in the second and third sub-periods, the results that were obtained with the estimation of the rate of depreciation of the escudo for a one-year horizon (Table 1). The credibility of the escudo fluctuation band was guaranteed by the rise in Portuguese and Spanish interest rates. The rise in Spanish rates probably operated (or was perceived by Portuguese authorities) as a constraint on monetary policy in Portugal. Nevertheless, we cannot be satisfied with these results. In fact, given the aforementioned â2 parameter instability and the irrelevance of the first sub-period regression, we must be somewhat guarded as regards the conclusions put forward. As a result, the question arose as to whether the breakdown into sub-periods marked by the occurrence of a realignment was in fact the best way to understand the possible changes in the influences of interest rate differentials on the depreciation rate of the escudo. In fact, the realignments do not seem to mark any significant changes in the evolution of the escudo within the band. Using "simple" credibility tests10, Abreu 10 In the “simple test”, the devaluation expectation is computed by adjusting interest rate differentials for the maximum and minimum rates of possible currency depreciation within the band. This method has the advantage of taking into account all the observations, but, on the other hand, the corresponding test of credibility of the exchange rate band is frequently inconclusive. 13 (1999a) drew conclusions as regards the relevance of three other sub-periods. The first one, from the date when the escudo joined the ERM to August 1992, was characterized by strong and stable credibility; the second, between September 1992 and July 1993, was marked by a growing nervousness in the markets (increased volatility), with credibility being restored in leaps and bounds and only temporarily after each realignment; the third period, between July 1993 and March 1995, was marked by an extreme lack of credibility of the escudo band, which was much greater than that of the peseta. We therefore re-estimated the depreciation rate of the escudo and the peseta within the band, for all time horizons, in keeping with equation (2), this time taking into consideration the sub-periods from Abreu 1999a). We also re-estimated the rate of depreciation of the escudo, for the time horizon of one month, separately for each subperiod between realignments, in keeping with equation (3). The results of these estimations are shown in tables 4, 5 and 6. Results in table 4 show remarkable differences with regard to the identification of regimes. The coefficient â03 is now significant for all time horizons, and â02 is significant for the 3-month horizon. On the other hand, by taking these periods into account, we were able to identify more clearly the influence that, as a whole, pressures exerted on the peseta have on the behavior of the escudo within the band. The â3 coefficient for the escudo is negative and statistically significant for time horizons of 3 months and one year. It therefore seems possible to confirm the hypothesis of a contagion effect of the peseta on the escudo for these horizons. The Spanish authorities increased their interest rates to strengthen the peseta, an action in which they were successful (for â3 is also negative for the peseta, table 5), and the Portuguese authorities accompanied them in order to prevent the escudo from being 14 attacked, with equal success. Table 6 allows us to determine the variation, between the different sub-periods, in the influence of interest rates on the expected escudo depreciation rate for one month. The coefficients â2 and â3 are significant for the first two sub-periods11. Until August 1992, â2 was negative whilst â3 was positive. This combination represents, a priori, a competitive relationship between the two currencies: an increase in Portuguese interest rates strengthens the escudo, whilst Spanish rates exert pressure in the opposite direction. A rise in interest rates in the neighboring country would expose Portugal to an outflow of capital that would accentuate the depreciation of the escudo. Between August 1992 and July 1993, â2 and â3 were both negative. We therefore find once again the situation envisaged by the hypothesis of a credible band guaranteed by an effective monetary policy which closely follows what is happening in Spain. In the third sub-period, after August 1993, â2 and â3 were not statistically significant. • BRIEF ASSESSMENT OF THE CORRELATION BETWEEN PORTUGUESE AND SPANISH INTEREST RATES AND SEARCH FOR CAUSALITIES An assessment of the correlation, as well as possible relationships of causality, between Portuguese and Spanish interest rates, would seem indispensable for the interpretation of the aforementioned results. By observing the coefficients of correlation between the Portuguese and Spanish interest rates presented in Table 7, the following may be concluded: 11 Considering a 10% significance level. 15 i. that there is an important positive correlation between Portuguese and Spanish interest rates throughout the whole period of the escudo’s participation in the ERM; ii. that this link is stronger whenever the maturity of the rates is longer; iii. that the strength of this link seems to vary greatly throughout the period of the escudo’s participation in the ERM. After being very strong until July 1993, it diminished thereafter, to the point of becoming negative, for time horizons of one month and three months. The results referred to above attest that there is a strong interdependence12 between Portuguese and Spanish interest rates, and that this actually increased after the 1992 crises. This interdependence remained positive until the widening of the bands in 1993, a period in which both the peseta and the escudo were under pressure. After July 1993, the correlation becomes negative, in a period when only the Portuguese escudo was under attack. The interdependence between escudo and peseta interest rates can be refined by using econometric techniques which look for causal relations. It is important to stress that the causality established in this way does not necessarily reflect the underlying existence of a fundamental economic law. The econometric causality that links two variables together simply indicates that knowledge of past values of one variable improves prediction ("cause") with regard to the other. The results of the causality tests applied to the period from 6 April 1992 to 7 March 1995 are shown in table 8. These results suggest that there is a relationship of instantaneous causality, in one direction only, from the peseta to the escudo, after the 16 escudo’s integration into the exchange rate mechanism. No other relationship of causality can be confirmed, neither in the same direction (but nonetheless considering past values), nor in the opposite direction (from the escudo to the peseta). 3.3. PRINCIPAL CONCLUSIONS ON THE RATE OF DEPRECIATION OF THE ESCUDO WITHIN THE BAND With regard to the influence exercised by interest rate differentials, and therefore by monetary policy, on the expected rate of depreciation of the escudo within the band, the principal results obtained allow us to deduce that the influence of the variation in (Portuguese and Spanish) interest rates on the position of the escudo within the band did not have the same sign throughout the EMS period. In fact, for the time horizon of one month, results in tables 1 and 4 show no significant â2 and â3 values, if one considers these coefficients fixed for the entire EMS period. For the one-month horizon, statistically significant coefficients of interest rate differentials were only obtained in the case of separate estimations of the depreciation rate within each sub-period. Results in table 6 lead us to conclude that: i) From the moment when the escudo joined the ERM until the summer of 1993, Portuguese monetary policy could be regarded to have been successful (in relation to the exchange rate objective); the increase in domestic interest rates making it possible to strengthen the escudo. In fact, 12 We must bear in mind that correlation coefficients may be biased upward during periods of high volatility (Forbes and Rigobon, 1999), though this may be less of a problem with interest rates compared to exchange rates or other financial variables. 17 the â2 coefficient, which expresses the efficiency of monetary policy in relation to the exchange rate objective, is negative during this period13. ii) During a very short period only, right at the very beginning of the participation of the escudo in the ERM (between April and August 1992), the negative â2 coefficient14 is linked to a positive â3. This situation would seem to indicate that during the first months of the escudo’s participation in the ERM, monetary policy was effective and not constrained by the behavior of the Spanish monetary authorities, with the relationship between the currencies demonstrating the typical features of a competitive relationship. iii) Between August 1992 and July 1993, both â2 and â3 are negative and significant. It would therefore seem that during this phase, monetary policy continued to show a tendency to be effective in Portugal (since the increase in Portuguese rates reduced expectations of a depreciation of the escudo), although it was nonetheless largely determined by the behavior of the Spanish authorities. A rise in interest rates in Spain (in view of the strengthening of the peseta) was very quickly followed by Portuguese authorities, who thus prevented the escudo from being attacked. A study of the interdependence between the interest rates of the two countries seems to support this interpretation. In fact, we found a strong correlation between the interest rates of the two countries during this period. Furthermore, the instantaneous causality noted from the peseta interest rates to those of the escudo also seems to confirm this idea. 13 This apparent efficiency of monetary policy does, however, need to be viewed with a certain amount of reservation, since we have eliminated the observations preceding a realignment. 18 iv) In the first two periods, in which the coefficients were seen to be significant15, the coefficient associated with Spanish interest rates was always stronger, which would seem to account for the intensity of the exchange rate constraint exerted by Spain. v) After July 1993, it is not possible to make any judgment; either about the influence of Portuguese interest rates, or about the influence of Spanish rates on the behavior of the escudo within the band, since neither â2 nor â3 are significant. As far as Portuguese interest rates are concerned, this fact seems to demonstrate a tendency towards a reduction in the efficiency of the interest rate policy. In the case of the Spanish influence, the fact that â3 is not statistically significant confirms that there was no real symbiosis in the behavior of the exchange rates of the two currencies. In this phase, and unlike the earlier ones, the peseta was not under attack (the peseta was simply hovering close to the band’s ceiling of 6% throughout most of this period) and it was the escudo that was constantly subject to tensions in the foreign exchange market. The results for â3 in this period, coupled with the results obtained for previous periods, seem to suggest the existence of a one-way transfer of the weakness of the peseta onto that of the escudo. The recursive estimation of â3 (results not shown) seems to confirm this hypothesis, namely that the influence of the peseta on the escudo was exerted at moments when the peseta was "weak", and not throughout each sub-period analyzed here. 14 15 Significant at the 10% level. At least at the 10% level. 19 3.4. ESTIMATIONS OF THE EXPECTED RATE OF DEVALUATION The expected rate of devaluation of the escudo and the peseta in relation to the mark is obtained by deducting the expected rate of depreciation within the band from the interest rate differential on the euromarket between the currency of the country being considered and that of Germany, in keeping with equation (1). On the one hand, the longer the horizon of expectation, the higher the expected rate of devaluation, which is natural, given that a realignment is all the more likely to occur when the time period considered is longer. On the other hand, the shorter the horizon of expectation, the higher the volatility of the expected rate of devaluation, which is explained by the relatively high volatility of short rates in comparison with long rates. Let us now focus on the shortest horizon: one month. Figure 2 retraces the evolution of the expected rate of devaluation (by time unit) of the escudo and the peseta against the mark for the time horizon of one month. The estimations chosen for the depreciation rate are those shown in tables 1 and 216. Three sub-periods can be clearly identified. i) The first of these sub-periods, beginning at the moment of entry into the ERM and continuing until August 1992, is marked by a weak volatility of the devaluation rate. On the other hand, the devaluation rate, which is negative throughout the sub-period, expresses a strong level of credibility in regard to the exchange rate band. This credibility of the escudo exchange rate band was only 16 This choice arises from the fact that the estimation of equations (2) displays better quality than that centred exclusively on the differential between the exchange rate and the central parity (xt). Opting for sub-periods between realignments vis-à-vis "regimes" allows us, first of all, not to introduce information gaps within each sub-period (which would inevitably influence the constants associated with each period), for, in order to avoid sudden leaps in the exchange rate within the band (the depreciation rate is conditionally estimated in the absence of a realignment), the estimation requires the elimination of "m" observations before each realignment. Secondly, this choice makes it easier to establish comparisons with the behaviour of other currencies. 20 to be seen again in a somewhat temporary fashion after the realignment of May 1993. ii) After August 1992, this state of grace enjoyed by the escudo came to a sudden end. During this second sub-period, between August 1992 and July 1993, phases of credibility followed on from periods when a realignment was seen as imminent, i.e. phases in which the only common feature was an enormous volatility of the expected rate of devaluation. The devaluation of November 1992 brought little more than a temporary respite, for the rate of devaluation only briefly remained below zero. After the realignment, the rate of devaluation shot upwards again, becoming positive after March 1993. Credibility was once more temporarily restored for a brief two-month period after the realignment of May 1993. iii) Finally, in the third sub-period, between July 1993 and the realignment in March 1995, there was an almost continuous expectation of a devaluation of the escudo, whilst, at the same time, its volatility was reduced. During this long period of roughly 20 months, two crises in the credibility of the escudo can be distinguished: the first one in October 1993, the other in June 1994. An analysis of Figure 2 allows us to compare the evolution of the expected rates of devaluation of the escudo and the peseta. There are several points which should be pointed out here: i) It clearly seems that the credibility of the bands of the two currencies evolved in a practically concerted fashion following the second realignment of the peseta 21 (first of the escudo) and until the widening of the bands, in August 1993. Thereafter, their evolutions diverged quite dramatically. ii) The moments of greatest weakness of the peseta led almost systematically to a realignment of parities, which therefore brought with them a devaluation of the escudo. The only exception was the realignment of September 1992, after which expectations of a devaluation of the escudo remained very high. iii) The lack of credibility of the escudo exchange rate band, which had increased in the second half of 1993 and continued throughout 1994, does not seem to have contaminated the credibility of the peseta. The realignment of March 1995 only arose after the rapid rise in the expected rate of devaluation of the peseta after December 1994 and at a time when the escudo seemed to have partly regained some of the credibility that it had lost. 4. CONCLUSION The preceding analysis shows that the participation of the escudo in the ERM was marked by a repeated lack of credibility after the autumn of 1992. In particular, the ± 6% escudo exchange rate band, which had been implicitly defended by the Portuguese monetary authorities since August 1993, showed itself to be continuously lacking in credibility. There are some fundamental reasons which justify the speculative attacks suffered by the Portuguese currency in 1992, particularly the over-appreciation of the escudo real exchange rate which accumulated from the second half of the eighties (Abreu 1999b, Kempa and Nelles 1999, Williamson 2000a). What is more difficult to understand is 22 why the Portuguese escudo continued to be under pressure until March 1995, once this real over-appreciation had been absorbed by the firsts two realignments. A credible explanation for this behavior is suggested by the strong link noted between the escudo and the peseta. Yet, as we have seen, this was not a marriage of the "for better or worse" type. The escudo seems to have followed the peseta when the expected rate of devaluation of the latter moved upwards, but it did not, however, pull the peseta down with it in its moments of "weakness", nor did it benefit from mirroring the peseta in its periods of greater vigor. In other words, it seems that the credibility of the peseta was not a sufficient condition, but was instead a necessary condition, for the credibility of the escudo. As Eichengreen, Rose and Wyplosz (1997) have clearly pointed out, we speak of contagion in foreign exchange markets when there exists a systematic effect on the probability of a speculative attack which stems from attacks on other currencies, and is therefore an additional effect above and beyond those of domestic “fundamentals”. In the Portuguese and Spanish case, this spillover seems to be well explained by the anticipated mimetic behavior of the Portuguese monetary authorities vis a vis the Spanish monetary authorities. We evaluate this phenomenon as a particular case of the escape clause approach to new models of currency crises. A country which adheres to a fixed exchange rate arrangement makes that committment in view of certain benefits. The commitment to maintain the exchange rate within a band is limited, in the sense that the country’s policymaker can always exercise the escape clause, i.e. devalue, if the benefits associated with the fixed peg are outweighed by the costs of the same. From this perspective, a currency crisis is a 23 situation in which private agents perceive that monetary authorities are on the brink of abandoning the prevailing exchange rate policy. What would seem to be particular to the Portuguese case is that market expectations of the monetary authorities’ evaluation of benefits and costs depend mostly on the behavior of Spanish monetary authorities in response to a speculative attack. If Spanish authorities respond to an attack with a devaluation of the peseta, it is supposed that this will sharply increase the cost of the prevailing exchange rate policy in Portugal. A devaluation of the peseta thus leads investors to revise their beliefs on the continuity of the Portuguese commitment to the exchange rate system. 24 REFERENCES Abreu, Margarida (1999a), “As crises cambiais do escudo entre 1992 e 1995” Boletim de Ciências Económicas de Coimbra. Abreu, Margarida (1999b), “Exchange rate market tensions in a small open economy during the transition to EMU: can Asian countries learn from the Portuguese experience en 1992-1995”, Instituto Superior de Economia e Gestão, Economics Department Working Paper, WP7/1999/DE/CISEP, ISSN 0874-4548. Abreu, Margarida, and Laurence Scialom (2000), “Prudential Policy and Exchange Rate Regime: The Lessons of the Asian Crisis” in M. Abreu and V. Mendes, eds, What Financial System for the Year 2000?, Lisboa: Bolsa de Valores de Lisboa e Porto. Ayuso, J. and all (1992), "Efficiency in the Peseta Forward Exchange Rate Market", Centre for Economic Policy Research Discussion Paper 627, February. Bartolini, Leonardo, Alessandro Prati (1999), “Soft exchange rate bands and speculative attacks: theory and evidence from the ERM since August 1993”, Journal of International Economics, 49, pp. 1-29. Bertola, G., and R. Caballero (1990), "Target Zones and Realignments", Centre for Economic Policy Research Discussion Paper 398, Mars. Calvo, Guillermo (1996), "Capital Flows and Economic Management", Federal Reserve Bank of San Francisco Working Paper in Applied Economic Theory 96-02. Campa, José M., P. H. Kevin Chang (1998), “ERM Realigments Risk and Its Economic Determinants as Reflected in Cross-Rate Options”, The Economic Journal, 108 (July), pp. 1046-1066. Caramazza, Francesco, Luca Ricci, Ranil Salgado (2000), “Trade and Financial Contagion in Currency Crises”, IMF Working Paper nº 00/55, March. Cassola, N. (1995), "Indicadores de Credibilidade de um Regime Cambial: os casos do escudo e da peseta no Sistema Monetário Europeu", Bank of Portugal, Research Department, February. Eichengreen, Barry, A. Rose, and C. Wyplosz (1997), “Contagious Currency Crises”, Scandinavian Economic Review 98, 4. Eichengreen, Barry, and C. Wyplosz (1993), "The Unstable EMS", INSEAD Working paper 93/38/EPS, Fontainebleau, France. Favero, Carlo A., Francesco Giavazzi (2000), “Looking for Contagion: Evidence From the ERM”, CEPR Discussion Paper Series nº 2591, October. Froot, K. A., and M. Obstefeld (1989), "Exchange Rate Dynamics under Stochastic Regime Shifts: A Unified Approach", Harvard Institute of Economic Research Discussion Paper 1451. _____(1991), "Exchange Rate Dynamics Under Stochastic Regime Shifts: a unified approach", Journal of International Economics, (31), 203-229. Glick, Reuven, and Andrew Rose (1999), "Contagion and Trade: Why are Currency Crises Regional?", Journal of International Money and Finance, (18)4, 603-617. Goldstein, Morris, Graciela Kaminsky, and Carmen Reinhart (2000), Assessing Financial Vulnerability: An Early Warning System for Emerging Markets, Institute for International Economics. Gómez-Puig, Marta, José G. Montalvo (1997), “A new indicator to assess the credibility of the EMS”, European Economic Review, 41, pp. 1511-1535. Isard, P. (1995), Exchange Rate Economics, Cambridge University Press, Cambridge, U.K.. Jeanne, Olivier (1999), "Currency Crises: A perspective on Recent Theoretical Developments", Centre for Economic Policy Research Discussion paper 2170, June. Kaminsky, Graciela (1998), “Currency and Banking Crises: The Early Warnings of Distress”, Board of Governors of the Federal Reserve System, International Finance Discussion Paper nº 629, October. Kempa, Bernd, Michael Nelles (1999), “Misalignments and Real Exchange Rates and the Credibility of Nominal Currency Bands”, Weltwirtschaftliches Archiv, vol. 135 (4), pp.613-627. 25 Kodres, Laura, and Matthew Pritsker (1998), "A Rational Expectation Model of Financial Contagion", Board of Governors of the Federal Reserve System, Finance and Economic Discussion paper 1998-48, December. Moutot, P., and D. Namur (1991),"Taux de Change et Zone Cibles: Une Synthèse de la Littérature Récente", CRIEGE - URA CNRS 1420, Cahier de Recherche 91 204. Paris: Centre National de Recherche Scientifique. Peria, Maria Soledad Martinez (1999), "A Regime Switching Approach to Studying Speculative Attacks : A Focus on Monetary System Crises", World Bank Policy Research Working Paper nº 2132, June. Rose, A. K., and L. Svensson (1991), "Expected and Predicted Realignments: the FF/DM Exchange Rate during the EMS", Institute for International Economic Studies, Seminar Paper 485. Stockholm: Institute for International Economic Studies. Sutherland, A. (1995), "Currency crisis models: bridging the gap between old and new approaches", C. Bordes, E. Girardin and J. Melitz, eds, European Currency Crises and After, pp. 57-81. Manchester: Manchester University Press. Svensson, L. (1990a), "The Term Structure of Interest Rates Differentials in a Target Zone: Theory and Swedish Data", The National Bureau of Economic Research, Working Paper 3374. _____(1991a), "Assessing Target Zone Credibility: Mean Reversion and Devaluation Expectations in the EMS", Centre for Economic Policy Research, Discussion Paper 580, September. _____(1991b), "The Simplest Test of Target Zone Credibility", Centre for Economic Policy Research Discussion Paper 493, January. Williamson, John (2000a), “Capital Mobility, Contagion, and Crises”, M. Abreu and V. Mendes, eds, What Financial System for the Year 2000?. Lisboa: Bolsa de Valores de Lisboa e Porto. Williamson, John (2000b), Exchange Rate Regimes for Emerging Markets: Reviving the Intermediate Option. Washington: Institute for International Economics. 26