- Email: [email protected]
Explaining ERM realignments: Insights from optimising models of currency crises
Journal of Macroeconomics 25 (2003) 491–507 www.elsevier.com/locate/econbase
Explaining ERM realignments: Insights from optimising models of currency crises F. Gulcin Ozkan
*
Department of Economics and Related Studies, University of York, Heslington, York YO10 5DD, UK CEPR, London, UK Received 19 April 1999; accepted 6 August 2002
Abstract This paper attempts to provide empirical evidence on the determinants of the realignments throughout the European exchange rate mechanism (ERM). Motivated by the implications of optimising currency crisis models, we relate the probability of ‘‘crises’’ to a set of macroeconomic fundamentals. By using a conditional binominal logit model we show that regime switches are strongly influenced by movements in industrial production, foreign interest rates, competitiveness and imports as well as in foreign exchange reserves. These findings are consistent with the general propositions of recent currency crises models. Ó 2003 Elsevier Inc. All rights reserved. JEL classification: F31; F32 Keywords: Currency crisis; Parity changes; Probability of devaluations; ERM
1. Introduction This paper attempts to provide more insights into the understanding of the determinants of currency crises following the recent turmoil in international financial markets. Currency and financial crises experienced by Mexico in 1994–1995, by Indonesia, Korea, Malaysia, the Phillippines and Thailand in 1997 and more recently by Turkey in 2001 and Argentina in 2002 all revived the interest in understanding the roots of currency collapses. The issue of currency crises has already been a lively research area and has been a subject of investigation in international finance literature
*
Tel.: +44-01904-434673; fax: +44-01904-433759. E-mail address: [email protected] (F.G. Ozkan).
0164-0704/$ - see front matterÓ 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.jmacro.2002.08.001
492
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
since the seminal paper by Krugman (1979). The main proposition of the early literature (generally referred to as first generation models) is that balance of payments crises are precipitated by policy imbalances such as lax monetary and fiscal policies. In addition, the level of foreign exchange reserves is argued to be the main determinant of the duration of a fixed exchange rate regime. The events in the European exchange rate mechanism (henceforth the ERM) in 1992 and 1993 posed serious problems for the propositions of the early models. Although the crises which preceded the exit of the Pound and the Italian Lira from the ERM involved some loss of reserves reserve adequacy was clearly not an issue for these countries which could borrow almost unlimited amounts of foreign exchange due to the provisions of the European exchange rate system. In addition, it might have been possible for these governments to raise interest rates to whatever level necessary to avoid the crises. The fact that these countries chose to leave the system instead was widely interpreted as an indication that the authorities’ objective functions included some other key variables. 1 These observations might indicate that the decision to leave the ERM was to some extent an optimising decision on the part of those countries rather than an action only forced upon them by either policy imbalances or speculators. This has been at the centre of the new approach to the currency crises. These so-called optimising models of currency crises view regime changes as conscious decisions taken by policy makers who continuously weigh the marginal costs versus benefits of being in a fixed exchange rate regime. Main examples of these are Obstfeld (1994, 1996a), Ozkan and Sutherland (1995, 1998), Davies and Vines (1995), Bensaid and Jeanne (1997) and Andersen (1998) among others. These second generation models all share the common feature of modelling policy makers as having well-defined utility functions and thus clear preferences in terms of being in a particular exchange rate regime. The main implication of these models is that a fixed exchange rate regime lasts as long as the policy maker derives more utility from being in it than in either changing the rate or leaving the regime altogether. In other words, a country’s commitment to the fixed exchange rate regime is not state-invariant (see, for example, Flood and Marion, 1998). Motivated by the implications of these new currency crises models, this paper empirically investigates the roots of Ôcurrency crises’ during the ERM period by extending the set of potential determinants, as suggested by this recent literature. Our data set covers the ERM experience between 1979 and 1992. The reasons for constraining the panel to the ERM members and leaving out the Latin American and East Asian experiences are two-fold. First of all, in the case of recent Latin American and East Asian crises, currency collapses can not be analysed in isolation from financial and banking crises (see, for example, Radelet and Sachs, 1998; Kaminsky and Reinhart, 1998). Secondly, although empirical studies on the ERM abound most of the existing research on the European experience focuses on the determinants of expected
1
High interest rates are a huge political cost in Britain due to the fact that a large proportion of mortgages are at variable rate. In the case of Italy, repayments on public borrowing soar as a result of rising interest rates.
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
493
changes in the exchange rate (see, for example, Chen and Giovannini, 1993; Caramazza, 1993; Thomas, 1994; Rose and Svensson, 1994). 2 Here we explore the determinants of actual rather than expected realignments. 3 Therefore, by incorporating an extended set of potential variables we provide fresh evidence on the causes of crises during one of the important examples of pegged exchange rate regimes. The first practical task in carrying out this exercise is how to define Ôcurrency crises’ and distinguish between different crises events. Although it is straightforward to classify the events leading to the decisions of the UK and Italy to leave the ERM in September 1992 as ‘‘currency crises’’, the same is not true of a large number of other changes in central parities throughout the period 1979–1992. Despite this, in this paper we aim to study the sources of all regime switches. Therefore, we attempt to empirically analyse the macroeconomic determinants of pressure leading to all realignments of central parities during the life of the ERM. This is because all of these changes signal regime switches though to a varying degree. As stated above, the contribution of the optimising models of currency crises is to broaden the set of fundamentals that are derived from the objectives of and the constraints faced by the policy makers that affect their decision to leave the existing exchange rate regime. It is crucial to point out, however, that it is also possible for a shift in market expectations to push an otherwise sound economy into a crisis. As Flood and Marion (1998) point out, in such a situation anything that helps to coordinate speculators’ actions may bring about a sudden attack on the currency. This highlights the difficulty in empirically isolating between the implications of the first and second generation models. 4 This is, however, not the purpose of the current study. In this paper, we are mostly concerned with identifying common characteristics of devaluations that could represent sources of macroeconomic trade-offs that were faced by policy makers during the ERM. Affirmative evidence to this would be in line with general propositions of these new models that Ôcurrency crises’ are precipitated by a broader set of Ôfundamentals’ that affect the interactions between policy makers and the markets. 5 Our findings suggest that regime switches in a particular period are influenced by the movements in industrial production, 2 The two exceptions are Eichengreen et al. (1995, 1996). The first of these carries out a comprehensive study of the empirical regularities of macroeconomic data around a number of exchange rate events inclusive of those throughout the ERM. The second tests for the contagious nature of speculative attacks in a sample of 20 countries including the EMS members. 3 The estimates of the actual realignments reveal information about the policy makers’ preferences while that of the expected realignments reflect information about the market participants’ perception of those preferences. These two alternative approaches should yield similar results in terms of the timing of parity changes as long as expectations are rational and correctly measured. However, apart from the practical difficulties in measuring expectations, such factors as the lack of information on the part of the financial markets about the preferences and the constraints of policy makers may lead to incorrect inference about the actual preferences of the policy makers when using expected parity changes. 4 Obstfeld (1996b) also points to this identification problem. 5 A growing body of empirical literature has emerged from testing the implications of these new optimising models of currency crises. Kaminsky et al. (1998) provide a unifying framework and identify a set of macroeconomic variables as the Ôleading indicators of currency crises’ by examining the existing empirical evidence. Our set of regressors are among the main indicators identified by them.
494
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
foreign interest rates, competitiveness and imports as well as the foreign exchange reserves. The rest of this paper is organised as follows. Section 2 explains the empirical specification. The main findings are discussed in Section 3 and finally Section 4 concludes the paper. A brief definition of data is provided in the Appendix A.
2. Empirical specification The purpose of the empirical analysis here is to shed some light on the macroeconomic determinants of devaluations throughout the ERM period. As explained above we use ‘‘crisis’’ here as a general term referring to periods of pressure on the existing parity as well as more extreme cases. Therefore, in this context crisis for a particular country is defined as the realignment of the domestic currency’s central parity against the ECU. There were 43 devaluations between March 1979 and December 1992 in total. 6 A realignment is defined as a change in the central parities relative to the ECU. A devaluation is then effectively the fall in the value of the currency in question against those currencies whose bilateral parities against the ECU remained unchanged. The regressand is a binary choice variable which takes the value 1 for the months in which the domestic currency is devalued against the ECU and 0 otherwise. This suggests that our set of Ôcrises’ excludes unsuccessful speculative attacks. The main reason for this is that in this paper we specifically focus on identifying a set of macroeconomic variables the deterioration of which leads to a change in the actual value of the currency in question. In so doing, our results provide new evidence on the potential for a number of indicators to actually bring about the policy maker’s response. Our main focus as the determinants of actual parity changes is motivated by the elusive nature of the predictability of devaluations and the choice of regressors by the prominence of the policy makers’ preferences in the recent currency crises. The dates and sizes of all 43 realignments between 1979 and 1992 are listed in Table 1. A general discrete response model for realignments can be written as P ðzit ¼ 1Þ ¼ F ðai þ b0 xit Þ;
ð1Þ
where i ¼ 1; 2; . . . ; N and t ¼ 1; 2; . . . ; T and P ðzit ¼ 1) is the probability of devaluation for country i at time t and xit is the vector of macroeconomic variables that influence this probability. The ai ’s are country specific unobserved variables that are assumed to be constant over time. Since the sample of the current empirical analysis is the ERM countries over a period of their participation in the mechanism the data set has both time-series and cross-sectional features.
6
Although there are only 14 parity changes in the sample period (March 1979–December 1992 inclusive) the total number of devaluations is 43 since at each parity changes there was more than one country that experienced a change in the value of its currency.
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
495
Table 1 Dates and the sizes of ERM realignments relative to ECU, 1979–1992 Date 24.09.1979 30.10.1979 05.10.1981 22.02.1982 14.06.1982 21.03.1983 21.07.1985 07.04.1986 04.08.1986 12.01.1987 07.01.1990 14.09.1992 16.09.1992 23.11.1992
Currencies BF
DK
DM
FF
IP
IL
DG
SP
PE
UKS
0.0 0.0 0.0 )8.5 0.0 +1.5 +2.0 +1.0 0.0 +2.0 0.0 0.0 0.0 0.0
)2.9 )4.8 0.0 )3.0 0.0 +2.5 +2.5 +1.0 0.0 0.0 0.0 0.0 0.0 0.0
+2.0 0.0 +5.5 0.0 +4.25 +5.5 +2.0 +3.0 0.0 +3.0 0.0 0.0 0.0 0.0
0.0 0.0 )3.0 0.0 )5.75 )2.5 2.0 )3.0 0.0 0.0 0.0 0.0 0.0 0.0
0.0 0.0 0.0 0.0 0.0 )3.5 +2.0 0.0 )8.0 0.0 0.0 0.0 0.0 0.0
0.0 0.0 3.0 0.0 )2.75 )2.5 )6.0 0.0 0.0 0.0 )3.8 )7.0 s –
0.0 0.0 +5.5 0.0 +4.25 +3.5 +2.0 +3.0 0.0 +3.0 0.0 0.0 0.0 0.0
– – – – – – – – – –
– – – – – – – – – – –
– – – – – – – – – –
0.0 0.0 +6.0
0.0 e –
0.0 0.0 +5.0 +6.0
Notes: Source is Burda and Wyplosz (1997). BF is the Belgian Franc, DK is the Danish Kroner, DM is the German Mark, FF is the French Franc, IP is the Irish Punt, IL is the Italian Lira, DG is the Dutch Guilder, SP is the Spanish Peseta, PE is the Portuguese Escudo and UKS is the British Sterling. s stands for Italy’s suspension of its membership on 16, September 1992 while e is used for the UK’s exit from the system. Italy followed the UK in ending participation on the 22nd of September.
The logit model for the panel data to be used in this analysis can be represented by
7
P ðzit ¼ 1Þ ¼ F ai þ
L X
bi j ii;tj þ
j¼1
þ
L X j¼1
bej ei;tj þ
L X
bp j pi;tj þ
j¼1 L X j¼1
bmj mi;tj þ
L X j¼1
L X
byj yi;tj þ
L X
!
bdj di;tj
j¼1
bresj resi;tj ;
ð2Þ
j¼1
where the ai ’s are country-specific fixed effects, L is the maximum lag, i is interest rates in Germany, p is inflation in Germany, y is the industrial production index, d is the DM/Dollar exchange rate, e is exports, m is imports, res is foreign exchange reserves. It is well-known that (see, for example, Baltagi, 1995, p. 179; Green, 1993, p. 657) in the presence of these country specific effects ai ’s cannot be consistently estimated for a fixed T . In this case the estimation procedure should aim to maximize the conditional likelihood function instead of the unconditional likelihood function for the inconsistency of the maximum likelihood estimators of ai ’s not to be transmitted into the inconsistency of that of b’s. Therefore, in what follows we adopt a conditional logit model to estimate the probability of devaluations. 8
7 8
The choice of the fixed-effects version is motivated by the small number of cross-sectional units. Estimations are carried out by using the statistical software package STATA.
496
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
As to the set of variables, all variables are expressed in logs except interest rates and inflation. Interest rates and inflation in Germany are included because of the asymmetric functioning of the ERM in which Germany acted as the centre country throughout. The DM/Dollar exchange rate is added to consider the effect of competitiveness vis- a-vis the non-ERM countries on the existing ERM parities. This is because an increase in the DM/Dollar exchange rate, within the framework of fixed parities against the DM, amounts to an improvement in the member countries’ competitiveness vis- a-vis the Ôrest of the world’. This choice of the RHS variables is driven by the explicit implications of the above mentioned recent currency crises models. The main assertion that high interest rates in Germany were translated into output losses which were costly for member countries, as argued by Ozkan and Sutherland (1995, 1998) and Davies and Vines (1995) is tested by the inclusion of German interest rates and domestic industrial production index. The link between the terms of trade shocks and the viability of a fixed exchange rate regime is studied by Andersen (1998). It is straightforward to generalise this argument to a relationship between the trade performance and the pressure on the existing parities. This is tested by including competitiveness variables; foreign (German) inflation, and the DM/Dollar exchange rate and trade performance variables; trade balance, exports and imports. In addition to these, RHS variables set includes foreign exchange reserves variable and a number of binary variables. The model adopted here is a general one including several past values of the independent variables. The choice of such a dynamic approach is motivated by the notion that currency pressure can build up over longer periods of time than usually studied in the empirical literature. 9
3. Estimation results Eq. (2) is estimated by the conditional maximum likelihood method using monthly observations for a panel of all the ERM member countries including the six long-term members; Belgium, Denmark, France, Italy, Netherlands and Ireland as well as the ‘‘new members’’; Spain, Portugal and the UK. There are two specifications that are reported in Table 2. Column 1 reports the estimation results when trade performance is measured by the trade balance variable. We arrive at the specification reported in Column 2 by replacing the trade balance variable with exports and imports separately.
9
Edin and Vredin (1993) utilize a binominal probit model to estimate the probability of devaluations and its determinants for the Nordic Countries by using monthly data for the period 1978–1989. It is shown that output growth, the central parity and foreign exchange reserves were the main determinants of the probability of devaluations for this group of countries in the studied period. Similarly Klein and Marion (1997) estimate devaluation probabilities for 16 Latin American Countries for the period 1956–1991 by using a binominal logit model. The real exchange rate, the net asset position and the openness of the economy are shown to influence the probability of devaluing against the US dollar.
Table 2 Panel data conditional logit estimates of the determinants of realignments for all nine ERM countries (1)
it1 it2 it3 pt1 pt2 pt3 yt1 yt2 yt3 dt1 dt2 dt3 tbt1 tbt2 tbt3 et1 et2 et3 mt1 mt2 mt3 rest1 rest2 rest3 pd dev new
1.78 )2.91 1.51 0.24 )1.29 0.77 10.00 4.55 )16.55 )20.64 0.69 19.92 0.01 )0.01 )0.01 – – – – – – )5.92 3.72 0.94 0.23 )1.96 )2.64
(2) (3.65) ()3.44) (2.85) (0.49) ()1.79) (1.52) (1.51) (0.68) ()2.52) ()3.10) (0.07) (2.81) (0.55) ()0.15) ()0.66)
()2.12) (0.96) (0.34) (0.50) ()2.51) ()2.38)
0.38
)0.28 )2.00 )0.03 )0.01
)1.26
1.83 )3.09 1.51 0.13 )1.20 0.93 12.31 5.76 )20.17 )20.44 4.63 15.62 – – – 1.07 0.90 )1.98 )3.13 )3.76 6.89 )6.31 3.64 1.61 0.37 )1.72 )2.73
(3.56) ()3.39) (2.68) (0.24) ()1.49) (1.68) (1.64) (0.82) ()2.80) ()2.97) (0.43) (2.03)
(0.53) (0.45) ()0.92) ()1.45) ()1.66) (2.55) ()2.18) (0.94) (0.58) (0.78) ()2.17) ()1.86)
0.25
)0.14 )2.1 )0.19
)0.01
0.00
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
Dependent variable: P ðzit ¼ 1Þ
)1.06
497
498
Table 2 (continued) (1)
(2)
N Nr LRI LR(1), groupwise Het. LR(1), general Het. RESET(LR) LR
1042 43 0.29 0.07 0.53 1.48 10.26
1042 43 0.32 0.67 0.13 0.07 11.11
Notes: (i) t-ratios are in the parentheses. (ii) N r is the number of realignments in the relevant sample period. (iii) LRI, the likelihood ratio index, is similar to the R2 in conventional regression model and is given by 1 ln L= ln Lo , where ln Lo is the log likelihood when all slope coefficients are zero and ln L is the log likelihood from the relevant specification. (iv) Specification tests carried out for heteroscedasticity take the form of variable addition tests. Orme (1987) suggests that heteroscedasticity can be detected by testing the significance of an extra set of constructed regressors in binary choice models. If one attempts to test for group-wise heteroscedasticity the constructed variable should be w ^ i ¼ ðxi bÞvi where vi is an index which takes different values for different countries ^ it ¼ ðxi bÞnit . LR (likelihood but does not change over time. In the case of the general form of heteroscedasticity the additional variable can be formed as m ratio statistic) reported for the general form of heteroscedasticity is derived by testing the significance of this constructed variable when nit is the level of foreign exchange reserves. The reason for using the reserves variable is that it may take very different values over time and between countries. Other variables such as the industrial production index are also used (not reported). The relevant v2 table values are v2ð1Þ ¼ 6:64 (1%) and v2ð1Þ ¼ 3:84 (5%). (v) RESET (LR) reports tests of functional form mis-specification for binary choice models following Horowitz (1994), that are based on testing the significance of the squared fitted values from relevant specifications. (vi) LR reports the likelihood ratio statistic arising from testing specification 1 (reported in Table 2) to specification 2 (reported in Table 3). The relevant v2 table values for both these LR tests are same as given in (iv).
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
Dependent variable: P ðzit ¼ 1Þ
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
499
The results presented in Table 2 suggest that rises in interest rates in Germany and falls in domestic industrial production increased the odds of a parity change during the ERM. In addition, movements in industrial production take three months to exert influence on the likelihood of a devaluation. That is, the policy maker may still prefer to stay in the fixed exchange rate regime even when there are output losses because the potential benefits of the fixed rate regime may outweigh these costs. This may also result from the fact that output losses turn into job losses only in time. Further lag values of i are also significant and have alternating signs suggesting potential explanatory powers for the changes in this variable on the likelihood of a parity change. As stated above, the right-hand side variable set includes two measures of competitiveness, the DM/Dollar exchange rate and the inflation in Germany both of which are expected to reduce the odds of a devaluation. Our results confirm the view that a rise in DM/Dollar exchange rate––a sign of competitiveness of the home country versus the Ôrest of the world’––reduces the need for a realignment within the ERM. 10 As to the other measure, although the parameter estimate of pt2 is negative as expected, this variable is significant only at 7% level in the first specification and is not significantly different from zero in other specifications. In addition to these measures of competitiveness, the specification reported in Column 1 also includes trade balance variable which reflects, to some extent, the outcome of the evolution in the above mentioned competitiveness measures. As is seen from the estimated coefficients the trade balance variable is not significant. Column 2 reports the estimation results when this variable is replaced by exports and imports separately. The reason for including the trade balance variable (and exports and imports variables in Column 2) as well as the competitiveness variables is due to the potential non-contemporaneous nature of the relationship between the competitiveness measures and the trade variables and/or a potentially different lag structure of the effects of these two sets of variables. This is confirmed by the empirical results presented in Table 2. The positive and significant coefficient of mt3 is testament to the pressure put on the exchange rate as a result of a rise in imports. This is an interesting finding which may reflect that what happens in the export and/or import sectors may carry more weight than the balancing position in the current account in creating pressures on the exchange rate. One of the other variables that is expected to have an impact on the odds of a parity change is the level of foreign exchange reserves. As explained above this is one of the key determinants of the timing of speculative attacks in the framework of the earlier currency crises models (see, for example, Krugman, 1979; Flood and Garber, 1984). In the optimising currency crises models it has much less significance. However, authorities could use reserves to fend off untimely attacks. As is seen from
10
It was argued that the depreciating dollar in autumn 1992 led to ERM members losing competitiveness, which put some pressure on the system (The Economist, 19 September, 1992 and Eichengreen and Wyplosz (1993)).
500
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
the estimates, the level of foreign exchange reserves in the previous month has a negative and significant effect on the probability of parity changes in both specifications. One of the main features of the optimising currency crises models is the explicit treatment of the policy makers’ preferences. As a result one would expect governments with different political ideologies to follow different policies once in office and thus to have different impacts on the economy (see, for example, Hibbs, 1977; Alesina, 1988; Alesina and Roubini, 1990). More specifically, left-wing parties are expected to be less concerned about inflation and be more expansionary than right-wing parties. Therefore, left-wing governments are expected to be less concerned about the inflationary consequences of devaluations and would be more welcoming about their output raising effects. The variable pd is included to capture these political effects. However, we do not find any support for this hypothesis in the ERM period. 11 In addition to these macroeconomic variables, we have introduced a dummy variable to test for the potential link between the realisation of a parity change in the previous three months and its probability in the current period. This is measured by a binary variable, dev, which takes the value 1 if there was a parity change in either t 1, t 2 or t 3, and 0 otherwise. The negative and significant parameter estimate of dev in all specifications suggests that there is a significantly reduced likelihood of a devaluation in the current month if the currency in question is devalued in the past 3 months. Also among the regressors in both specifications is another binary variable included to examine if the likelihood of regime changes is any different in different sub-periods of the sample period, 1979–1992. The period after 1987 is usually referred to as the ‘‘New EMS’’ over which there was a higher degree of capital mobility but also a much increased borrowing facility among the member central banks. The binary variable, new, (for the ÔNew EMS’), therefore, takes the value 1 for all months after August 1987 and zero otherwise. The negative and significant coefficient of new suggests that in spite of the removal of capital controls, this period was associated with more stability not less. 12 It is seen from the estimation results reported in Table 2 that the coefficients of most of the variables have alternating signs in consecutive periods. The summed values reported in adjacent columns to estimation results are close to zero and although they have the expected signs it is not possible to reject the null hypothesis that they are not significantly different from zero. Taken together with the alternating signs of the coefficients, this suggests that it might be the rates of changes of these variables that affect the odds of a parity change. Therefore, we have re-estimated the model by
11
One reason for the insignificance of this variable might be the almost universal adoption of antiinflationary policies during the 1980s, following unprecedented levels of inflation experienced in the wake of the oil crises. This led to various forms of disinflationary policies on the part of both the left and rightwing governments. Among those was the ÔFranc fort’ policy of the left-wing French government in the early 1980s which even led to the abandonment of a program of domestic expansion in order to prevent the devaluation of the franc (see, for example, Sachs and Wyplosz, 1986). 12 This point is also highlighted by Artus and Bourguinat (1994).
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
501
replacing the regressors with their rates of change. New results are reported in Table 3. The coefficient estimates suggest that the probability of a devaluation in this sample was determined positively by a rise in the growth rate of German interest rates and a fall in that of the DM/Dollar exchange rate. However, the results also reveal that the growth rate of the foreign exchange reserves does not have any significant effect on the probability of a parity change. Similarly, contrary to the effects of their levels, growth rates of industrial production increases the likelihood of a parity change while that of imports reduces it. As before, the occurrence of a devaluation in the previous three months reduces this probability. Likewise, estimation results suggest that the odds of parity change is reduced in the new-EMS period. As to the preferability of one model over the other, we carry out RESET tests of mis-specifications for both models (see, for example, Horowitz (1994) on RESET tests for binary choice models). As neither of the models could be rejected on these bases, we resorted to the likelihood ratio tests to choose between the two, the results of which are now reported in Table 2. The inference from this favours the first model. Table 3 Panel data conditional logit estimates of the determinants of realignments for all nine ERM countries: rates of changes Dependent variable: P ðzit ¼ 1Þ dit1 dit2 dpt1 dpt2 dyt1 dyt2 ddt1 ddt2 dtbt1 dtbt2 det1 det2 dmt1 dmt2 drest1 drest2 pd dev new N Nr LRI LR (1), groupwise Het. LR (1), general Het. RESET (LR)
(1)
(2)
1.94 )1.63 0.66 )0.89 37.43 57.62 )11.24 )12.86 )0.02 )0.01
(4.11) ()2.84) (1.43) ()1.81) (1.33) (2.06) ()2.31) ()2.72) ()1.36) ()0.21)
0.31 )0.23 95.05 24.10 )0.03
– – – – )16.52 )9.96 0.61 )1.91 )3.27
()1.49) ()0.83) (1.34) ()2.43) ()4.23)
1042 43 0.25 0.12 3.81 0.006
Note: All the test statistics are as specified above.
)26.48
2.07 )1.69 0.74 )1.07 33.98 66.08 )10.87 )13.01 – – 15.71 19.77 )24.45 )53.89 )19.48 )11.25 0.64 )1.82 )3.35 1042 43 0.29 0.37 1.52 0.95
(4.24) ()2.89) (1.49) ()2.06) (1.07) (2.14) ()2.14) ()2.50)
(1.25) (1.52) ()1.60) ()3.03) ()1.73) ()0.99) (1.43) ()2.25) ()4.17)
0.38 )0.33 100.06 )23.88
35.48 )78.34 )30.73
502
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
estimated probabilities
0.2
0.15
0.1
0.05
0
1979 1980 1982 1983 1984 1986 1987 1988 1990 1991 1992
years Fig. 1. Belgium.
estimated probabilities
0.25 0.2 0.15 0.1 0.05 0
1979 1980 1982 1983 1984 1986 1987 1988 1990 1991 1992
years Fig. 2. Denmark.
Alternatively, one can also use LRI indices of individual models (see, for example, Wooldridge, 2000, Chapters 9 and 17). Comparing LRI indices reported in both tables produces the same ranking between these two models. Now turn to the estimated probabilities of realignments in Figs. 1–6, which are the plots of estimated and actual devaluations for a group of ERM members. 13 These figures reveal that the estimated devaluations match the actual devaluations reasonably well. It is also clear that macroeconomic variables were not cause for much concern in the aftermath of 1987–88 for the long-term members as compared with their earlier experience. In contrast, the initial periods of the new members’ participation in the ERM seem turbulent. The estimated devaluation probabilities are non-zero for most of their membership and actual devaluations are matched quite 13 Given the result of the LR test and the comparison of the LRI’s, the first specification is used in estimating these probabilities.
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
503
estimated probabilities
0.2
0.15
0.1
0.05
0
1979 1980 1982 1983 1984 1986 1987 1988 1990 1991 1992
years Fig. 3. France. 0.16
estimated probabilities
0.14 0.12 0.1 0.08 0.06 0.04 0.02 1979 1980 1981 1982 1984 1985 1986 1987 1988 1989 1991 1992
years Fig. 4. Italy.
well by the model estimates. Figs. 5 and 6 suggest that there was sufficient basis for potential parity changes for the new-members during 1992. For the UK, for example, there were ‘‘fundamental signals’’ as early as May 1992. Indeed as the recession deepened throughout the spring there were calls for an interest rate cut. The findings in this paper together with a background of deepening recession throughout the UK’s membership calls for alternative ways of measuring the credibility of parities. Masson (1995), for example, shows that the credibility of Sterling was low much earlier than September 1992 by incorporating an unemployment measure into the credibility estimations.
4. Conclusions There are two main findings of this study. First of all, devaluations during the ERM were strongly influenced by movements in industrial production,
504
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
estimated probabilities
0.5 0.4 0.3 0.2 0.1 0
1991 1991 1991 1991 1991 1991 1992 1992 1992 1992 1992 y
Fig. 5. UK.
estimated probabilities
0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
1989 1990 1990 1990 1991 1991 1991 1992 1992 1992
years Fig. 6. Spain.
competitiveness, foreign interest rates and imports as well as in foreign exchange reserves. This piece of evidence is consistent with optimising models of currency crisis where the well-being of policy makers are related to these variables. In fact, this link may be stronger in the empirical analyses of non-ERM pegs. This is because the link between macroeconomic variables and the optimal response by the authorities is weakened by the institutional framework of the ERM which required difficult negotiations with other participants at times of parity changes. Having such an institutional framework is, however, precisely the factor that renders this kind of systems more credible and less vulnerable. In addition, we show that pressure on currencies develops over time. More specifically, the probability of devaluation in the present period is influenced by macroeconomic developments in up to past three months. These findings lead one to believe that a country’s ability to operate a suc-
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
505
cessful exchange rate policy is closely linked to macroeconomic variables and to the economic and political structure required to implement such a policy. Having established the link between the probability of regime switches and the evolution of macroeconomic variables that defines the well-being of the policy maker, the obvious policy implication stands out. Fixed exchange rate regimes can be prolonged by raising the benefits and/or reducing the costs associated with them. One can argue that the reason why it took so long for the ERM crises of 1992–1993 to emerge after the start of the Europe-wide recession was high benefits of maintaining the status quo especially in order to qualify for joining the core countries of the system in shaping the future of an integrated Europe.
Appendix A. Data The monthly data series used in estimation were taken from two separate sources. The exchange rate data were from the Bank of International Settlements (BIS) database. The monthly averages of the nominal exchange rates were calculated by using ECU exchange rates at 2:15 pm Brussels time (prior to September 1988, at 2:30 pm) as communicated by the Commission of the European Communities. The rates are expressed as national currency units per ECU. ECU central rates were taken from the 1994 issue of European Economy, published by the European Commission Directorate General for Economic and Financial Affairs. Data were collected for Germany, and the six long-term members of the ERM: Belgium, Denmark, France, Italy, Netherlands, Ireland and the three new members; the United Kingdom, Spain and Portugal. Data on the long-term members are for the period March 1979–December 1992 except for Italy for which the sample period ends in September 1992 when the country suspended its ERM membership. For the new members samples vary. For the United Kingdom the sample period is defined by the length of the country’s participation in the system; October 1990–September 1992. For Spain and Portugal the samples start with their entry to the mechanism, June 1989 and April 1992 respectively and end in December 1992. The second set of data is from International Financial Statistics (IFS) of the International Monetary Fund. Reserves are the IFS foreign assets series. Interest rates are short term money market rates which are the rates at which short term borrowing is affected between financial institutions. Inflation is the rate of change of the consumer price index. The trade balance is the difference between f.o.b exports and c.i.f imports. Output is the monthly industrial production index (1985 ¼ 100).
References Alesina, A., 1988. Macroeconomics and politics. NBER Macroeconomics Annual, 13–61. Alesina, A., Roubini, N., 1990. Political cycles in OECD economies. Review of Economic Studies 59, 633– 688.
506
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
Andersen, T., 1998. Shocks and the viability of fixed exchange rate regime. Open Economies Review 9, 139–156. Artus, P., Bourguinat, H., 1994. The stability of the EMS. In: Alfred, S. (Ed.), European Monetary Integration. Longman. Baltagi, B.H., 1995. Econometric Analysis of Panel Data. John Wiley. Bensaid, B., Jeanne, O., 1997. The instability of fixed exchange rate regimes when raising the nominal interest rates is costly. European Economic Review 41, 1461–1478. Burda, M., Wyplosz, C., 1997. Macroeconomics: A European Text. Oxford University Press. Caramazza, F., 1993. French–German interest rate differentials and time varying realignment risk. IMF Staff Papers 40, 567–583. Chen, Z., Giovannini, A., 1993. The determinants of realignments expectations under the EMS: Some empirical irregularities. NBER Working Paper 4291. Davies, G., Vines, D., 1995. Equilibrium currency crises: Are multiple equilibria self-fulfilling or history dependent? CEPR Discussion Paper 1239. Edin, P.A., Vredin, A., 1993. Devaluation risk in target zones: Evidence from the Nordic countries. The Economic Journal 103, 161–175. Eichengreen, B., Wyplosz, C., 1993. The unstable EMS. Brookings Paper on Economic Activity (1), 51– 143. Eichengreen, B., Rose, A., Wyplosz, C., 1995. Exchange rate mayhem: The antecedents and aftermath of speculative attacks. Economic Policy 21, 251–312. Eichengreen, B., Rose, A., Wyplosz, C., 1996. Contagious currency crises: First tests. Scandinavian Journal of Economics 98, 463–484. Flood, R., Garber, P., 1984. Collapsing exchange rate regimes: Some linear examples. Journal of International Economics 17, 1–13. Flood, R., Marion, N., 1998. Perspectives on the recent currency crisis literature. NBER Working Papers No. 6380. Hibbs, D., 1977. Political parties and macroeconomic policy. American Political Science Review 71, 1467– 1487. Horowitz, J., 1994. Bootstrap-based critical values for the information matrix test. Journal of Econometrics 61, 395–411. Green, W., 1993. Econometric Analysis. Prentice Hall. Kaminsky, G., Reinhart, C., 1998. Financial crises in Asia and Latin America: Then and now. American Economic Review, Papers and Proceedings, vol. 88, pp. 444–448. Kaminsky, G., Lizondo, S., Reinhart, C., 1998. Leading indicators of currency crises. IMF Staff Papers 45, 1–48. Klein, M., Marion, N., 1997. Explaining the duration of exchange rate pegs. Journal of Development Economics 54, 387–404. Krugman, P., 1979. A theory of balance of payments crises. Journal of Money, Credit, and Banking 11, 311–325. Masson, P., 1995. Gaining and losing credibility: The case of the United Kingdom. The Economic Journal 105, 571–582. conomiques et Monetaires 43, 189– Obstfeld, M., 1994. The Logic of Currency Crises. Cahiers E 213. Obstfeld, M., 1996a. Models of currency crises with self-fulfilling features. European Economic Review 40, 1037–1047. Obstfeld, M., 1996b. Comment on ‘‘Are currency crises self-fulfilling’’. In: Krugman, P. (Ed.), NBER Macroeconomics Annual, pp. 393–403. Orme, C., 1987. Specification tests for binary data models, University of Nottingham, Department of Economics, Discussion Paper, No. 65. Ozkan, G., Sutherland, A., 1995. Policy measure to avoid a currency crisis. The Economic Journal 105, 510–519. Ozkan, G., Sutherland, A., 1998. A currency crisis model with an optimising policy maker. Journal of International Economics 44, 339–364.
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
507
Radelet, S., Sachs, J., 1998. The East Asian financial crisis: Diagnosis, remedies, prospects. Brookings Papers on Economic Activity (1), pp. 1–76. Rose, A., Svensson, L., 1994. European exchange rate credibility before the fall. European Economic Review 38, 1185–1216. Sachs, J., Wyplosz, C., 1986. The economic consequences of President Mitterand. Economic Policy 2, 261– 322. Thomas, A., 1994. Expected devaluation and economic fundamentals. IMF Staff Papers 41, 262–285. Wooldridge, J., 2000. Introductory Econometrics. South Western College Publishing.
Explaining ERM realignments: Insights from optimising models of currency crises F. Gulcin Ozkan
*
Department of Economics and Related Studies, University of York, Heslington, York YO10 5DD, UK CEPR, London, UK Received 19 April 1999; accepted 6 August 2002
Abstract This paper attempts to provide empirical evidence on the determinants of the realignments throughout the European exchange rate mechanism (ERM). Motivated by the implications of optimising currency crisis models, we relate the probability of ‘‘crises’’ to a set of macroeconomic fundamentals. By using a conditional binominal logit model we show that regime switches are strongly influenced by movements in industrial production, foreign interest rates, competitiveness and imports as well as in foreign exchange reserves. These findings are consistent with the general propositions of recent currency crises models. Ó 2003 Elsevier Inc. All rights reserved. JEL classification: F31; F32 Keywords: Currency crisis; Parity changes; Probability of devaluations; ERM
1. Introduction This paper attempts to provide more insights into the understanding of the determinants of currency crises following the recent turmoil in international financial markets. Currency and financial crises experienced by Mexico in 1994–1995, by Indonesia, Korea, Malaysia, the Phillippines and Thailand in 1997 and more recently by Turkey in 2001 and Argentina in 2002 all revived the interest in understanding the roots of currency collapses. The issue of currency crises has already been a lively research area and has been a subject of investigation in international finance literature
*
Tel.: +44-01904-434673; fax: +44-01904-433759. E-mail address: [email protected] (F.G. Ozkan).
0164-0704/$ - see front matterÓ 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.jmacro.2002.08.001
492
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
since the seminal paper by Krugman (1979). The main proposition of the early literature (generally referred to as first generation models) is that balance of payments crises are precipitated by policy imbalances such as lax monetary and fiscal policies. In addition, the level of foreign exchange reserves is argued to be the main determinant of the duration of a fixed exchange rate regime. The events in the European exchange rate mechanism (henceforth the ERM) in 1992 and 1993 posed serious problems for the propositions of the early models. Although the crises which preceded the exit of the Pound and the Italian Lira from the ERM involved some loss of reserves reserve adequacy was clearly not an issue for these countries which could borrow almost unlimited amounts of foreign exchange due to the provisions of the European exchange rate system. In addition, it might have been possible for these governments to raise interest rates to whatever level necessary to avoid the crises. The fact that these countries chose to leave the system instead was widely interpreted as an indication that the authorities’ objective functions included some other key variables. 1 These observations might indicate that the decision to leave the ERM was to some extent an optimising decision on the part of those countries rather than an action only forced upon them by either policy imbalances or speculators. This has been at the centre of the new approach to the currency crises. These so-called optimising models of currency crises view regime changes as conscious decisions taken by policy makers who continuously weigh the marginal costs versus benefits of being in a fixed exchange rate regime. Main examples of these are Obstfeld (1994, 1996a), Ozkan and Sutherland (1995, 1998), Davies and Vines (1995), Bensaid and Jeanne (1997) and Andersen (1998) among others. These second generation models all share the common feature of modelling policy makers as having well-defined utility functions and thus clear preferences in terms of being in a particular exchange rate regime. The main implication of these models is that a fixed exchange rate regime lasts as long as the policy maker derives more utility from being in it than in either changing the rate or leaving the regime altogether. In other words, a country’s commitment to the fixed exchange rate regime is not state-invariant (see, for example, Flood and Marion, 1998). Motivated by the implications of these new currency crises models, this paper empirically investigates the roots of Ôcurrency crises’ during the ERM period by extending the set of potential determinants, as suggested by this recent literature. Our data set covers the ERM experience between 1979 and 1992. The reasons for constraining the panel to the ERM members and leaving out the Latin American and East Asian experiences are two-fold. First of all, in the case of recent Latin American and East Asian crises, currency collapses can not be analysed in isolation from financial and banking crises (see, for example, Radelet and Sachs, 1998; Kaminsky and Reinhart, 1998). Secondly, although empirical studies on the ERM abound most of the existing research on the European experience focuses on the determinants of expected
1
High interest rates are a huge political cost in Britain due to the fact that a large proportion of mortgages are at variable rate. In the case of Italy, repayments on public borrowing soar as a result of rising interest rates.
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
493
changes in the exchange rate (see, for example, Chen and Giovannini, 1993; Caramazza, 1993; Thomas, 1994; Rose and Svensson, 1994). 2 Here we explore the determinants of actual rather than expected realignments. 3 Therefore, by incorporating an extended set of potential variables we provide fresh evidence on the causes of crises during one of the important examples of pegged exchange rate regimes. The first practical task in carrying out this exercise is how to define Ôcurrency crises’ and distinguish between different crises events. Although it is straightforward to classify the events leading to the decisions of the UK and Italy to leave the ERM in September 1992 as ‘‘currency crises’’, the same is not true of a large number of other changes in central parities throughout the period 1979–1992. Despite this, in this paper we aim to study the sources of all regime switches. Therefore, we attempt to empirically analyse the macroeconomic determinants of pressure leading to all realignments of central parities during the life of the ERM. This is because all of these changes signal regime switches though to a varying degree. As stated above, the contribution of the optimising models of currency crises is to broaden the set of fundamentals that are derived from the objectives of and the constraints faced by the policy makers that affect their decision to leave the existing exchange rate regime. It is crucial to point out, however, that it is also possible for a shift in market expectations to push an otherwise sound economy into a crisis. As Flood and Marion (1998) point out, in such a situation anything that helps to coordinate speculators’ actions may bring about a sudden attack on the currency. This highlights the difficulty in empirically isolating between the implications of the first and second generation models. 4 This is, however, not the purpose of the current study. In this paper, we are mostly concerned with identifying common characteristics of devaluations that could represent sources of macroeconomic trade-offs that were faced by policy makers during the ERM. Affirmative evidence to this would be in line with general propositions of these new models that Ôcurrency crises’ are precipitated by a broader set of Ôfundamentals’ that affect the interactions between policy makers and the markets. 5 Our findings suggest that regime switches in a particular period are influenced by the movements in industrial production, 2 The two exceptions are Eichengreen et al. (1995, 1996). The first of these carries out a comprehensive study of the empirical regularities of macroeconomic data around a number of exchange rate events inclusive of those throughout the ERM. The second tests for the contagious nature of speculative attacks in a sample of 20 countries including the EMS members. 3 The estimates of the actual realignments reveal information about the policy makers’ preferences while that of the expected realignments reflect information about the market participants’ perception of those preferences. These two alternative approaches should yield similar results in terms of the timing of parity changes as long as expectations are rational and correctly measured. However, apart from the practical difficulties in measuring expectations, such factors as the lack of information on the part of the financial markets about the preferences and the constraints of policy makers may lead to incorrect inference about the actual preferences of the policy makers when using expected parity changes. 4 Obstfeld (1996b) also points to this identification problem. 5 A growing body of empirical literature has emerged from testing the implications of these new optimising models of currency crises. Kaminsky et al. (1998) provide a unifying framework and identify a set of macroeconomic variables as the Ôleading indicators of currency crises’ by examining the existing empirical evidence. Our set of regressors are among the main indicators identified by them.
494
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
foreign interest rates, competitiveness and imports as well as the foreign exchange reserves. The rest of this paper is organised as follows. Section 2 explains the empirical specification. The main findings are discussed in Section 3 and finally Section 4 concludes the paper. A brief definition of data is provided in the Appendix A.
2. Empirical specification The purpose of the empirical analysis here is to shed some light on the macroeconomic determinants of devaluations throughout the ERM period. As explained above we use ‘‘crisis’’ here as a general term referring to periods of pressure on the existing parity as well as more extreme cases. Therefore, in this context crisis for a particular country is defined as the realignment of the domestic currency’s central parity against the ECU. There were 43 devaluations between March 1979 and December 1992 in total. 6 A realignment is defined as a change in the central parities relative to the ECU. A devaluation is then effectively the fall in the value of the currency in question against those currencies whose bilateral parities against the ECU remained unchanged. The regressand is a binary choice variable which takes the value 1 for the months in which the domestic currency is devalued against the ECU and 0 otherwise. This suggests that our set of Ôcrises’ excludes unsuccessful speculative attacks. The main reason for this is that in this paper we specifically focus on identifying a set of macroeconomic variables the deterioration of which leads to a change in the actual value of the currency in question. In so doing, our results provide new evidence on the potential for a number of indicators to actually bring about the policy maker’s response. Our main focus as the determinants of actual parity changes is motivated by the elusive nature of the predictability of devaluations and the choice of regressors by the prominence of the policy makers’ preferences in the recent currency crises. The dates and sizes of all 43 realignments between 1979 and 1992 are listed in Table 1. A general discrete response model for realignments can be written as P ðzit ¼ 1Þ ¼ F ðai þ b0 xit Þ;
ð1Þ
where i ¼ 1; 2; . . . ; N and t ¼ 1; 2; . . . ; T and P ðzit ¼ 1) is the probability of devaluation for country i at time t and xit is the vector of macroeconomic variables that influence this probability. The ai ’s are country specific unobserved variables that are assumed to be constant over time. Since the sample of the current empirical analysis is the ERM countries over a period of their participation in the mechanism the data set has both time-series and cross-sectional features.
6
Although there are only 14 parity changes in the sample period (March 1979–December 1992 inclusive) the total number of devaluations is 43 since at each parity changes there was more than one country that experienced a change in the value of its currency.
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
495
Table 1 Dates and the sizes of ERM realignments relative to ECU, 1979–1992 Date 24.09.1979 30.10.1979 05.10.1981 22.02.1982 14.06.1982 21.03.1983 21.07.1985 07.04.1986 04.08.1986 12.01.1987 07.01.1990 14.09.1992 16.09.1992 23.11.1992
Currencies BF
DK
DM
FF
IP
IL
DG
SP
PE
UKS
0.0 0.0 0.0 )8.5 0.0 +1.5 +2.0 +1.0 0.0 +2.0 0.0 0.0 0.0 0.0
)2.9 )4.8 0.0 )3.0 0.0 +2.5 +2.5 +1.0 0.0 0.0 0.0 0.0 0.0 0.0
+2.0 0.0 +5.5 0.0 +4.25 +5.5 +2.0 +3.0 0.0 +3.0 0.0 0.0 0.0 0.0
0.0 0.0 )3.0 0.0 )5.75 )2.5 2.0 )3.0 0.0 0.0 0.0 0.0 0.0 0.0
0.0 0.0 0.0 0.0 0.0 )3.5 +2.0 0.0 )8.0 0.0 0.0 0.0 0.0 0.0
0.0 0.0 3.0 0.0 )2.75 )2.5 )6.0 0.0 0.0 0.0 )3.8 )7.0 s –
0.0 0.0 +5.5 0.0 +4.25 +3.5 +2.0 +3.0 0.0 +3.0 0.0 0.0 0.0 0.0
– – – – – – – – – –
– – – – – – – – – – –
– – – – – – – – – –
0.0 0.0 +6.0
0.0 e –
0.0 0.0 +5.0 +6.0
Notes: Source is Burda and Wyplosz (1997). BF is the Belgian Franc, DK is the Danish Kroner, DM is the German Mark, FF is the French Franc, IP is the Irish Punt, IL is the Italian Lira, DG is the Dutch Guilder, SP is the Spanish Peseta, PE is the Portuguese Escudo and UKS is the British Sterling. s stands for Italy’s suspension of its membership on 16, September 1992 while e is used for the UK’s exit from the system. Italy followed the UK in ending participation on the 22nd of September.
The logit model for the panel data to be used in this analysis can be represented by
7
P ðzit ¼ 1Þ ¼ F ai þ
L X
bi j ii;tj þ
j¼1
þ
L X j¼1
bej ei;tj þ
L X
bp j pi;tj þ
j¼1 L X j¼1
bmj mi;tj þ
L X j¼1
L X
byj yi;tj þ
L X
!
bdj di;tj
j¼1
bresj resi;tj ;
ð2Þ
j¼1
where the ai ’s are country-specific fixed effects, L is the maximum lag, i is interest rates in Germany, p is inflation in Germany, y is the industrial production index, d is the DM/Dollar exchange rate, e is exports, m is imports, res is foreign exchange reserves. It is well-known that (see, for example, Baltagi, 1995, p. 179; Green, 1993, p. 657) in the presence of these country specific effects ai ’s cannot be consistently estimated for a fixed T . In this case the estimation procedure should aim to maximize the conditional likelihood function instead of the unconditional likelihood function for the inconsistency of the maximum likelihood estimators of ai ’s not to be transmitted into the inconsistency of that of b’s. Therefore, in what follows we adopt a conditional logit model to estimate the probability of devaluations. 8
7 8
The choice of the fixed-effects version is motivated by the small number of cross-sectional units. Estimations are carried out by using the statistical software package STATA.
496
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
As to the set of variables, all variables are expressed in logs except interest rates and inflation. Interest rates and inflation in Germany are included because of the asymmetric functioning of the ERM in which Germany acted as the centre country throughout. The DM/Dollar exchange rate is added to consider the effect of competitiveness vis- a-vis the non-ERM countries on the existing ERM parities. This is because an increase in the DM/Dollar exchange rate, within the framework of fixed parities against the DM, amounts to an improvement in the member countries’ competitiveness vis- a-vis the Ôrest of the world’. This choice of the RHS variables is driven by the explicit implications of the above mentioned recent currency crises models. The main assertion that high interest rates in Germany were translated into output losses which were costly for member countries, as argued by Ozkan and Sutherland (1995, 1998) and Davies and Vines (1995) is tested by the inclusion of German interest rates and domestic industrial production index. The link between the terms of trade shocks and the viability of a fixed exchange rate regime is studied by Andersen (1998). It is straightforward to generalise this argument to a relationship between the trade performance and the pressure on the existing parities. This is tested by including competitiveness variables; foreign (German) inflation, and the DM/Dollar exchange rate and trade performance variables; trade balance, exports and imports. In addition to these, RHS variables set includes foreign exchange reserves variable and a number of binary variables. The model adopted here is a general one including several past values of the independent variables. The choice of such a dynamic approach is motivated by the notion that currency pressure can build up over longer periods of time than usually studied in the empirical literature. 9
3. Estimation results Eq. (2) is estimated by the conditional maximum likelihood method using monthly observations for a panel of all the ERM member countries including the six long-term members; Belgium, Denmark, France, Italy, Netherlands and Ireland as well as the ‘‘new members’’; Spain, Portugal and the UK. There are two specifications that are reported in Table 2. Column 1 reports the estimation results when trade performance is measured by the trade balance variable. We arrive at the specification reported in Column 2 by replacing the trade balance variable with exports and imports separately.
9
Edin and Vredin (1993) utilize a binominal probit model to estimate the probability of devaluations and its determinants for the Nordic Countries by using monthly data for the period 1978–1989. It is shown that output growth, the central parity and foreign exchange reserves were the main determinants of the probability of devaluations for this group of countries in the studied period. Similarly Klein and Marion (1997) estimate devaluation probabilities for 16 Latin American Countries for the period 1956–1991 by using a binominal logit model. The real exchange rate, the net asset position and the openness of the economy are shown to influence the probability of devaluing against the US dollar.
Table 2 Panel data conditional logit estimates of the determinants of realignments for all nine ERM countries (1)
it1 it2 it3 pt1 pt2 pt3 yt1 yt2 yt3 dt1 dt2 dt3 tbt1 tbt2 tbt3 et1 et2 et3 mt1 mt2 mt3 rest1 rest2 rest3 pd dev new
1.78 )2.91 1.51 0.24 )1.29 0.77 10.00 4.55 )16.55 )20.64 0.69 19.92 0.01 )0.01 )0.01 – – – – – – )5.92 3.72 0.94 0.23 )1.96 )2.64
(2) (3.65) ()3.44) (2.85) (0.49) ()1.79) (1.52) (1.51) (0.68) ()2.52) ()3.10) (0.07) (2.81) (0.55) ()0.15) ()0.66)
()2.12) (0.96) (0.34) (0.50) ()2.51) ()2.38)
0.38
)0.28 )2.00 )0.03 )0.01
)1.26
1.83 )3.09 1.51 0.13 )1.20 0.93 12.31 5.76 )20.17 )20.44 4.63 15.62 – – – 1.07 0.90 )1.98 )3.13 )3.76 6.89 )6.31 3.64 1.61 0.37 )1.72 )2.73
(3.56) ()3.39) (2.68) (0.24) ()1.49) (1.68) (1.64) (0.82) ()2.80) ()2.97) (0.43) (2.03)
(0.53) (0.45) ()0.92) ()1.45) ()1.66) (2.55) ()2.18) (0.94) (0.58) (0.78) ()2.17) ()1.86)
0.25
)0.14 )2.1 )0.19
)0.01
0.00
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
Dependent variable: P ðzit ¼ 1Þ
)1.06
497
498
Table 2 (continued) (1)
(2)
N Nr LRI LR(1), groupwise Het. LR(1), general Het. RESET(LR) LR
1042 43 0.29 0.07 0.53 1.48 10.26
1042 43 0.32 0.67 0.13 0.07 11.11
Notes: (i) t-ratios are in the parentheses. (ii) N r is the number of realignments in the relevant sample period. (iii) LRI, the likelihood ratio index, is similar to the R2 in conventional regression model and is given by 1 ln L= ln Lo , where ln Lo is the log likelihood when all slope coefficients are zero and ln L is the log likelihood from the relevant specification. (iv) Specification tests carried out for heteroscedasticity take the form of variable addition tests. Orme (1987) suggests that heteroscedasticity can be detected by testing the significance of an extra set of constructed regressors in binary choice models. If one attempts to test for group-wise heteroscedasticity the constructed variable should be w ^ i ¼ ðxi bÞvi where vi is an index which takes different values for different countries ^ it ¼ ðxi bÞnit . LR (likelihood but does not change over time. In the case of the general form of heteroscedasticity the additional variable can be formed as m ratio statistic) reported for the general form of heteroscedasticity is derived by testing the significance of this constructed variable when nit is the level of foreign exchange reserves. The reason for using the reserves variable is that it may take very different values over time and between countries. Other variables such as the industrial production index are also used (not reported). The relevant v2 table values are v2ð1Þ ¼ 6:64 (1%) and v2ð1Þ ¼ 3:84 (5%). (v) RESET (LR) reports tests of functional form mis-specification for binary choice models following Horowitz (1994), that are based on testing the significance of the squared fitted values from relevant specifications. (vi) LR reports the likelihood ratio statistic arising from testing specification 1 (reported in Table 2) to specification 2 (reported in Table 3). The relevant v2 table values for both these LR tests are same as given in (iv).
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
Dependent variable: P ðzit ¼ 1Þ
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
499
The results presented in Table 2 suggest that rises in interest rates in Germany and falls in domestic industrial production increased the odds of a parity change during the ERM. In addition, movements in industrial production take three months to exert influence on the likelihood of a devaluation. That is, the policy maker may still prefer to stay in the fixed exchange rate regime even when there are output losses because the potential benefits of the fixed rate regime may outweigh these costs. This may also result from the fact that output losses turn into job losses only in time. Further lag values of i are also significant and have alternating signs suggesting potential explanatory powers for the changes in this variable on the likelihood of a parity change. As stated above, the right-hand side variable set includes two measures of competitiveness, the DM/Dollar exchange rate and the inflation in Germany both of which are expected to reduce the odds of a devaluation. Our results confirm the view that a rise in DM/Dollar exchange rate––a sign of competitiveness of the home country versus the Ôrest of the world’––reduces the need for a realignment within the ERM. 10 As to the other measure, although the parameter estimate of pt2 is negative as expected, this variable is significant only at 7% level in the first specification and is not significantly different from zero in other specifications. In addition to these measures of competitiveness, the specification reported in Column 1 also includes trade balance variable which reflects, to some extent, the outcome of the evolution in the above mentioned competitiveness measures. As is seen from the estimated coefficients the trade balance variable is not significant. Column 2 reports the estimation results when this variable is replaced by exports and imports separately. The reason for including the trade balance variable (and exports and imports variables in Column 2) as well as the competitiveness variables is due to the potential non-contemporaneous nature of the relationship between the competitiveness measures and the trade variables and/or a potentially different lag structure of the effects of these two sets of variables. This is confirmed by the empirical results presented in Table 2. The positive and significant coefficient of mt3 is testament to the pressure put on the exchange rate as a result of a rise in imports. This is an interesting finding which may reflect that what happens in the export and/or import sectors may carry more weight than the balancing position in the current account in creating pressures on the exchange rate. One of the other variables that is expected to have an impact on the odds of a parity change is the level of foreign exchange reserves. As explained above this is one of the key determinants of the timing of speculative attacks in the framework of the earlier currency crises models (see, for example, Krugman, 1979; Flood and Garber, 1984). In the optimising currency crises models it has much less significance. However, authorities could use reserves to fend off untimely attacks. As is seen from
10
It was argued that the depreciating dollar in autumn 1992 led to ERM members losing competitiveness, which put some pressure on the system (The Economist, 19 September, 1992 and Eichengreen and Wyplosz (1993)).
500
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
the estimates, the level of foreign exchange reserves in the previous month has a negative and significant effect on the probability of parity changes in both specifications. One of the main features of the optimising currency crises models is the explicit treatment of the policy makers’ preferences. As a result one would expect governments with different political ideologies to follow different policies once in office and thus to have different impacts on the economy (see, for example, Hibbs, 1977; Alesina, 1988; Alesina and Roubini, 1990). More specifically, left-wing parties are expected to be less concerned about inflation and be more expansionary than right-wing parties. Therefore, left-wing governments are expected to be less concerned about the inflationary consequences of devaluations and would be more welcoming about their output raising effects. The variable pd is included to capture these political effects. However, we do not find any support for this hypothesis in the ERM period. 11 In addition to these macroeconomic variables, we have introduced a dummy variable to test for the potential link between the realisation of a parity change in the previous three months and its probability in the current period. This is measured by a binary variable, dev, which takes the value 1 if there was a parity change in either t 1, t 2 or t 3, and 0 otherwise. The negative and significant parameter estimate of dev in all specifications suggests that there is a significantly reduced likelihood of a devaluation in the current month if the currency in question is devalued in the past 3 months. Also among the regressors in both specifications is another binary variable included to examine if the likelihood of regime changes is any different in different sub-periods of the sample period, 1979–1992. The period after 1987 is usually referred to as the ‘‘New EMS’’ over which there was a higher degree of capital mobility but also a much increased borrowing facility among the member central banks. The binary variable, new, (for the ÔNew EMS’), therefore, takes the value 1 for all months after August 1987 and zero otherwise. The negative and significant coefficient of new suggests that in spite of the removal of capital controls, this period was associated with more stability not less. 12 It is seen from the estimation results reported in Table 2 that the coefficients of most of the variables have alternating signs in consecutive periods. The summed values reported in adjacent columns to estimation results are close to zero and although they have the expected signs it is not possible to reject the null hypothesis that they are not significantly different from zero. Taken together with the alternating signs of the coefficients, this suggests that it might be the rates of changes of these variables that affect the odds of a parity change. Therefore, we have re-estimated the model by
11
One reason for the insignificance of this variable might be the almost universal adoption of antiinflationary policies during the 1980s, following unprecedented levels of inflation experienced in the wake of the oil crises. This led to various forms of disinflationary policies on the part of both the left and rightwing governments. Among those was the ÔFranc fort’ policy of the left-wing French government in the early 1980s which even led to the abandonment of a program of domestic expansion in order to prevent the devaluation of the franc (see, for example, Sachs and Wyplosz, 1986). 12 This point is also highlighted by Artus and Bourguinat (1994).
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
501
replacing the regressors with their rates of change. New results are reported in Table 3. The coefficient estimates suggest that the probability of a devaluation in this sample was determined positively by a rise in the growth rate of German interest rates and a fall in that of the DM/Dollar exchange rate. However, the results also reveal that the growth rate of the foreign exchange reserves does not have any significant effect on the probability of a parity change. Similarly, contrary to the effects of their levels, growth rates of industrial production increases the likelihood of a parity change while that of imports reduces it. As before, the occurrence of a devaluation in the previous three months reduces this probability. Likewise, estimation results suggest that the odds of parity change is reduced in the new-EMS period. As to the preferability of one model over the other, we carry out RESET tests of mis-specifications for both models (see, for example, Horowitz (1994) on RESET tests for binary choice models). As neither of the models could be rejected on these bases, we resorted to the likelihood ratio tests to choose between the two, the results of which are now reported in Table 2. The inference from this favours the first model. Table 3 Panel data conditional logit estimates of the determinants of realignments for all nine ERM countries: rates of changes Dependent variable: P ðzit ¼ 1Þ dit1 dit2 dpt1 dpt2 dyt1 dyt2 ddt1 ddt2 dtbt1 dtbt2 det1 det2 dmt1 dmt2 drest1 drest2 pd dev new N Nr LRI LR (1), groupwise Het. LR (1), general Het. RESET (LR)
(1)
(2)
1.94 )1.63 0.66 )0.89 37.43 57.62 )11.24 )12.86 )0.02 )0.01
(4.11) ()2.84) (1.43) ()1.81) (1.33) (2.06) ()2.31) ()2.72) ()1.36) ()0.21)
0.31 )0.23 95.05 24.10 )0.03
– – – – )16.52 )9.96 0.61 )1.91 )3.27
()1.49) ()0.83) (1.34) ()2.43) ()4.23)
1042 43 0.25 0.12 3.81 0.006
Note: All the test statistics are as specified above.
)26.48
2.07 )1.69 0.74 )1.07 33.98 66.08 )10.87 )13.01 – – 15.71 19.77 )24.45 )53.89 )19.48 )11.25 0.64 )1.82 )3.35 1042 43 0.29 0.37 1.52 0.95
(4.24) ()2.89) (1.49) ()2.06) (1.07) (2.14) ()2.14) ()2.50)
(1.25) (1.52) ()1.60) ()3.03) ()1.73) ()0.99) (1.43) ()2.25) ()4.17)
0.38 )0.33 100.06 )23.88
35.48 )78.34 )30.73
502
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
estimated probabilities
0.2
0.15
0.1
0.05
0
1979 1980 1982 1983 1984 1986 1987 1988 1990 1991 1992
years Fig. 1. Belgium.
estimated probabilities
0.25 0.2 0.15 0.1 0.05 0
1979 1980 1982 1983 1984 1986 1987 1988 1990 1991 1992
years Fig. 2. Denmark.
Alternatively, one can also use LRI indices of individual models (see, for example, Wooldridge, 2000, Chapters 9 and 17). Comparing LRI indices reported in both tables produces the same ranking between these two models. Now turn to the estimated probabilities of realignments in Figs. 1–6, which are the plots of estimated and actual devaluations for a group of ERM members. 13 These figures reveal that the estimated devaluations match the actual devaluations reasonably well. It is also clear that macroeconomic variables were not cause for much concern in the aftermath of 1987–88 for the long-term members as compared with their earlier experience. In contrast, the initial periods of the new members’ participation in the ERM seem turbulent. The estimated devaluation probabilities are non-zero for most of their membership and actual devaluations are matched quite 13 Given the result of the LR test and the comparison of the LRI’s, the first specification is used in estimating these probabilities.
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
503
estimated probabilities
0.2
0.15
0.1
0.05
0
1979 1980 1982 1983 1984 1986 1987 1988 1990 1991 1992
years Fig. 3. France. 0.16
estimated probabilities
0.14 0.12 0.1 0.08 0.06 0.04 0.02 1979 1980 1981 1982 1984 1985 1986 1987 1988 1989 1991 1992
years Fig. 4. Italy.
well by the model estimates. Figs. 5 and 6 suggest that there was sufficient basis for potential parity changes for the new-members during 1992. For the UK, for example, there were ‘‘fundamental signals’’ as early as May 1992. Indeed as the recession deepened throughout the spring there were calls for an interest rate cut. The findings in this paper together with a background of deepening recession throughout the UK’s membership calls for alternative ways of measuring the credibility of parities. Masson (1995), for example, shows that the credibility of Sterling was low much earlier than September 1992 by incorporating an unemployment measure into the credibility estimations.
4. Conclusions There are two main findings of this study. First of all, devaluations during the ERM were strongly influenced by movements in industrial production,
504
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
estimated probabilities
0.5 0.4 0.3 0.2 0.1 0
1991 1991 1991 1991 1991 1991 1992 1992 1992 1992 1992 y
Fig. 5. UK.
estimated probabilities
0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
1989 1990 1990 1990 1991 1991 1991 1992 1992 1992
years Fig. 6. Spain.
competitiveness, foreign interest rates and imports as well as in foreign exchange reserves. This piece of evidence is consistent with optimising models of currency crisis where the well-being of policy makers are related to these variables. In fact, this link may be stronger in the empirical analyses of non-ERM pegs. This is because the link between macroeconomic variables and the optimal response by the authorities is weakened by the institutional framework of the ERM which required difficult negotiations with other participants at times of parity changes. Having such an institutional framework is, however, precisely the factor that renders this kind of systems more credible and less vulnerable. In addition, we show that pressure on currencies develops over time. More specifically, the probability of devaluation in the present period is influenced by macroeconomic developments in up to past three months. These findings lead one to believe that a country’s ability to operate a suc-
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
505
cessful exchange rate policy is closely linked to macroeconomic variables and to the economic and political structure required to implement such a policy. Having established the link between the probability of regime switches and the evolution of macroeconomic variables that defines the well-being of the policy maker, the obvious policy implication stands out. Fixed exchange rate regimes can be prolonged by raising the benefits and/or reducing the costs associated with them. One can argue that the reason why it took so long for the ERM crises of 1992–1993 to emerge after the start of the Europe-wide recession was high benefits of maintaining the status quo especially in order to qualify for joining the core countries of the system in shaping the future of an integrated Europe.
Appendix A. Data The monthly data series used in estimation were taken from two separate sources. The exchange rate data were from the Bank of International Settlements (BIS) database. The monthly averages of the nominal exchange rates were calculated by using ECU exchange rates at 2:15 pm Brussels time (prior to September 1988, at 2:30 pm) as communicated by the Commission of the European Communities. The rates are expressed as national currency units per ECU. ECU central rates were taken from the 1994 issue of European Economy, published by the European Commission Directorate General for Economic and Financial Affairs. Data were collected for Germany, and the six long-term members of the ERM: Belgium, Denmark, France, Italy, Netherlands, Ireland and the three new members; the United Kingdom, Spain and Portugal. Data on the long-term members are for the period March 1979–December 1992 except for Italy for which the sample period ends in September 1992 when the country suspended its ERM membership. For the new members samples vary. For the United Kingdom the sample period is defined by the length of the country’s participation in the system; October 1990–September 1992. For Spain and Portugal the samples start with their entry to the mechanism, June 1989 and April 1992 respectively and end in December 1992. The second set of data is from International Financial Statistics (IFS) of the International Monetary Fund. Reserves are the IFS foreign assets series. Interest rates are short term money market rates which are the rates at which short term borrowing is affected between financial institutions. Inflation is the rate of change of the consumer price index. The trade balance is the difference between f.o.b exports and c.i.f imports. Output is the monthly industrial production index (1985 ¼ 100).
References Alesina, A., 1988. Macroeconomics and politics. NBER Macroeconomics Annual, 13–61. Alesina, A., Roubini, N., 1990. Political cycles in OECD economies. Review of Economic Studies 59, 633– 688.
506
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
Andersen, T., 1998. Shocks and the viability of fixed exchange rate regime. Open Economies Review 9, 139–156. Artus, P., Bourguinat, H., 1994. The stability of the EMS. In: Alfred, S. (Ed.), European Monetary Integration. Longman. Baltagi, B.H., 1995. Econometric Analysis of Panel Data. John Wiley. Bensaid, B., Jeanne, O., 1997. The instability of fixed exchange rate regimes when raising the nominal interest rates is costly. European Economic Review 41, 1461–1478. Burda, M., Wyplosz, C., 1997. Macroeconomics: A European Text. Oxford University Press. Caramazza, F., 1993. French–German interest rate differentials and time varying realignment risk. IMF Staff Papers 40, 567–583. Chen, Z., Giovannini, A., 1993. The determinants of realignments expectations under the EMS: Some empirical irregularities. NBER Working Paper 4291. Davies, G., Vines, D., 1995. Equilibrium currency crises: Are multiple equilibria self-fulfilling or history dependent? CEPR Discussion Paper 1239. Edin, P.A., Vredin, A., 1993. Devaluation risk in target zones: Evidence from the Nordic countries. The Economic Journal 103, 161–175. Eichengreen, B., Wyplosz, C., 1993. The unstable EMS. Brookings Paper on Economic Activity (1), 51– 143. Eichengreen, B., Rose, A., Wyplosz, C., 1995. Exchange rate mayhem: The antecedents and aftermath of speculative attacks. Economic Policy 21, 251–312. Eichengreen, B., Rose, A., Wyplosz, C., 1996. Contagious currency crises: First tests. Scandinavian Journal of Economics 98, 463–484. Flood, R., Garber, P., 1984. Collapsing exchange rate regimes: Some linear examples. Journal of International Economics 17, 1–13. Flood, R., Marion, N., 1998. Perspectives on the recent currency crisis literature. NBER Working Papers No. 6380. Hibbs, D., 1977. Political parties and macroeconomic policy. American Political Science Review 71, 1467– 1487. Horowitz, J., 1994. Bootstrap-based critical values for the information matrix test. Journal of Econometrics 61, 395–411. Green, W., 1993. Econometric Analysis. Prentice Hall. Kaminsky, G., Reinhart, C., 1998. Financial crises in Asia and Latin America: Then and now. American Economic Review, Papers and Proceedings, vol. 88, pp. 444–448. Kaminsky, G., Lizondo, S., Reinhart, C., 1998. Leading indicators of currency crises. IMF Staff Papers 45, 1–48. Klein, M., Marion, N., 1997. Explaining the duration of exchange rate pegs. Journal of Development Economics 54, 387–404. Krugman, P., 1979. A theory of balance of payments crises. Journal of Money, Credit, and Banking 11, 311–325. Masson, P., 1995. Gaining and losing credibility: The case of the United Kingdom. The Economic Journal 105, 571–582. conomiques et Monetaires 43, 189– Obstfeld, M., 1994. The Logic of Currency Crises. Cahiers E 213. Obstfeld, M., 1996a. Models of currency crises with self-fulfilling features. European Economic Review 40, 1037–1047. Obstfeld, M., 1996b. Comment on ‘‘Are currency crises self-fulfilling’’. In: Krugman, P. (Ed.), NBER Macroeconomics Annual, pp. 393–403. Orme, C., 1987. Specification tests for binary data models, University of Nottingham, Department of Economics, Discussion Paper, No. 65. Ozkan, G., Sutherland, A., 1995. Policy measure to avoid a currency crisis. The Economic Journal 105, 510–519. Ozkan, G., Sutherland, A., 1998. A currency crisis model with an optimising policy maker. Journal of International Economics 44, 339–364.
F.G. Ozkan / Journal of Macroeconomics 25 (2003) 491–507
507
Radelet, S., Sachs, J., 1998. The East Asian financial crisis: Diagnosis, remedies, prospects. Brookings Papers on Economic Activity (1), pp. 1–76. Rose, A., Svensson, L., 1994. European exchange rate credibility before the fall. European Economic Review 38, 1185–1216. Sachs, J., Wyplosz, C., 1986. The economic consequences of President Mitterand. Economic Policy 2, 261– 322. Thomas, A., 1994. Expected devaluation and economic fundamentals. IMF Staff Papers 41, 262–285. Wooldridge, J., 2000. Introductory Econometrics. South Western College Publishing.