Since the Realignment of currencies in the exchange rate mechanism (ERM) of the European Monetary System (EMS) in January 1987, the differential between French and German interest rates has narrowed considerably.1 Nevertheless, the continuing existence of a differential suggests that market participants do not rule out a realignment of central parities in the EMS before exchange rates are irrevocably fixed in the final stage of economic and monetary union (EMU). Specifically, the positive interest rate differential in favor of Germany across the maturity spectrum implies a perceived risk of devaluation of the franc-mark exchange rate. Indeed, attempts to quantify the various factors that might explain the French-German interest rate differential find that they account for only part of the differential.2
The persistence of a positive interest rate differential suggests that an announced commitment to a fixed exchange rate may not be sufficient to eliminate devaluation risk completely. Economic performance and the authorities’ policy approach also influence investors’ expectations. Thus, exchange rate policy may not be fully credible if, for instance, problems of unemployment, a weak external position, or other perceived weaknesses cast doubt on the authorities’ ability to maintain their commitment. In the same vein, a policy of keeping the exchange rate close to the upper (weak) edge of the fluctuation band may also diminish credibility.
This paper examines expectations of a realignment of the franc-mark central parity over the period February 1987 to November 1991. The aim is to estimate the expected rate of realignment and to attempt to identify the factors that might influence market participants’ expectations of parity changes.
A commonly used measure of the expected rate of realignment of a currency is the differential between the interest rates on assets denominated in the domestic currency and those denominated in a foreign currency. This measure is imprecise, however, especially for interest rates at the short end of the maturity spectrum, because interest rate differentials are affected by expected exchange rate changes within the fluctuation band of the exchange rate mechanism. In what follows, estimates of the time-varying expected rate of realignment are first constructed by adjusting interest rate differentials on 3-, 6-, and 12-month Eurofranc and Euromark deposits for the expected rate of change of the franc-mark exchange rate within the fluctuation band. The latter, in turn, is obtained by assuming that the exchange rate inside the band follows a mean-reversion process. The calculated expected parity changes are then regressed on a number of macroeconomic variables that agents are thought to consider in forming expectations of a currency’s possible realignment. A similar analysis is also conducted with long-term government bond yields. Inflation differentials, competitiveness, unemployment rates, relative fiscal situations, and the position of the exchange rate within the band are found to play an important role in the formation of exchange rate expectations.
I. Interest Rate Differentials, Exchange Rate Bands, and Credibility
Under current EMS arrangements, interest rates on financial assets that differ only in their currency of denomination can diverge because of the discounting of two types of exchange rate uncertainty: the day-today fluctuations in exchange rates within the ERM bands and the possibility of a realignment of the centra! rates. It is widely accepted that since the January 1987 realignment the EMS target zones have become much more credible. This has allowed devaluation risk premia to decline and interest rate differentials to narrow considerably. For example, the differential on 12-month Eurofranc and Euromark interest rates (Figure 1) declined from a mean differential of 4.7 percentage points in 1987 to 0.25 percentage points in 1991.
Figure 1.Parity Deviations of Franc-Mark Exchange Rate and 12-Month Interest Rate Differential
a Eurofranc rate minus Euromark rate.
b Percent deviation.
A simple way to test the credibility of the EMS target zones is tc suppose that investors at time t expect with certainty that there will be no realignment (or change in the width of the exchange rate band) up tc time t + m, the time of maturity of a given asset. Then, at time t + n the spot exchange rate is expected with certainty to be bounded by
where S is the domestic currency price of foreign exchange (franc/mark and SL and SU are the lower and upper margins of the exchange rate band Hence, the net profit from a forward safe of one unit of foreign currency is bounded by
where Fmt is the forward exchange rate at time t for maturity m (measured in years). Further assuming that arbitrage eliminates certain positive minimum profits, it follows that if the forward exchange rate lies outside the exchange rate bands, then the target zone is not credible. If the forward exchange rate lies within the exchange rate band, then the test is inconclusive.3 Alternatively, invoking covered interest parity, the domestic interest rate under full credibility is bounded by
where it*m and itm are the domestic currency (franc) and foreign currency (mark) interest rates on assets of the same default risk and maturity m, expressed as annualized rates of return. If at any time t the domestic interest rate is outside these “credibility bounds,” then agents expect with positive probability that over the time to maturity, t + m, the exchange rate band will shift by way of either a realignment or an increase in the band’s width.
Eurofranc interest rates for 3-, 6-, and 12-month maturities and their corresponding “credibility bounds” are shown in Figure 2. The three-month interest rate was almost always within its credibility bounds; hence, the exchange rate band may or may not have been credible. The one-year interest rate, however, was consistently above the upper bound from January 1987 to March 1990, and thereafter consistently inside the bounds. Thus, it may be inferred that until March 1990 agents expected with positive probability a devaluation of the franc-mark exchange rate. In the case of one-year interest rates, the test is less inconclusive as the credibility bounds are significantly narrower than those for three-month interest rates. The next section outlines a more precise empirical method (from Bertola and Svensson (1991)) for extracting the implicit expected rate of realignment from exchange rates and interest rate differentials.4
Figure 2.Eurofranc Interest Rates and “Credibility Bounds”
II. A Model of the Expected Rate of Realignment
Let s, sL, and sU denote the natural logarithms of S, SL, and SU (the franc-mark spot exchange rate and its lower and upper intervention rates, respectively). By definition, the exchange rate can be decomposed as
where ct = (StL + StU) /2 is the log of the central parity of the franc-mark exchange rate and s, is the deviation of the franc from the central parity. Taking first differences of equation (4), the total expected rate of change of the franc from time t to t + m, conditional on information available at time t, is equal to the expected rate of realignment (that is, the change in the central parity) plus the expected rate of change of the franc within the band;
Assuming uncovered interest parity (UIP)
it follows that
UIP is an appropriate assumption if the foreign exchange risk premium is small. Theoretical support is provided by Svensson (1990), who argues that the risk premium is likely to be small in exchange rate target zones, even in the presence of devaluation risk. Empirical support of UIP for the franc-mark rate is provided by Rose and Svensson (1991), Andersen and Sorensen (1991), and, indirectly, Frankel and Phillips (1991).
From equation (1) and the definition of s̄t, it can be shown that the rate of change of the exchange rate within the band is bounded by
which combined with equation (7) gives the following maximal bounds for the expected rate of realignment:
These bounds are shown as the solid lines in Figures 3 and 4. As can be seen, the bounds for the expected parity changes are wider the shorter the maturity. This reflects the fact that the maximal bounds for the expected rate of change of the exchange rate within the band are wider for shorter maturities.
Figure 3.Expected Rate of Change of Franc-Mark Exchange Rate Within the Intervention Band Figure 4.Bounds for Expected Rate of Realignment of Franc-Mark Exchange Rate
III. Estimation of Expected Exchange Rate Changes Within the Band
To calculate the expected rate of realignment from equation (7), it is necessary to filter out expectations of exchange rate changes within the fluctuation band. As noted earlier, the simplest procedure is to assume that the expected change in the exchange rate within the band is zero—that is, Et Δs̄t+m/m = 0—in which case the expected change in the central parity is simply equal to the interest rate differential. This case can be ruled out a priori, however, since the exchange rate within the band cannot follow a random walk. Indeed, an augmented Dickey-Fuller test rejects the hypothesis that s has a unit root.5 An alternative procedure, following Bertola and Svensson (1991), Svensson (1990, 1991), and Rose and Svensson (1991), assumes initially that the future exchange rate within the band may be well approximated by the current exchange rate.6 The change in the exchange rate within the band, conditional upon no realignment, may thus be estimated from
|3 months||6 months||12 months|
|F-statistic (β1 = 0)||21.34||43.93||91.64|
|Number of observations||60||57||51|
Two points about the estimation of equation (10) should be noted. One, a consequence of the target zone model is that the conditional distribution of the exchange rate within the band is heteroskedastic. Two, the projection horizons employed (m = 3, 6, and 12 months) are longer than the sampling interval of the data (monthly). The use of overlapping observations implies that the error terms will follow a moving-average process of order m—1. Ordinary least squares will give consistent estimates of the coefficients, but their standard errors will be inappropriate. Consequently, generalized method of moment (GMM) estimates have been used for the standard errors, which are robust to heteroskedasticity and serial correlation as in Newey and West (1987).
The results indicate that the exchange rate within the band is mean reverting: within the band, the expected exchange rate is closer to the long-run mean of the exchange rate at longer time horizons. The estimated slopes are negative for all maturities, as implied by mean reversion, and are targe relative to their standard errors. Moreover, the longer is the maturity, the larger is the estimated slope in absolute value. The estimates also show that the expected change of the franc-mark exchange rate within the band has narrowed over the January 1987 to March 1992 period and has gradually turned from an expected depreciation to an expected appreciation by the end of the period (Figure 3).
An attempt was made to refine the estimates in Table 1 by the inclusion of additional explanatory variables in equation (9). First, s2 and s3 were included so as to capture possible nonlinearities in the relationship. They were found to be insignificant. Second, the mark-dollar interest rate differential for the corresponding maturity was included on the assumption that its movements may affect the franc-mark exchange rate (Artus and others (1991)). This, too, proved insignificant. The estimates in Table 1 are thus used to calculate the expected rate of realignment of the franc-mark exchange rate. It is shown in Figure 4 for different maturities.
IV. Explaining the Expected Rate of Realignment
This section examines whether the calculated expected rate of realignment can be explained by generally observed macroeconomic variables.7 It is assumed that in forming expectations of a currency’s possible realignment, agents consider a number of factors, at home and abroad, that may cause a change in the central parity. These include inflation differentials, changes in foreign exchange reserves, fiscal developments, unemployment rates, relative money supply growth, and other macro-economic variables. They may also include other factors that influence market sentiment, such as the authorities’ perceived policy behavior or commitment.
The results of regressing the calculated expected rates of realignment on a selected set of macroeconomic variables that are believed to enter agents’ information sets are shown in Table 2.8 For all three maturities the results are quite similar. More than 70 percent of the variation in the expected rate of devaluation is explained by the change in foreign exchange reserves, the government financing requirement (as a ratio to GDP) of France relative to Germany, the inflation differential, France’s export price competitiveness relative to Germany’s, the unemployment rate, and the position of the franc-mark rate within the fluctuation band. Relative money supply growth rates and the trade balance were also included, but they were found to have no additional explanatory power.
|Variableb||3 months||6 months||12 months|
|Change in foreign exchange||-0.118||-0.097||-0.088|
|Export price competitivenesse||-0,245||-0.249||-0.243|
|Deviation of franc-mark||1.502||1.256||0.800|
|rate from upper edge of band||(6.63)||(6.41)||(4.61)|
|Number of observations||58||58||58|
The government financing requirement and inflation variables are found to be positively related, and the competitiveness variable negatively related, to the expected rate of realignment. Thus, decreases in France’s government financing requirement and in its inflation rate relative to Germany’s lower the expected rate of devaluation of the franc, as does an improvement in France’s export price competitiveness relative to Germany’s. Moreover, the null hypothesis that the coefficient on the inflation differential is equal to one cannot be rejected at the 5 percent significance level. The change in foreign exchange reserves enters the equations for the expected rate of realignment with the expected negative sign: an increase in reserves decreases the expected rate of devaluation. Its statistical significance, however, depends on the specification of the equation. In particular, when the unemployment rate is also included its statistical significance declines considerably. The unemployment rate itself, however, is highly significant.9 This raises an interesting point. In a recent paper, Drazen and Masson (1992) extend the notion of policy credibility to encompass not only the role of government policies in signaling the “type” of government (how “tough” it is, to use their terminology) but also the situation in which a government finds itself—since in very adverse circumstances even a policymaker with a reputation for being “tough” may renege on a commitment. Applied to the EMS, this suggests that the increased credibility of the EMS reflects the dominance of the signaling motive for setting policies as governments maintain their commitment not to realign. However, under the Drazen-Masson notion of credibility, expectations of realignment also reflect pressures to increase employment and growth after a period of restrictive policies. The expected rate of devaluation will, therefore, be positively correlated with the rate of unemployment.
The position of the exchange rate within the fluctuation band—expressed as the deviation of the franc-mark rate from the upper (weak) edge of the band—may be interpreted as capturing intangible market sentiment about the credibility of the target zone. Its statistical significance suggests that when the franc trades close to the upper intervention margin, market expectations of a realignment intensify and a devaluation risk premium is built into franc interest rates. As an alternative way of capturing market sentiment, the deviation of the franc-mark rate from the upper band was replaced in the regression equations by a dummy variable that takes increasing values as the exchange rate approaches the upper intervention margin.10 The results were not qualitatively different.11 The statistical significance of these variables suggests that the recent policy of allowing the franc to strengthen in the ERM may pay off by way of reducing the devaluation risk premia on French short-term interest rates. One interpretation of these results is that agents may consider it more difficult to defend a currency that is close to the weak edge of the band by sales of reserves (since reserves are limited) than it is to sterilize capital inflows when a currency is at the strong edge of the intervention margin. In the event of a shock, therefore, a country whose currency is in the weak part of the band may have to raise interest rates in order to maintain its currency within the band. If this is seen as introducing a conflict between the domestic and the exchange rate objectives of monetary policy, the credibility of the central parity may suffer.
The analysis thus far has been conducted using short-term Euromarket deposit rates. Euromarket rates have the advantage over domestic money market rates of not being affected by the existence of capital controls. Since capital controls were being phased out over the sample period, it seemed preferable to use Euromarket rates.12 Euromarket rates are not available for long maturities, however, and the latter have an advantage in that the expected rate of mean reversion of the exchange rate within the band becomes very small at long horizons.13 Since E, Δ
Table 3 reports the results of regressing the differential in the yield on long-term government bonds on the same set of macroeconomic variables used in the equation for the expected rate of realignment derived from Euromarket rates. The results are broadly similar. The inflation rate, competitiveness, and the unemployment rate are again the major factors influencing devaluation expectations.14 The relative government financing requirement, however, is not as significant as in the regression equations based on short-term rates, and the market sentiment variables are insignificant. This last result is not surprising, as the current position of the franc in the intervention band should be of little relevance in forming expectations of exchange rates several years hence: views of the long-term viability of the central rate are shaped by the unfolding of more fundamental economic factors.
|Change in foreign exchange reserves||-0.123||-0.055|
|Government financing requirement||0.562||0.179|
|Export price competitiveness||-0.150||-0.169|
|Deviation of franc-mark||0.086||…|
|rate from upper edge of band||(0.61)|
|Number of observations||58||58|
The differentials between French and German interest rates have narrowed considerably since the EMS realignment in January 1987 as the franc-mark exchange rate band has become increasingly credible. Expectations of realignments of the central parity have been found to be influenced by the evolution of fundamental economic factors such as inflation differentials, competitiveness, unemployment, government financing requirements, and foreign reserves. France’s favorable economic performance, especially with regard to inflation, the external position, and its fiscal situation, has allowed the implicit expected rate of devaluation to decrease considerably. The results further suggest that the devaluation risk premium on franc interest rates could be reduced further and differentials with respect to Germany additionally narrowed by an improved labor market performance and, in the case of short-term rates, by a strengthened position of the franc in the ERM band.
Consumer prices: IMF, International Financial Statistics, line 64.
Exchange rates: IMF, average of daily observations.
Export prices: IMF, International Financial Statistics, line 74.
Foreign exchange reserves: IMF, International Financial Statistics, line 11.d (total reserves minus gold).
Gross domestic products: IMF, International Financial Statistics, line 99b.c for France, line 99a.c for Germany (monthly values obtained by interpolating from industrial production index).
Government financing requirements: IMF, International Financial Statistics, line 84.
Industrial production indices: IMF, International Financial Statistics, line 66c.
Interest rates (Euromarket rates): IMF; 7-10 year government bond yields:
Banque de France (average of daily observations). Unemployment rates: OECD, Analytical Data Base.
Andersen, Torben M., and JanRose Sorensen, “Exchange Rate Risks, Interest Rates and European Monetary Integration,”Memo 1991-18 (Aarhus: Centre for International Economics, University of Aarhus, 1991).
Artus, Patrick, “Peut-on comprendre I’écart de taux d’éinteret entre la France et l’Allemagne?”Document de Travail No. 1992-01/E (Paris: Caisse de Depots et Consignations, 1992).z
Artus, Patrick, and others, “Transmission of U.S. Monetary Policy to Europe and Asymmetry in the European Monetary System,”European Economic Review, Vol. 35 (1991). pp. 1369–84.
Bertola, Giuseppe, and Lars E.O.Svensson, “Stochastic Devaluation Risk and the Empirical Fit of Target-Zone Models,”NBER Working Paper No. 3576 (Cambridge, Mass.: National Bureau of Economic Research, 1991).
Chen, Zhaohui, and AlbertoGiovannini, “Estimating Expected Exchange Rates Under Target Zones,”NBER Working Paper No. 3955 (Cambridge, Mass.: National Bureau of Economic Research, 1992).
Drazen, Allan, and Paul R.Masson, “Credibility of Policies Versus Credibility of Policymakers” (unpublished;Washington: International Monetary Fund, 1992).
Frankel, Jeffrey, and StevenPhillips, “The European Monetary System: Credible at Last?”NBER Working Paper No. 3819 (Cambridge, Mass.: National Bureau of Economic Research, 1991).
Geadah, Sami, TapioSaavalainen, and Lars E.O.Svensson, “The Credibility of Nordic Exchange Rate Bands: 1987-91,”IMF Working Paper 92/3 (Washington: International Monetary Fund, 1992).
Giovannini, Alberto, “European Monetary Reform: Progress and Prospects,”Brookings Papers on Economic Activity 2 (1990), pp. 217–91.
Koen, Vincent, “Testing the Credibility of the Belgian Hard Currency Policy,”IMF Working Paper 91/79 (Washington: International Monetary Fund, 1991).
Lindberg, Hans, Lars E.O.Svensson, and Paul, Soderlind, “Devaluation Expectations: The Swedish Krona 1982-1991,”Seminar Paper No. 495 (Stockholm: Institute for International Economic Studies, 1991).
Newey, Whitney K., and Kenneth D.West, “A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix,”Econometrica, Vol. 55 (1987), pp. 703–8.
Rose, Andrew K., and Lars E.O.Svensson, “Expected and Predicted Realignments: The FF/DM Exchange Rate During the EMS,”International Finance Discussion Paper No. 395 (Washington: Board of Governors of the Federal Reserve System, 1991).
Svensson, Lars E.O., “The Foreign Exchange Risk Premium in a Target Zone with Devaluation Risk,”Discussion Paper No. 494 (London: Centre for Economic Policy Research, 1990).
Francesco Caramazza, a Senior Economist in the Monetary and Exchange Affairs Department, was a Senior Economist in the European I Department when this paper was written. He holds a doctorate from Johns Hopkins University. The author would like to thank Michael Deppler, John Green, and Harilaos Vittas for helpful comments and Susan Becker for research assistance.
Short-term interest rate differentials widened sharply in September 1992 and in the months following the ERM crisis. Recently, short-term differentials have narrowed significantly.
See, for instance, Artus (1992).
This test of target zone credibility was first applied by Svensson (1990) to the Swedish krona and has since been applied to a number of currencies by Giovannini (1990), Koen (1991), and Geadah, Saavalainen, and Svensson (1992).
For an application to the Swedish krona, see Lindberg, Svensson, and Soderlind (1991).
The regression of As, on st-1, three lags of Δ
Supporting evidence is also reported by Chen and Giovannini (1992).
The two-step estimation procedure used in this paper, namely first estimating the expected change in the exchange rate within the intervention band and, subsequently, estimating the determinants of the expected rate of realignment (constructed by adjusting interest rate differentials for the estimates from the first step), is not the only possible estimation strategy. For instance, one could estimate simultaneously the expected change in the exchange rate within the band and the determinants of the expected rate of realignment by estimating
on the rational expectations assumption that Et(st+m—st), conditional on no realignment, is equal to (st+m—st). In this case, the disturbance of the estimated equation would include the expectations error st+m—Et(st+m). The point to note is that, whether in two steps or simultaneously, estimation of the expected rate of realignment requires an estimate of the expected change in the exchange rate within the band. Tests of hypotheses of the determinants of expected changes in the central parity are thus joint tests of the expected changes in the exchange rate within the band. It should also be noted that the estimated standard errors are conditional on the estimated series of realignment risk.
Since numerous variables can be considered, the results reported in this section should be regarded as an examination of some plausible determinants of market expectations.
The unemployment rate is more appropriately measured with respect to its natural rate. But assuming that over the sample period the natural rate is constant, it makes no difference—except for the constant term—whether the actual unemployment rate or its deviation from the natural rate is used. In the regression using the 12-month Euromarket rates, for example, the estimated constant term is -0.95 when the natural rate is assumed to be 8.5 percent and -0.25 (and not significantly different from zero) when the natural rate is assumed to be 9 percent. In view of the uncertainty about the perceived value of the natural rate, and since it makes no difference for the other coefficients, the constant term is not reported.
Specifically, the dummy variable takes values of one when the franc-mark rate is in the top half of the intervention band, two when it is in the top quarter, and three when it is in the top eighth.
The estimated coefficients on the dummy variable (with r-statistics in parentheses) in equations using the 3-, 6-, and 12-month interest rates, respectively, are 1.205 (6.89), 1.00 (6.41), and 0.66 (5.33).
Regulations other than capital controls, political and default risk, information and transaction costs, and other factors may also introduce a wedge between domestic market and Euromarket rates. Changes in domestic market rates may, therefore, be due to actual or anticipated changes in these characteristics rather than in exchange risk. Empirically, however, capital controls have been found to constitute the major explainable component of spreads between Euromarket and domestic market rates.
The results in the preceding section indicate that the expected rate of mean reversion decreases as the horizon lengthens.