Empirica (2011) 38:555578 DOI 10.

1007/s10663-010-9146-z ORIGINAL PAPER

Behavioural equilibrium exchange rate and total misalignment: evidence from the euro exchange rate
Nikolaos Giannellis Minoas Koukouritakis

Published online: 3 November 2010 Springer Science+Business Media, LLC. 2010

Abstract In the present paper we attempt to investigate whether the nominal exchange rate of the euro against the currencies of the four major trading partners of the eurozone, namely China, Japan, the UK and the USA, converges or not to its equilibrium level. Applying recent unit root and system cointegration techniques in the presence of structural shifts in the data, our results indicate that there exist an equilibrium relationship between each of the euro/yuan, euro/yen, euro/UK pound and euro/US dollar nominal exchange rates and the fundamentals dened by the monetary model. Following these results, our modied Behavioural Equilibrium Exchange Rate model suggests that at the end of the estimated period, the euro/ Chinese yuan and the euro/UK pound nominal exchange rates follow an equilibrium process. Our empirical results also imply that the single European currency is considered as overvalued against the US dollar, while it is considered as undervalued against the Japanese currency. Keywords Euro BEER Misalignment rate Cointegration Structural shifts E43 F15 F42

JEL classication

1 Introduction The present paper aims to evaluate the dynamic behaviour of the single European currency, by examining the likelihood of emergence of signicant future
N. Giannellis Department of Economics, University of Ioannina, Ioannina, Greece e-mail: ngianel@cc.uoi.gr M. Koukouritakis (&) Department of Economics, University of Crete, Crete, Greece e-mail: minoas@econ.soc.uoc.gr

123

556

Empirica (2011) 38:555578

uctuations of the euro exchange rate against four major currencies, namely the Chinese yuan, the Japanese yen, the UK pound and the US dollar. Besides unanticipated disturbances, future exchange rate instability is not expected to be high if the above nominal exchange rates are not signicantly away from their equilibrium rates. The intuition is that even if the exchange rate is currently stable but, signicantly misaligned, exchange rate stability may be weakened in the future. In other words, a highly misaligned exchange rate is expected to be volatile in the future in its attempt to move close to its equilibrium rate. The above dependence is because exchange rate stability and equilibrium are closely related notions. Exchange rate volatility corresponds to short-run uctuations of the exchange rate around its long-run trend, while exchange rate misalignment refers to a signicant deviation of the observed exchange rate from its equilibrium rate. Given that the exchange rate links the domestic economy with the rest of the world, a signicantly misaligned exchange rate can have important negative consequences in the Euro area. For instance, an undervalued euro (i.e. below its equilibrium value) may lead to inationary pressures, while an overvalued euro (i.e. above its equilibrium value) implies possible competitiveness problems. These situations suggest the necessity of estimating the equilibrium exchange rate of the euro against the currencies of Euro areas major trading partners. Departing from traditional theories of equilibrium exchange rates, such as the Purchasing Power Parity (PPP), recent empirical applications are based on the estimation of alternative equilibrium exchange rate models, such as the Fundamental Equilibrium Exchange Rate (FEER), the Natural Real Exchange Rate (NATREX) and the Behavioural Equilibrium Exchange Rate (BEER).1 In the present study, the theoretical model that we have chosen is a combination of the BEER model (Clark and MacDonald 1998) with the monetary model of exchange rate determination (Frenkel 1976). The properties of the BEER model that we develop are the decomposition of the exchange rate misalignment, and most importantly, the distinction between current and total equilibrium exchange rate. Namely, the key fact of the BEER model utilised in our theoretical model is that the exchange rate may be misaligned if the macroeconomic fundamentals deviate from their sustainable values. Thus, our BEER model differs from standard BEER models in the set of macroeconomic fundamentals.2 This paper contributes to the literature of equilibrium exchange rate determination by strengthening the theoretical background of the BEER model. Our novelty lies on the fact that we combine the theoretical assumptions of the monetary model with specic characteristics of the BEER methodology. This is an important issue, since the BEER approach, whose building idea is the uncovered interest rate parity (UIP) condition, does not actually rely on any theoretical model. Hence, the combination of the advantages of both models (monetary model and BEER) allows us to get results from a competitive model.
1 2

See Egert (2004) for an analytical presentation of these models.

The use of non-monetary variables, such as the terms of trade, current account, net foreign asset position, debt ratio and others is more appropriate for the estimation of the real effective exchange rates. Since we investigate nominal exchange rates, the fundamentals of the monetary model are suitable for estimating a bilateral nominal exchange rate.

123

Empirica (2011) 38:555578

557

Furthermore, we account for structural breaks in the data. In our context, this is also an important issue because the monetary and scal policies of the Euro area, China, Japan, the UK and the USA are likely to have caused structural shifts in the level and trend of the bilateral exchange rates under consideration. Since the presence of structural breaks in the data are known to have signicant effects on the properties and interpretation of standard unit root and cointegration tests, we employ recently developed tests that are valid in the presence of structural shifts in the data; we discuss these issues extensively in Sect. 3 below. In what follows, rstly, we use monthly data available from the 1999:01 and unit root tests in the presence of structural breaks in the data (Lee and Strazicich 2003, 2004), in order to test for unit roots and to determine endogenously the possible structural breaks. Secondly, we use recently developed cointegration tests that allow tkepohl and his associates in for structural breaks in the data (Johansen et al. 2000; Lu several papers noted below) in order to establish an equilibrium relationship between ` -vis the Chinese yuan, the Japanese yen, the UK the nominal euro exchange rate vis-a pound and the US dollar and the fundamental variables that dened by the monetary model of exchange rate determination. Thirdly, from the above relationship we estimate the total equilibrium exchange rate (i.e. the BEER) in order to investigate if the nominal euro exchange rate against the currencies of China, Japan, the UK and the USA is overvalued or undervalued in relation to its equilibrium level. In brief, our empirical results indicate that, at the end of our sample, the euro is overvalued in relation to the US dollar, undervalued in relation to the Japanese yen and moves towards its equilibrium value in relation to the Chinese yuan and the UK pound. The rest of the paper is organised as follows. Section 2 discusses the theoretical framework of the BEER model and the way that this methodology can be combined with the theoretical assumptions of the monetary model. Section 3 describes the data and outlines the unit root and cointegration tests in the presence of structural breaks. Section 4 discusses the results for the current and total equilibrium exchange rates, while Sect. 5 contains some concluding remarks.

2 Theoretical framework The estimation of the equilibrium exchange rate is based on the BEER approach of Clark and MacDonald (1998). This approach estimates the exchange rate misalignment in accordance with the deviations of the actual exchange rate from its estimated value, which is derived from the long-run relationship between the exchange rate and the macroeconomic fundamentals. The advantage of the BEER model is that the exchange rate is a function of variables that have a direct effect on the exchange rate. In other words, the equilibrium exchange rate is driven by the sustainable (equilibrium) values of the fundamentals that affect the actual exchange rate in the long run and not by overall macroeconomic balance. The BEER approach does not actually rely on any theoretical model and the equilibrium rate is designated by the long-run behaviour of the macroeconomic variables. However, a theoretical concept is required. Stein (2001) presents an evaluation of studies based on the BEER approach, in which the authors have in

123

558

Empirica (2011) 38:555578

mind a theoretical model based on the condition of simultaneous internal and external balance, but they do not clearly specify it. This implies that the building idea of the BEER approach is the UIP condition. In this study, we strengthen the theoretical background of the BEER model by assuming that the fundamentals that affect the exchange rate are those dened by the monetary model of exchange rate determination. Our approach extends the standard BEER approach in the sense that we do rely on a theoretical model. The derivation of the current equilibrium exchange rate based on the monetary model, is presented below.3 The main expression of the monetary model is as follows4: st mt m 1 t / yt yt l rt rt : Applying the UIP condition, (1) becomes: st mt m t / yt yt lEt Dst1 :

2 Since the PPP holds all the time, it turns out that Et Dst1 Et pt1 Et p t 1 , where p and p* are the domestic and foreign ination rates, respectively. Then the exchange rate equation becomes: 3 st mt m t / yt yt l Et pt1 Et pt1 : Finally, assuming that ination expectations are rationally formed (i.e. the market agents have perfect foresight), we derive the current equilibrium exchange rate: 4 st mt m t / y t y t l p t 1 pt 1 : Following Clark and MacDonald (1998), we set as Z1 a vector of macroeconomic fundamentals that affect the exchange rate in the long run, as Z2 a vector of macroeconomic fundamentals that affect the exchange rate in the medium run and as T a vector of variables that affect the exchange rate in the short run. Then, the nominal exchange rate is dened as follows: st b1 Z1t b2 Z2t sTt ut 5

where, b1, b2 and s are reduced form coefcients and ut is the error term. The current values of the medium-run and long-run fundamentals give the current equilibrium exchange rate: "t b1 Z1t b2 Z2t ; s Subtracting (6) from (5), we get the current misalignment: "t sTt ut : st s 7 6

Equation (6) is equivalent to (4) if Z1 and Z2 include the variables of the monetary model. However, it should be noted that the fundamentals of the monetary model have a long-run effect on the exchange rate. Thus, all variables are included to
3

For more details about the monetary model, see MacDonald and Taylor (1991, 1992), MacDonald (1995) and Groen (2000).

S is the nominal exchange rate and m, y, r represent the domestic real money supply, real income and interest rate, respectively, while the asterisk denotes the respective foreign variables.

123

Empirica (2011) 38:555578

559

vector Z1. Equation (7) corresponds to the series that results by subtracting (4) from the actual exchange rate. In our analysis we need to estimate the total misalignment that is the deviation of the actual exchange rate from the total equilibrium exchange rate (i.e. the BEER). To estimate the BEER, we replace Z1 and Z2 in (5) with the long-run (or ~2 , respectively.5 These values are ~1 and Z equilibrium) values of the fundamentals, Z taken by ltering the fundamentals from speculative and cyclical factors. Maintaining the theoretical affairs of the monetary model, the BEER is: ~t y ~ t 1 p ~ ~ ~ ~t m 8 BEER m t u y t l p t 1 : Comparing the BEER with the actual exchange rate, st, we nd how the latter deviates from the former. If st, is greater (less) than the BEER, the exchange rate is said to be overvalued (undervalued). Thus, the total misalignment rate is: ~1t b2 Z ~2t : nt s t b 1 Z 9

Finally, by adding and subtracting the current equilibrium exchange rate, " s, from the right-hand side of (9) and using (7), we can decompose the source of exchange rate misalignment, n: ~1t b2 Z2t Z ~2t : nt sTt ut b1 Z1t Z 10

Equation (10) illustrates the sources of exchange rate deviation from its equilibrium value. These are: (1) the transitory factors that have a short-run effect on the exchange rate, (2) the disturbance term and, most importantly, (3) the deviations of the macroeconomic fundamentals from their long-run (or equilibrium) values. Since one of the novelties of the present paper is that we take into account the presence of structural breaks in the data, we describe in Sect. 3 below the unit root and system cointegration tests in the presence of structural breaks that will be used in this study for estimating the BEER for the euro exchange rates.

3 Unit roots and cointegration with structural breaks Table 1 presents the data used in the present paper along with their sources. Our sample is consisted of monthly observations from 1999:01 to 2008:08. During this period several events have taken place in the economies of the sample countries that probably caused structural breaks in their time series. Since the presence of structural breaks have signicant effects on the properties and interpretation of standard ADF-type unit root tests and Johansen-type cointegration tests, we employ recently developed tests that take structural shifts into account. Note here that the time span of the data set is restricted by the time of launch of the euro.6 We did not
5 6

Although vector Z2 is considered as empty, it is presented for exponential purposes.

Although the sample period is relatively small, it is suitable for our analysis. Our study does not imply the detection of a steady-state rate around which the exchange rate should move for the next decades. Actually, our equilibrium analysis aims to nd, based on the information from the last 10 years, whether the euro exchange rate is consistent with the macroeconomic fundamentals and to discuss the

123

560 Table 1 Description of data Country Euro area China Japan UK USA WE AE AG AE Euro exchange rate Consumer price index 64H 64 64 64H 64

Empirica (2011) 38:555578

Money supply M2 (59MBU) M2 (59MB) M2 (59MBA) M4a M2 (59MB)

Industrial production 66 66 66 66 66

All data were obtained from the International Financial Statistics (IFS) of the International Monetary Fund. The entry in each cell refers to the line of the IFS database
a

M4 for the UK was obtained from the Bank of England. For China and Japan, the IFS data series report the nominal exchange rates against the US dollar. To obtain the nominal euro exchange rates, we used cross exchange rates. The time span is 1999:012008:08

use ECU rates before 1999 because the two rates are not equivalent for two reasons: Firstly, euro is the common currency of the 16 EMU members, while ECU was a basket of currencies of the EU members. Thus, the currencies that the euro replaced are not identical to the currencies that constituted the ECU.7 Secondly, the ECU was an internal accounting unit, while the euro is a real currency represented by ofcial banknotes and coins and whose exchange rate is determined on foreign exchange markets and is inuenced by monetary and scal policies in the Euro area. Note also that for industrial production and money supply we used raw instead of seasonally adjusted data. The reason is that the use of seasonally adjusted data can lead to mistaken inference on economic relationships, and thus to a signicant loss of valuable information on seasonal behaviour in economic time series (see Ghysels and Perron 1993; Lee and Siklos 1997; Franses and McAleer 1998). 3.1 Unit root tests with structural breaks We test for unit roots in the data using the two-break and one-break LM (Lagrange Multiplier) tests that developed by Lee and Strazicich (2003, 2004). These tests determine the structural breaks endogenously from the data. Also, by including breaks under both the null and alternative hypotheses, a rejection of the null hypothesis of a unit root implies unambiguously trend stationarity. Consider the two-break LM unit root test for the process yt generated by y t d0 Z t e t ; et bet1 ALet ; et $ iid 0; r2 11

where A(L) is a k-order polynomial and Zt is a vector of exogenous variables, whose components are determined by the type of breaks in yt. Lee and Strazicich (2003) include two breaks in the level (Model A) or two breaks in both the level and trend (Model C) of yt. Equation (11) shows that yt has a unit root if b = 1, while it is trend
Footnote 6 continued macroeconomic aftermath of a possible misalignment. We believe that 10 years of data are enough to conduct this kind of equilibrium analysis.
7

For instance, Denmark and the UK that were participating in the ECU did not adopt the euro. In contrast, Austria and Finland that adopted the euro in 1999 were not participating in the ECU.

123

Empirica (2011) 38:555578

561

stationary if b \ 1. According to the LM principle, a unit root test statistic can be obtained from the test regression ~t1 Dyt d0 DZt /S
k X i1

~ti ut ; hi D S

12

~ Zt ~ ~t yt w d; t 2; . . .; T , in which ~ d is a vector of coefcients in the where S x ~ y 1 Z1 ~ d , where y1 and Z1 are the rst obserregression of Dyt on DZt and w x vations of yt and Zt, respectively, and ut is a white noise error term. The lagged ~ti correct for serial correlation in ut. The unit root null hypothesis is differences of S described by / = 0 in (12) and is tested by the LM test statistic: ~ s t -statistic for the hypothesis / 0 13

To endogenously determine the location of the two breaks (kj = TBj/T, j = 1, 2) the two-break minimum LM test statistic is determined by a grid search over k: LMs inf f~ skg
k

14

The critical values for this test are invariant to the break locations (kj) for Model A but depend on the break locations for Model C. In this study, when the two-break LM test results showed that only one break is signicant for some variables, we computed the one-break LM test. We did this not only because the one-break LM test appears more appropriate, but also because we wanted to determine if including two breaks instead of one can adversely affect the power to reject the unit root hypothesis for these variables. 3.2 Cointegration tests with structural breaks Structural breaks in the data can distort substantially standard inference procedures for cointegration. Thus, it is necessary to account for possible breaks in the data before cointegration inference. In the recent literature in a VAR framework, there are two main approaches: Johansen et al. (2000) (JMN) extend the standard VECM with additional dummy variables in order to account for q exogenous breaks in the deterministic components of a vector-valued stochastic process. JMN obtain critical or p-values, for the multivariate counterparts of models A and C, using the response surface method. Consider the simple case with only level shifts in the constant term l of an observed p-dimensional time series Yt, t = 1, , T, of possibly I(1) variables. JMN divide the sample observations into q sub-samples, according to the location of the break points, each of length Tj - Tj-1 for j = 1, , q and 0 = T0 \ T1 \ \ Tq = T. They assume the following VECM(k) for Yt, conditional on the rst k observations of each sub-sample YTj1 1 ; . . .; YTj1 k : DYt PYt1 lDt
k 1 X i 1

Ci DYti

q k X X i1 j2

gji Dj;ti et ;
0

et $ iidN 0; X; 15

where l = (l1, , lq) and Dt = (D1,t, , Dq,t) are of dimension (p 9 q) and (q 9 1), respectively, and the Dj, ts are dummy variables, such that Dj,t = 1 for

123

562

Empirica (2011) 38:555578

Tj-1 ? k ? 1 B t B Tj and Dj,t = 0 otherwise, for j = 1, , q. The cointegration hypothesis can then be tested by the likelihood ratio statistic LRJMN T
p X ir0 1

ln1 ^ ki

16

^0 s can be obtained by solving the related generalised where the eigenvalues k j eigenvalue problem, based on estimation of the VECM(k) in (15). tkepohl and his associates (see Lu tkepohl The second approach developed by Lu tkepohl 2000; Trenkler et al. 2008) (LST). and Saikkonen 2000; Saikkonen and Lu LST assume that the structural breaks have occurred only in the deterministic part and do not affect the stochastic part of the process Yt. Thus, LST set up the data generation process (DGP) for Yt by adding its deterministic part lt to its stochastic part Xt, where the latter is an unobservable zero-mean stochastic VAR process, and use dummy variables to account for exogenous shifts in lt. LST propose a two-step procedure to test for cointegration and they use LM-type and LR-type test statistics. For LR-type tests, consider the case of a single shift in the level of Yt. Assuming an exogenous break at time TB in the level of lt, LST specify the following DGP for Yt Yt lt Xt l0 l1 t ddt Xt ; t 1 ; . . .; T ; 17a

where t is a linear time trend, li(i = 0, 1) and d are unknown (p 9 1) parameter vectors, dt is a dummy variable and where the unobserved stochastic error Xt is assumed to follow a VAR(k) process with VECM representation DXt PXt1
k 1 X i1

Ci DXti et ;

et $ iidN 0; X;

t 1; . . .; T :

17b

It is also assumed that the components of Xt are at most I(1) variables and cointegrated (i.e. P = ab0 ) with cointegrating rank r0, where 0 \ r0 B p. Given the DGP in (17a) and (17b), LST initially obtain estimates of the parameter vectors l0, l1 and d using a feasible GLS procedure under H0(r0): rank(P) = r0 vs. H1(r0): ^0 , l ^1 and ^ rank(P) [ r0. Having the estimated parameters l d, one can then compute ^ ^ ^0 l ^1 t ddt . In the second step an LR-type test the de-trended series Xt Yt l for the H0 of cointegration is applied to the de-trended series, by replacing Xt with ^t in the VECM (17b) and computing the LR or trace statistic: X LRLST T
p X i r 0 1

ln1 ~ ki ; ;

18

where the eigenvalues ~ k0i s are obtained by solving a generalised eigenvalue problem, along the lines of Johansen (1988). Critical or p-values for a single level shift can be computed by the response surface techniques (Trenkler 2008). Additionally, Trenkler et al. (2008) derive asymptotic results and p-values for the case of one level shift and one trend break in the Yt process, and show that the asymptotic distribution of the LR statistic in (18) depends on the break location. They also discuss the general case of q [ 1 break points.

123

Empirica (2011) 38:555578

563

Since the JMN and LST approaches have different nite sample properties, we tkepohl et al. employ both the LRJMN and LRLST test statistics in our analysis. Lu (2003) studied the statistical properties of their tests in the case of level shifts, and compared them to the JMN test. They found that their tests have better size and power properties than the JMN test in nite samples. For that reason, if the results of the JMN and LST tests are different, we will use those of the latter test.

4 Empirical results 4.1 Unit Root Results with Structural Breaks Tables 2 and 3 report the unit root results from the two- and one-break LM tests, respectively. We tested each time series for a unit root using the two-break LM test at the 1-, 5- and 10% levels of signicance. When this test showed that only one structural break is signicant we employed the one-break LM test. The number of lags, k, in (12), was determined using a general to specic procedure at each combination of break points (k1, k2) for the two-break test, and at each single break point k for the one-break test. Initially, we set the lag-length at k = 12, and examined the signicance of the last lagged term, at the 10% level. The procedure was repeated until the last lagged term was found to be signicantly different than zero, at which point the procedure stops.8 As shown in the last column of Table 2, the unit root hypothesis with two structural breaks cannot be rejected for all nominal exchange rates, for the Euro area/USA output differential and for the Euro area/China and Euro area/UK money supply differentials. It is also shown in this column that the unit root hypothesis with two breaks is strongly rejected for the Euro area/UK output differential and for the ination rate differential in the cases of Euro area/China, Euro area/UK and Euro area/USA. Table 3 reports the results for the cases that one break is signicant. As shown in the last column of this table, the unit root hypothesis with one break cannot be rejected for the Euro area/China and Euro area/Japan output differentials, for the Euro area/Japan and Euro area/USA money supply differentials and for the Euro area/Japan ination rate differential. Since the results in Table 3 are consistent with the results of Table 2 regarding the null hypothesis, there does not seem to be any detectable loss of power in using the two-break LM test to test the unit root hypothesis for the cases of Table 3. Finally, as shown in column three of Tables 2 and 3, Model A ts the data best for Euro area/China and Euro area/Japan cases, while Model C ts the data best for the Euro area/UK and Euro area/USA cases, over the sample period. Column ve of Tables 2 and 3 reports the endogenously estimated structural breaks in each series. Not surprisingly, these breaks correspond closely to specic events that have taken place in the countries that we examine. In the Euro area/China case, our results indicate that the money supply and ination differentials appear a break in level in 2001, which coincides with a period
8

The Gauss codes of J. Lee are available at the website http://www.cba.ua.edu/*jlee/gauss.

123

564 Table 2 Two-break LM unit root test results Variables Euro area/China s (yy*) (mm*) (pp*) Euro area/Japan s (yy*) (mm*) (pp*) Euro area/UK s (yy*) (mm*) (pp*) Euro area/USA s (yy*) (mm*) (pp*) Model A Critical values 1% -4.54 5% -3.84 10% -3.50 Model A A A A A A A A C C C C C C C C ^ k 11 12 12 11 7 12 12 11 11 11 12 11 11 12 12 12 Model C Break points k ^ k1 ; ^ k2 k = (0.2, 0.4) k = (0.2, 0.6) k = (0.2, 0.8) k = (0.4, 0.6) ^B T

Empirica (2011) 38:555578

^ k2 k1 ; ^ NA NA NA NA NA NA NA NA 0.2, 0.8 0.2, 0.6 0.2, 0.8 0.2, 0.4 0.2, 0.8 0.2, 0.8 0.4, 0.6 0.2, 0.8

LM -2.36 -2.72 -2.61 -3.95* -2.80 -3.15 -2.57 -3.38 -4.69 -6.54** -5.06 -7.05** -4.43 -4.18 -4.89 -7.73**

2004:10, 2006:04 2001:11n, 2003:01 2001:05, 2007:01 2001:11, 2004:11 2000:07, 2001:04 2000:09, 2003:01n 2001:01, 2002:02n 2000:04n, 2000:07 2002:12, 2006:08 2000:05, 2003:10 2002:12, 2006:04 2000:06, 2001:07 2002:03, 2005:03 2000:09, 2005:05 2000:12n, 2003:11 2000:10, 2005:11

Critical values 1% -6.16 -6.41 -6.33 -6.45 5% -5.59 -5.74 -5.71 -5.67 10% -5.27 -5.32 -5.33 -5.31

s is the nominal euro exchange rate; y is the industrial production, m is the money supply and p is the ^ is the estimated number of lags in the test ination rate. s, y and m are expressed in natural logarithms. k ^B denotes the estimated break points. NA means not regression (12) to correct for serial correlation. T affected by the break points. The critical values for Models A and C are from Table 2 of Lee and Strazicich (2003)
n

Signies that the relevant break is not signicant at the 0.10 level of signicance

** (*) denotes rejection of the unit root hypothesis at the 0.01 (0.05) level of signicance

where the European Central Bank (ECB) proceeded to several reductions of its marginal lending facility rate (MLFR). At the same time, China boosted bank lending and loosened its scal policy, which lead to an increase of the credit and money supply growth. In the end of 2002 and for avoiding overheat of the economy, the central bank of China tightened its monetary policy in order to reduce aggregate demand. This action is may reected in the level shift of the output differential, in early 2003. In 2004 and with the ECBs MLFR unchanged at 3%, China tightened again its monetary policy in order to ght ination. This policy action coincides with the break in the nominal exchange rate and ination rate differential in that year. The nominal exchange rate appears a second break in 2006,

123

Empirica (2011) 38:555578 Table 3 One-break LM unit root test results Variables Euro area/China Euro area/Japan (yy*) (yy*) (mm*) (pp*) Euro area/USA Model A Critical values 1% -4.24 5% -3.57 10% -3.21 (mm*) Model A A A A C Model C Break points ^ k k = 0.5 Critical values 1% -5.11 5% -4.51 ^ k 12 12 12 12 12 ^B T 2003:01 2000:09 2001:01 2000:07 2003:07 ^ k NA NA NA NA 0.5

565

LM -2.29 -3.11 -2.52 -3.17 -3.30

10% -4.17

See notes in Table 2. The critical values for Models A and C are from Table 1 of Lee and Strazicich (2004)

while the money supply differential has a second break in early 2007. These breaks coincide with a period of several increases of the ECBs MLFR. Between March 2006 and March 2007, this rate increased from 3.5 to 4.75%. In the Euro area/Japan case, the nominal exchange rate appears to have two level shifts, while each of the output, money supply and ination differentials has a single break. All breaks occurred between late 2000 and early 2001 and coincide with the decisions of the Bank of Japan to (a) terminate the zero interest rate policy in August 2000, which was implemented in February 1999, and (b) introduce quantitative easing in March 2001, along with a zero interest rate again and a large expansion of monetary base. Unfortunately, these actions did not help Japan to ght deation successfully and to escape the liquidity trap. In the Euro area/UK case, all variables have two signicant breaks in both the level and the trend. Our results indicate that the output and ination differentials have a break in mid-2000. The break in output differential signies the beginning of a period of gradual slowdown of the UK economy, while the break in the ination differential is probably related with the signicant fall of the UK ination. The ination differential has a second break in mid-2001 which is probably related with an increase in the UK ination. During the 20002001 period, no remarkable changes were observed in the output growth and the ination rate of the Euro area. The nominal exchange rate and the money supply differential appear to have a structural break at the end of 2002 that is related with the beginning of a period of several decreases of the ECBs MLFR, which reduced from 4.25% in November 2002 to 3% in June 2003. During that period, the UK rates remained unchanged. The output differential has a second break at the end of 2003, which coincides with the beginning of a period of increasing growth of the Euro areas industrial production. Also, the nominal exchange rate and the money supply differential have a second break in 2006. In that year, the ECB gradually raised its MLFR from

123

566

Empirica (2011) 38:555578

3.25% in January to 4.5% in December, while the Bank of England slightly increased its base rate from 4.5 to 4.75%. Finally, in the Euro area/USA case, the output and ination differentials appear to have a structural break at the end of 2000, which signies a period of gradual slowdown of the US output growth. As a consequence, the unemployment rate increased and the ination rate was edging lower. As a result, the Federal Reserve (FED) cut the federal funds rate, bringing the cumulative reduction in that rate to 3% by August 2001. The nominal exchange rate has a break in early 2002, which is related with the beginning of a period followed the terrorist attacks in September 11, 2001. This period was characterized by a signicant cut of the FEDs federal fund rate, together with a remarkable increase of government spending by the Bush administration in order to nance the war against Afghanistan and Iraq. The single break in money supply differential in mid-2003 coincides (a) with a signicant increase in the US money supply, and (b) with a reduction of the ECBs MLFR from 3.75% in January 2003 to 3% in June 2003. The second break in the nominal exchange rate and the output differential appears in early 2005, while the second break in the ination differential appears in late 2005. This break initiates a period of (a) continuous depreciation of the US dollar against the euro due to the increasing decit in the US balance of payments, (b) increasing growth rate of US industrial production, and (c) increasing US ination rate mainly due to the increase of oil price. Also, since 2005, both the ECB and the FED were gradually raising the MLFR and the federal funds rate, respectively. 4.2 Cointegration results with structural breaks In this section we examine the cointegration results with structural breaks based on the JMN and LST procedures described in Sect. 3.2. In each case, the vector Yt contains the nominal exchange rate and the output, money supply and ination differentials. Since we are interested in determining the euro exchange rate against the four major currencies, we used the estimated structural breaks of the nominal exchange rate reported in Table 2. In the case of the JMN procedure we estimated the VECM in (15) for each case and computed the LRJMN test statistics and the corresponding response surface p-values.9 In the case of the LST procedure, we estimated the model in (17a) and (17b) by adjusting (17a) to account for the structural breaks specic to each case. Thus, for each of the Euro area/China and Euro area/Japan cases, which have two signicant breaks in the level, we added a second step dummy to (17a). For each of the Euro area/UK and Euro area/USA cases, which have two signicant breaks in both level and trend, we extended (17a) by adding a second step dummy and two linear trend dummies. Then, for each country we computed the LRLST test statistic and the corresponding response surface p-value using GAUSS routines. Table 4 reports the LRJMN and LRLST test statistics and p-values, for each case. The lag length, k, for each VECM, was selected using the Akaike information criterion (AIC). As shown in the table, the JMN test results indicate two
9

For these estimations we used the JMulti software, available at the website http://www.jmulti.de.

123

Empirica (2011) 38:555578 Table 4 The JMN and LST cointegration tests with structural breaks Country Euro area/China (p - r0) 4 3 2 1 Euro area/Japan 4 3 2 1 Euro area/UK 4 3 2 1 Euro area/USA 4 3 2 1 LRJMN (r0) 95.41** 56.76* 29.82 11.40 164.73** 102.79** 46.65** 14.10 110.32** 59.58** 32.88** 10.12 120.36** 66.28 37.04 13.75 LRLST (r0) 48.62** 17.62 4.50 0.04 54.55** 20.61 11.31 0.32 64.77** 24.83 8.58 0.35 53.78* 31.55 15.08 2.38 p-values JMN 0.006 0.079 0.248 0.408 0.000 0.000 0.002 0.172 0.000 0.001 0.022 0.327 0.003 0.203 0.384 0.626 p-values LST 0.022 0.580 0.910 0.998 0.004 0.354 0.256 0.960 0.004 0.511 0.881 0.999 0.060 0.176 0.382 0.910

567

^ k 3

^ denotes the estimated lag length in the VECM k ** and * denote rejection of the null hypothesis at the 0.05 and the 0.10 level of signicance, respectively

cointegrating vectors for the Euro area/China case, three vectors for the Euro area/ Japan and Euro area/UK cases and a single vector for the Euro area/USA case. The LST test results indicate a single cointegrating vector in each case. As noted in Sect. 3.2, the LST is preferable than the JMN test in nite samples. Thus, we can conclude that there is a single cointegrating vector in each case. Having established a valid relationship, which can be interpreted as the long-run relationship, between the nominal exchange rate and the fundamentals, we estimate the corresponding VECMs, based on (17a) and (17b). Table 5 presents the estimated coefcients of the reduced form equations, normalised on the nominal exchange rate, along with the results from the long-run exclusion test. The latter investigates if any of the macroeconomic fundamentals can be excluded from the cointegrating space (i.e. do not affect the exchange rate in the long run). Using the LR test statistic, our results imply that the real income differential can be excluded from the cointegrating equation for the Euro area/China and Euro area/USA cases. Similarly, money supply and ination rate differentials should be excluded from the Euro area/ Japan and Euro area/UK cointegrating equations, respectively. Thus, these variables cannot explain the long-run behaviour of the respective exchange rates. Regarding the implied structural breaks, the long-run exclusion test shows that both breaks are statistically signicant for the Euro area/China case. In contrast, both structural changes should be excluded for the Euro area/USA cointegrating equation. For the Euro area/Japan case the rst break is statistically insignicant, while for the Euro area/UK case, the second break seems to not affect the nominal exchange rate in the long run. In order to check the robustness of our results, we also estimated the

123

568 Table 5 Estimated coefcients of the cointegrating vectors (raw data) Parameter estimates b(yy*) b(mm*) b(pp*) Intercept Trend SB1 SB2 Euro area/China 0.049 (0.702) 2.032** (0.000) 0.066** (0.000) 1.023 (0.260) 0.018** (0.000) -0.313** (0.001) -0.327** (0.009) Euro area/Japan -4.163** (0.007) 2.328 (0.136) 2.015** (0.000) 0.129 (0.303) -0.019** (0.021) 0.136 (0.295) 0.443** (0.044)

Empirica (2011) 38:555578

Euro area/UK 3.618** (0.000) -0.116** (0.000) -0.065 (0.601) -0.156 (0.969) -0.007** (0.000) -0.002* (0.096) -0.002 (0.111)

Euro area/USA -0.157 (0.865) -1.050** (0.027) 0.559** (0.000) 0.080 (0.399) -0.003* (0.087) -0.003 (0.127) -0.002 (0.318)

The above cointegrating vectors are presented having the explanatory variables on the right-hand side. b0 s are the parameters of the cointegrating vectors, normalised on the nominal exchange rate. SB1 and SB2 are the rst and the second structural break, respectively. Numbers in parentheses are the p-values of the likelihood ratio test statistics for the long-run exclusion tests. For the Euro area/UK and Euro area/USA cases the structural breaks are trend breaks, while for the Euro area/China and Euro area/Japan cases the structural breaks are breaks in the constant term ** (*) denotes rejection of the null hypothesis at the 0.05 (0.10) level of signicance

VECMs based on (17a) and (17b) using seasonally adjusted data for the industrial production and the money supply, for each of the sample countries.10 These results are presented in Table 6 and are quite similar with those of Table 5, regarding either the sign or the statistical signicance of the explanatory variables. Thus, there is evidence of robustness of our results using raw data, which allows us to use them in the forthcoming analysis. Finally, we performed weak exogeneity tests, in order to investigate whether a variable can be considered as weakly exogenous to the long-run parameters. (i.e. if the corresponding adjustment coefcient is statistically zero). This test provides us information about the fundamentals that drive each system to its equilibrium. The results are presented in Table 7 and are based on the estimated VECMs using raw data. Starting from the Euro area/China case, our results indicate that the nominal exchange rate and the real income differential are found to be weakly exogenous to the exchange rate. This implies that the exchange rate itself and the real economic activity drive the exchange rate to its equilibrium. The driving force for the euro/ Japanese yen rate is the exchange rate itself, while for the euro/UK pound rate weak exogeneity has been established for the exchange rate, the money supply and the ination rate differentials. Finally, the euro/US dollar rate is driven to the long-run equilibrium by developments in the real economic activity. 4.3 Estimated current and total equilibrium exchange rates The evidence of cointegration allows us to claim that the monetary model can be considered as an equilibrium condition. However, the estimated coefcients of the variables are not always signed as the monetary model predicts. Removing any statistically insignicant coefcient, Table 5 shows that the sign of the money
10 Seasonally adjusted data for money supply and industrial production were obtained from the central bank and the ofcial statistical service of each sample country, respectively.

123

Empirica (2011) 38:555578 Table 6 Estimated coefcients of the cointegrating vectors (seasonally adjusted data) Parameter estimates b(yy*) b(mm*) b(pp*) Intercept Trend SB1 SB2 Euro area/China 0.270 (0.125) 2.468** (0.000) 0.077** (0.000) 1.228 (0.135) 0.025** (0.000) -0.302** (0.017) -0.563** (0.001) Euro area/Japan -6.165** (0.037) 1.611 (0.602) 1.638** (0.000) -7.794 (0.165) -0.015 (0.366) 0.136 (0.586) 0.600 (0.191) Euro area/UK 8.319** (0.006) -0.133** (0.000) -0.641 (0.101) -1.552 (0.869) -0.014** (0.047) -0.012** (0.003) -0.009* (0.051)

569

Euro area/USA 0.925 (0.166) -0.801** (0.003) 0.553** (0.000) -0.072 (0.148) -0.003** (0.018) -0.003 (0.310) -0.002 (0.361)

The above cointegrating vectors are presented having the explanatory variables on the right-hand side. b0 s are the parameters of the cointegrating vectors, normalised on the nominal exchange rate. SB1 and SB2 are the rst and the second structural break, respectively. Numbers in parentheses are the p-values of the likelihood ratio test statistics for the long-run exclusion tests. For the Euro area/UK and Euro area/USA cases the structural breaks are trend breaks, while for the Euro area/China and Euro area/Japan cases the structural breaks are breaks in the constant term ** (*) denotes rejection of the null hypothesis at the 0.05 (0.10) level of signicance Table 7 Adjustment coefcients and weak exogeneity tests Parameter estimates aS a(yy*) a(mm*) a(pp*)
0

Euro area/China -0.006 (0.679) 0.029 (0.851) 0.034** (0.000) 1.481** (0.000)

Euro area/Japan -0.011 (0.380) -0.076** (0.000) 0.007** (0.006) 0.981** (0.000)

Euro area/UK -0.026 (0.385) 0.414** (0.000) -0.004 (0.739) 0.238 (0.560)

Euro area/USA -0.251** (0.000) 0.001 (0.955) -0.036** (0.003) 2.157** (0.000)

a s are the adjustment coefcients. Numbers in parentheses are the p-values of the likelihood ratio test statistics for H0: ai = 0 ** (*) denotes rejection of the null hypothesis at the 0.05 (0.10) level of signicance

supply differential is as expected only in the Euro area/China case. In contrast, the results show that the equilibrium exchange rate of the euro against the UK pound and the US dollar is expected to appreciate when the eurozones monetary expansion is relatively higher. This positive sign can be possibly explained if we assume unchanged money demand. Then, a higher level of domestic money supply will reduce interest rate and thus, the expected ination. Given that eurozones money supply grows more than the foreign one and assuming rational expectations, the relatively lower eurozones ination makes domestic goods preferable, improves the trade balance and appreciates the domestic currency. Similarly, output differential has the expected negative sign in the euro/Japanese yen exchange rate equation. But, the euro/UK pound exchange rate is positively related to output differential. This means that a relatively higher output growth in the eurozone is expected to depreciate the euro. A possible explanation may be given by considering the effects of increased income on domestic consumption. Imports may increase more than exports leading to deterioration of the EMU members trade balances and thus, causing depreciating pressures on the euro. Finally, ination rate differential enters all the exchange rate equations with the

123

570

Empirica (2011) 38:555578

expected sign, except for the UK that is not statistically signicant. Specically, if the eurozones ination grows more than the foreign one, euro is expected to depreciate in the long-run. An interesting issue that arises from the above results is that the coefcient of ination rate differential is far from unity in all cases. In contrast to the PPP condition, which argues that exchange rate movements are induced by changes in relative price levels, the above nding implies that the nominal exchange rate does not fully reect changes in ination. Hence, the behaviour of the examined exchange rates is not solely explained by the PPP hypothesis, but it is also affected by other fundamental variables, as indicated by the theoretical assumptions of the monetary model. Now, we move on to the estimation and discussion of the behavioural equilibrium exchange rate. 4.3.1 Euro/Chinese yuan equilibrium exchange rate Based on Table 5, the euro/Chinese yuan equilibrium exchange rate is:11 st 2:032 mt m t 0:066 pt1 pt1 0:018trend 0:313SB1 0:327SB2 : 19 Equation (19) is the current equilibrium exchange rate and by subtracting this rate from the actual exchange rate we get the current misalignment rate. To estimate the BEER, we get the equilibrium values of the macroeconomic fundamentals via the Hodrick and Prescott (1997) (HP) lter. This smoothing approach estimates the long-run components of any given variable. However, its statistical properties have been criticised a lot. One issue is its poor performance near the end of the sample (Mise et al. 2005; Kaiser and Maravall 1999; Baxter and King 1999). To avoid this inconsistency, we followed Kaiser and Maravall (1999) and estimated optimal ARIMA forecasts. Then, we applied the HP lter to the extended series. Getting the smoothed values of the fundamentals and introducing them into (19), we estimate the BEER. Both rates along with the actual series are presented in Fig. 1. The left-hand side of the gure illustrates the relationship between the actual exchange rate and the current equilibrium exchange rate (LRER), while the righthand side of the gure plots the actual exchange rate with the BEER. If the actual exchange rate is higher (below) than any of the LRER and the BEER, euro is said to be undervalued (overvalued). Figure 1 show that the actual exchange rate is constantly below both rates. Thus, the euro (yuan) is monotonically overvalued (undervalued) for the whole period. Hardly surprising, this nding is in line with the common view of an undervalued yuan (Cline 2007; Coudert and Couharde 2005; Funke and Rahn 2005; MacDonald and Dias 2007).12 The undervaluation status of the Chinese yuan explains pretty well the huge increase in Chinese foreign exchange reserves and the expansion of the countrys
The industrial production (IP) differential and the intercept are not reported in (19) because were found to be statistically insignicant. Moreover, SB1 stands for the rst structural break in October 2004, while SB2 stands for the second break in April 2006.
12 For a comprehensive survey of recent estimates of the equilibrium effective and bilateral exchange ` -vis the US dollar, see Cline and Williamson (2007). rate of the Chinese currency vis-a 11

123

Empirica (2011) 38:555578

571

Euro/Chinese yuan
-1.0
Actual LRER

-1.0
Actual BEER

-1.5

-1.5

-2.0

-2.0

-2.5

-2.5

-3.0 99 00 01 02 03 04 05 06 07 08

-3.0 99 00 01 02 03 04 05 06 07 08

Fig. 1 Current and behavioural equilibrium euro per yuan exchange rate

global current account surplus.13 These facts reect the Chinese applied exchange rate policy during the October 1997July 2005 period, in which the nominal exchange rate was pegged to the US dollar. Figure 1 shows that, despite the permanent undervaluation of the yuan, the general trend of the actual exchange rate is consistent with the movements implied by the current and total equilibrium exchange rates. Apart from the period 19992002, in which the actual exchange rate moves upward but the BEER is relatively stable, the subsequent appreciation (depreciation) trend of the euro (yuan) is in line with the trend of the BEER. This implies that Chinese authorities have retained technically, by intervening in the foreign exchange (Forex) market, the exchange rate below its equilibrium value.14 However, this external imbalance has raised a current academic debate on the effectiveness of the above exchange rate policy and the necessity of reforms. The exchange rate revaluation, as a result of the abandonment of the xed regime, has changed the nature of exchange rate misalignment. Since July 2005, the actual exchange rate has begun to move towards BEER implying that the euro/yuan exchange rate follows an equilibrium process. Although the BEER implies that yuan should continue to depreciate, the actual exchange rate is moving upward reecting the need for appreciation of the Chinese currency. This difference in the trend does not necessarily imply further exchange rate misalignment, but it can be seen as a disequilibrium correction movement. At the end of the estimated period, the euro overvaluation rate (yuan undervaluation rate) is only about 1%, indicating that the exchange rate almost meets its equilibrium value.

China runs a growing trade surplus with the European countries and the US, but it runs increasing decits with developing and emerging Asian economies.
14 Of course, and especially for the period until 2005 when the Chinese currency was completely pegged to the US dollar, the dynamic behaviour of the euro/yuan exchange rate is partly explained by the movements of the euro/US dollar exchange rate, which is analysed below. It is indicative that both exchange rates and the corresponding equilibrium rates follow the same downward pathway (see, Figs. 1 and 4). Moreover, during this period, both estimated equilibrium rates imply identical misalignment status for the euro (i.e. overvaluation).

13

123

572

Empirica (2011) 38:555578

Euro/Japanese yen
3
Actual LRER

3
Actual BEER

2 1 0 -1 -2 -3 99 00 01 02 03 04 05 06 07 08

2 1 0 -1 -2 -3 99 00 01 02 03 04 05 06 07 08

Fig. 2 Current and behavioural equilibrium euro per yen exchange rate

4.3.2 Euro/Japanese yen equilibrium exchange rate The LRER of the euro per Japanese yen exchange rate is given by:15 st 4:163 yt y t 2:015 pt1 pt1 0:019trend 0:443SB2 :

20

In order to estimate of the BEER, we applied again the modied HP lter and then, we substituted the actual series in (20) by their ltered series. Both the LRER and the BEER are illustrated in the left-hand and right-hand side of Fig. 2, respectively. As shown, apart from the 1999 to 2001 period, the euro follows a stable appreciating trend. Both the LRER and BEER curves conrm the euro appreciation and they also imply a higher appreciation rate after 2004. Based on the BEER results, the estimated period can be decomposed into three sub-periods with different misalignment implications. The rst one, running from the beginning of our sample period until the rst half of 2001, implies that the euro was overvalued. In the subsequent period, from mid-2001 to early 2003, the euro/ yen exchange rate is very close to its equilibrium rate. In the nal sub-period from mid-2003 until the end of the estimated period, the euro is undervalued. Obviously, the Japanese yen is undervalued in the rst sub-period and overvalued during the nal sub-period. The last evidence of an overvalued yen contradicts the general view that the Japanese monetary authorities technically devaluate the exchange rate to retain the yen below its equilibrium level.16 However, our ndings refuse the above argument. We have reasons to believe that the depreciation of the yen (or the appreciation of the euro) is fully compatible with the macroeconomic status of the Japanese economy. Japan faces signicant internal and external imbalances: the current account surplus (external imbalance)
15 The intercept, the money supply differential and the rst structural break (SB1) in July 2000, are not reported in (20) because were found to be statistically insignicant. The second structural break (SB2), which is statistically signicant, is located in April 2001. 16 Similarly to the euro/yuan case, the euro/yen exchange rate is partly explained by the movements of the euro/US dollar exchange rate, because the Japanese currency is oriented more to the US dollar than the euro.

123

Empirica (2011) 38:555578

573

pushes the yen upward, while the continuous economic recession (internal imbalance) depreciates the yen.17 Besides to the extended slump and deation, Japanese economy also suffers from nancial sectors inefciency problems. These problems, along with the global nancial crisis, are deteriorating the efciency and the performance of domestic nancial intermediaries even more, and are expected to affect economic growth negatively.18 Thus, the Japanese authorities should continue the quantitative easing monetary policy, which may lead to a signicant devaluation of the yen. Furthermore, the yen is pressured downward by Japans high budget decit and permanent government debt, which force Japanese authorities to apply scal discipline. Expansionary monetary policy together with restrictive scal policy will probably cause expectations of yen depreciation. Moving one step further, we claim that the depreciating trend of the yen is not only normal but also that the depreciation rate is smaller than it should be. This argument is in line with the implications derived from Fig. 2. During the nal sub-period, the exchange rate follows a disequilibrium process. This can be explained by the pressures of the advanced Western economies for a stronger yen. As long as Japans internal imbalances remain together with the prevailing view that the Euro areas international trade is negatively affected by a weak yen, the exchange rate is not likely to approach its equilibrium level. 4.3.3 Euro/UK pound equilibrium exchange rate Similarly, the LRER is given by (21), while the BEER comes up by ltering the macroeconomic fundamentals.19 st 0:116 mt m 21 t 3:618 yt yt 0:007trend 0:002SB1 : Both rates are shown in Fig. 3. Although the LRER is highly volatile, one can observe the appreciating trend for the euro, which is clearly shown by the BEER line. Based on the BEER gure, we can decompose the whole estimated period into four sub-periods according to the sign of exchange rate misalignment. In 1999 the euro was overvalued. Then, the actual euro/UK pound exchange rate was close to its equilibrium level for the period from 2000 to mid-2001. The elimination of the euro overvaluation rate can be attributed to the upward movement (towards the BEER) of the actual exchange rate that started from 1999. Gomez et al. (2007) showed that a possible reason for the depreciation of the euro after its introduction is a higher transaction cost of the euro relative to the German mark. In our analysis, we argue ` -vis the UK pound was nothing more than a that the depreciation of the euro vis-a disequilibrium correction movement.
17

Rosenberg (2003) states that the Japanese yen is expected to depreciate because the internal imbalance will offset any positive developments on the external balance.

The effects of stock market developments on economic growth are, among others, analysed by Arestis and Demetriades (1997), De Avila (2003), Gardener et al. (2001), Levine and Zervos (1998).
19 The intercept, the ination rate differential and the second structural break (SB2) in August 2006, are not reported in (21) because were found to be statistically insignicant. The rst structural break (SB1), which is statistically signicant, is located in December 2002.

18

123

574

Empirica (2011) 38:555578

Euro/UK pound
1.0
Actual LRER

1.0
Actual BEER

0.5

0.5

0.0

0.0

-0.5

-0.5

-1.0 99 00 01 02 03 04 05 06 07 08

-1.0 99 00 01 02 03 04 05 06 07 08

Fig. 3 Current and behavioural equilibrium euro per pound exchange rate

In the subsequent period, from mid-2001 to the end of 2007, the euro (pound) is in general undervalued (overvalued). The actual exchange rate followed a decreasing trend since 2002, while the BEER started to decline since 2001. This implies that although both the actual and the equilibrium exchange rates follow similar pathways, the euro appreciation has been held by delay. The increase of the UK ination in the mid-2001 should have depreciated the pound, but this effect is observed only in the BEER. Instead, the actual exchange rate continues to increase until early 2002. Similarly, the actual exchange rate decreased to reach the BEER around mid-2003, but the following appreciation of the pound was not consistent with the BEER, which continued to decrease. These ndings imply that the euro/UK pound exchange rate had a tendency to remain above its equilibrium level. This result is consistent with Alberola et al. (1999), which mention that the transformation of the UK economy from a net creditor to a net debtor should depreciate the pound. Also, Wren-Lewis (2003) argues that the pound was overvalued against the euro due to the increased capital ows in the UK. Finally, from mid-2007 until the end of the estimated period, the actual exchange rate seems to follow an equilibrium process. It follows a decreasing pathway, while the BEER implies a stable exchange rate. This means that the euro/UK pound exchange rate moves towards its equilibrium value. The appreciation of the euro continues until the time that the actual exchange rate meets the BEER, in early 2008. Thus, at the end of the estimated period the euro/UK pound exchange rate does not deviate signicantly from its equilibrium rate. 4.3.4 Euro/US dollar equilibrium exchange rate Moving on to the nal exchange rate under investigation, the LRER is: st 1:05 mt m t 0:559 pt1 pt1 0:003trend
20

22

20 The two structural breaks (March 2003 and March 2005), along with the constant term and the IP differential, are not reported in (22) because were found to be statistically insignicant.

123

Empirica (2011) 38:555578

575

Euro/US dollar
1.0
Actual LRER

1.0
Actual BEER

0.5

0.5

0.0

0.0

-0.5

-0.5

-1.0 99 00 01 02 03 04 05 06 07 08

-1.0 99 00 01 02 03 04 05 06 07 08

Fig. 4 Current and behavioural equilibrium euro per dollar exchange rate

As shown in the left-hand side of Fig. 4, the actual exchange rate is mainly below the LRER apart from few and small in duration time periods. Also, the estimated BEER implies a persistent overvaluation of the euro against US dollar. Even though the BEER is consistent with the general trend of the actual exchange rate, we provide evidence of exchange rate misalignment in the sense that the euro has appreciated more than the macroeconomic fundamentals have indicated. In contrast to the above general nding, the euro was gradually depreciating against the US dollar during 19992002 sub-period. Although the BEER suggests an appreciating euro, this depreciating pathway can be considered as a disequilibrium correction movement. Indeed, the right-hand side of Fig. 4 illustrates that the actual exchange rate approached the BEER at least three times during the period 20002002. Rosenberg (2003) attributes the dollar appreciation to the positive market expectations about the long-run perspectives of the US economy. Also, Belloc and Federici (2008) explain the euro depreciation during this period through the portfolio balance model. These authors state that the excess supply of euro assets combined with the increasing condence on the US economy enforced the Euro areas residents to hold foreign (US) assets. Since the BEER is described by a linear curve with a negatively constant slope (Fig. 4), the magnitude of exchange rate misalignment depends on the behaviour of the actual exchange rate. During the 20022004 period the actual exchange rate is moving away from its equilibrium level, mainly due to the signicant appreciating process of the euro against the US dollar that begun in mid-2002. Belloc and Federici (2008) argue that the dollar depreciation was the outcome of the high increase of the US current account decit. Furthermore, Rosenberg (2003) argues that the depreciation of the dollar has been dictated by (a) the excessive increase in the US investment spending that led to an unsustainable savings-investment balance, and (b) the fact that the increase in the US productivity in late 1990s was not symmetrically distributed across all sectors of the US economy. Moreover, the rapid depreciation of the dollar was not irrelevant to the aftermaths of the terrorist attacks upon the USA on September 11, 2001. The depreciating process was

123

576

Empirica (2011) 38:555578

interrupted in the 20042006 period, in which the dollar appreciated against the euro. This movement has been motivated by the increase in the US output growth rate. However, neither the terrorist attack nor the higher output growth has been reected to the behaviour of the BEER. This is because the former as an unanticipated shock and the latter as a temporary effect cannot determine the BEER.21 From 2006 onwards, the dollar continues to depreciate against the euro as a result of the low level of the US output growth. Overall, our evidence implies a persistent overvaluation of the euro against the US dollar. Belloc and Federici (2008) estimate the NATREX and show an undervalued euro during the 19992003 period. But, their out-of-sample estimations imply overvaluation of the euro at the end of 2007. Similarly, Benassy-Quere et al. (2008) provide BEER and FEER estimates. The former implies an overvalued euro at 2005, while the latter shows an undervalued euro at the same time. When they considered the effect of asset price crash on US assets, their FEER estimate was very close to the actual exchange rate. Comparing our BEER estimate with the FEER and NATREX estimates of the recent literature, one can easily observe the difference in the misalignment implications. Our BEER model is based on the fundamentals of the monetary model, while the FEER and NATREX methodologies use as a building variable the current account decit. Thus, the continuously increasing US current account decit explains why the FEER and NATREX models imply that the US dollar was overvalued.

5 Concluding remarks In this paper we investigated if the nominal euro exchange rate against the currencies of China, Japan, the UK and the USA converges or not to its equilibrium level. We used the BEER model, having strengthened it with the theoretical assumptions of the monetary model of exchange rate determination. We evaluated these issues using cointegration techniques, in the presence of structural breaks in the data. Our ndings establish evidence of a valid long-run relationship between each of the four nominal exchange rates and the fundamentals dened by the monetary model. The BEER analysis indicates a general overvalued euro in relation to the Chinese yuan apart from the end of the estimated period, in which the exchange rate moves towards its equilibrium value. This nding well explains the huge increase of the countrys foreign exchange reserves and the expansion of its current account surplus. Our BEER results about the euro/yen nominal exchange rate indicate that even though the euro was overvalued until mid-2001, is considered as undervalued after mid-2003. The latter result contradicts the general view of an undervalued yen and can be explained by the depreciation expectations created from the combination of expansionary monetary and restrictive scal policies followed by the Japanese
21 As we already explained, the BEER is a function of the equilibrium values of the fundamentals. Namely, temporary shocks as well as the macroeconomic policy with only temporary effects on the economy can explain why exchange rates deviate from their equilibrium values in the long-run.

123

Empirica (2011) 38:555578

577

government, together with the pressures of the advanced Western economies for a stronger yen. The BEER results for the euro/UK pound exchange rate show that even though the euro was undervalued during the 20012007 period, the euro/UK pound exchange rate moves towards its equilibrium value from early 2008. Finally, our evidence implies that the euro is persistently overvalued against the US dollar, which means that it has appreciated more than the macroeconomic fundamentals have indicated. Concerning our motivation to examine the possibility of internal and external imbalance in the euro zone, we found evidence that would threaten stability for the euro/Japanese yen and the euro/US dollar exchange rates. This is because the observed misalignments may lead to future exchange rate uctuations even though ` -vis the exchange rates are currently stable. In contrast, the euro exchange rates vis-a the Chinese and the UK currencies are found to be very close to their equilibrium rates at the end of the estimated period and thus, we do not expect signicant future exchange rate uctuations. Additionally, we do not expect weakening of the eurozones internal balance due to the dynamic behaviour of the euro. The euro/Chinese yuan and the euro/UK pound exchange rates follow equilibrium processes, while the euro is overvalued against the US dollar and undervalued against the Japanese yen. Signicant internal imbalances would be expected if the euro were monotonically overvalued or undervalued in all the examined exchange rates. Hence, any loss of competitiveness ` -vis the US dollar can be offset by gains of due to the overvalued euro vis-a competitiveness due to the undervalued euro against the Japanese yen. The nal outcome depends on the relevant size of the misalignment rate and the relevant importance of euro zones trade with the USA and Japan.
Acknowledgments The authors are grateful to Carsten Trenkler for kindly providing them with the Gauss codes for the cointegration estimations with structural breaks, in Sect. 4.2. They also like to thank two anonymous referees for their constructive suggestions and helpful comments, which improved the quality and the exposition of the paper. Of course, the usual caveat applies.

References
Alberola E, Cervero S, Lopez H, Ubide A (1999) Global equilibrium exchange rates: euro, dollar, Ins, Outs and other major currencies in a panel cointegration framework. IMF working paper 99/175 Arestis P, Demetriades P (1997) Financial development and economic growth: assessing the evidence. Econ J 107:783799 Baxter M, King R (1999) Measuring business cycles: approximate band-pass lters for economic time series. Rev Econ Stat 81:575593 Belloc M, Federici D (2008) A two-country NATREX model for the euro/dollar. CESifo working paper no. 2290 Benassy-Quere A, Bereau S, Mignon V (2008) Equilibrium exchange rates: a guidebook for the eurodollar rate. CEPII working paper 2008-02 Clark P, MacDonald R (1998) Exchange rates and economic fundamentals: a methodological comparison of BEERs and FEERs. IMF working paper 67 Cline WR (2007) Estimating reference exchange rates. Workshop on policy to reduce global imbalances. Washington, February 89 Cline WR, Williamson J (2007) Estimates of the equilibrium exchange rate of the renminbi: is there a consensus and if not, why not? Peterson Institute for International Economics

123

578

Empirica (2011) 38:555578

Coudert V, Couharde G (2005) Real equilibrium exchange rates in China. CERII working paper 2005-01 De Avila DR (2003) Finance and growth in the EU: new evidence from the liberalization and harmonization of the banking industry. ECB working paper no 266 Egert B (2004) Assessing equilibrium exchange rates in CEE acceding countries: can we have DEER with BEER without FEER? A critical survey of the literature. BOFIT discussion papers 1/2004, Bank of Finland, Institute for Economies in Transition Franses PH, McAleer M (1998) Cointegration analysis of seasonal time series. J Econ Surv 12:651678 Frenkel J (1976) A monetary approach to the exchange rate: doctrinal aspects and empirical evidence. Scand J Econ 78:200224 Funke M, Rahn J (2005) Just how undervalued is the Chinese renminbi? World Econ 28:465489 Gardener E, Molyneux P, Moore B (2001) The impact of the single market program on EU banking. Serv Ind J 2:4770 Ghysels E, Perron P (1993) The effect of seasonal adjustment lters on tests for a unit root. J Econom 55:5798 Gomez M, Melvin M, Nardari F (2007) Explaining the early years of the euro exchange rate: an episode of learning about a new central bank. Europ Econ Rev 51:505520 Groen J (2000) The monetary exchange rate model as a long-run phenomenon. J Int Econ 52:299319 Hodrick RJ, Prescott EC (1997) Postwar US business cycles: an empirical investigation. J Money Credit Bank 29:116 Johansen S (1988) Statistical analysis of cointegration vectors. J Econ Dynam Control 12:231254 Johansen S, Mosconi R, Nielsen B (2000) Cointegration analysis in the presence of structural breaks in the deterministic trend. Econom J 3:216249 Kaiser R, Maravall A (1999) Estimation of the business cycle: a modied Hodrick-Prescott lter. Spanish Econ Rev 1:175206 Lee HS, Siklos PL (1997) The role of seasonality in economics time series: reinterpreting money-output causality in US data. Int J Forecast 13:381391 Lee J, Strazicich MC (2003) Minimum Lagrange multiplier unit root test with two structural breaks. Rev Econ Stat 85:10821089 Lee J, Strazicich MC (2004) Minimum LM unit root test with one structural break. Working paper, Department of Economics, Appalachian State University Levine R, Zervos S (1998) Stock markets, banks, and economic growth. Am Econ Rev 88:537558 tkepohl H, Saikkonen P (2000) Testing for the cointegrating rank of a VAR process with a time trend. Lu J Econom 95:177198 tkepohl H, Saikkonen P, Trenkler C (2003) Comparison of tests for the cointegrating rank of a VAR Lu process with a structural shift. J Econom 113:201229 MacDonald R (1995) Long-run exchange rate modeling: a survey of the recent evidence. Int Monet Fund Staff Pap 42:437489 MacDonald R, Dias P (2007) BEER estimates and target current account imbalances. Workshop on Policy to Reduce Global Imbalances. Washington, February 89 MacDonald R, Taylor MP (1991) The monetary model of the exchange rate: long-run relationships and coefcient restrictions. Econ Lett 37:179185 MacDonald R, Taylor MP (1992) Exchange rate economics: a survey. Int Monet Fund Staff Pap 39:157 Mise E, Kim T, Newbold P (2005) On suboptimality of the Hodrick-Prescott lter at time-series endpoints. J Macroecon 27:5367 Rosenberg MR (2003). The dollars equilibrium exchange rate: a market view. Institute for International Economics tkepohl H (2000) Testing for the cointegrating rank of a VAR process with structural Saikkonen P, Lu shifts. J Bus Econ Stat 18:451464 Stein J (2001) The equilibrium value of the euro/$ US exchange rate: an evaluation of research. CESifo working paper 525 Trenkler C (2008) Determining p-values for systems cointegration tests with a prior adjustment for deterministic terms. Comput Stat 23:1939 tkepohl H (2008) Testing for the cointegrating rank of a VAR process with Trenkler C, Saikkonen P, Lu level shift and trend break. J Time Ser Anal 29:331358 Wren-Lewis S (2003) Estimates of equilibrium exchange rates for sterling against the euro. HM Treasury, London

123

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.