Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Econometric Analysis of Croatia’s Proclaimed Foreign Exchange Rate

Econometric Analysis of Croatia’s Proclaimed Foreign Exchange Rate The officially proclaimed foreign exchange policy of the Croatian National Bank (CNB) is a managed float with a discretionary right of intervention on the Croatian kuna/euro foreign exchange (FX) market in order to maintain price stability. This paper examines the validity of three monetary policy hypotheses: the stability of the nominal exchange rate, the stability of exchange rate changes, and the exchange rate to inflation pass-through effect. The CNB claims a direct FX to inflation rate pass-through channel for which we find no evidence, but we find a strong link between FX rate changes and changes in M4, as well as between M4 changes and inflation. Changes in foreign investment Granger cause changes in monetary aggregates that further Granger cause inflation. Changes in FX rate Granger cause a reaction in M4 that indirectly Granger causes a further rise in inflation. Vector Autoregression Impulse Response Functions of changes in FX rate, M1, M4, and CPI confirm the Granger causalities in the established order. Keywords: central bank policies, monetary transmission effects, inflation targeting JEL classification: C22, E52, E58, F42 INTRODUCTION The Croatian National Bank (CNB) has recently changed its official policy from a free floating to a managed floating exchange regime (CNB 2013, CNB 2014). The CNB reserves the right to intervene on the currency markets and it did so more than 200 times in an 18 years period (1997-2014). The CNB has not officially determined an a priori upper or lower boundary or intervention point but it claims to maintain the stability of the kuna/euro foreign exchange (FX) rate in order to meet its primary objective of price stability. A similar approach was recently also taken by the Czech National Bank (CZNB 2014). The aim of the paper is to analyze the Croatian kuna/euro foreign exchange policy (FX policy) using Copyright © 2015 by the School of Economics and Business Sarajevo * Davor Mance, MA Assistant University of Rijeka, Faculty of Economics davor.mance@efri.hr Sasa Zikovi, PhD Associate Professor University of Rijeka, Faculty of Economics szikovic@efri.hr Diana Mance, PhD Senior Assistant University of Rijeka, School of Medicine diana.mance@uniri.hr an econometric approach. The three hypotheses tested in this paper and based upon CNB statements on their web pages are: (1) `the CNB does not predetermine the lower and upper level of the kuna exchange rate it is committed to defend (the upper and lower intervention point)'; (2) `the CNB participates in foreign exchange market transactions in order to prevent excessive exchange rate fluctuations in both directions', and (3) `the CNB maintains the stability of the kuna/ euro exchange rate in order to meet its primary objective of maintaining price stability' (CNB 2014). The stated reason for the proclaimed CNB policy is a presumed fast transmission channel between inflationary expectations and exchange rate changes (CNB 2014). This inflation pass-through effect is important as it prevents the exchange-rate policy from being an effective policy tool for employment and GDP growth. This is the reason why we analyze the statistical relationships and Granger causalities between Foreign Investments (FI), the foreign exchange (FX) rate and the consumer price index (CPI) via transmission monetary aggregates M1 and M4. The paper continues with a literature review with comments, followed by an explanation of our comprehensive data set and methodology. In the results and discussion section, statistical tests are consequently applied and commented on. LITERATURE REVIEW During the last 15 years several empirical studies on monetary transmission channels in Croatia have come to different conclusions. The presumed inability to model the primary FX rate time series because of its non-stationarity, non-normality, heteroscedasticity, and the presence of frequent structural breaks in the time series has motivated several other authors to model the FX returns instead, and to pursue various ARCH (Autoregressive Conditional Heteroscedasticity), TAR (Threshold Autoregression) and VAR (Vector Autoregression) approaches (Posedel 2006, Tica and Posedel 2009, Erjavec et al. 2012). Earlier exchange rate ­ inflation pass-through research did not come to converging results regarding the endogeneity of the exchange and inflation rate (Choudhri and Hakura 2001, Cukierman, Miller, and Neyapti 2002, Devereux and Engel 2003, Gagnon and Ihrig 2004, Mihaljek and Klau 2008). In regard to Croatia, Stucka (2004) found statistically significant J-curve effects with subsequent implications on investment, production, and international trade, with the latter having an influence on inflation. These are indirect effects that, according to the envelopment theory, should be disregarded in the direct assessment of FX - inflation pass-through effects. Subsequent research shows that the passthrough effect declines with increasing monetary stability and decreasing inflation (Mihaljek and Klau 2008). Monetary stability has, in this regard, a psychological memory effect and a non-linear relationship. For a small open economy such as Croatia, credit and liability euroization reduces the efficiency of the FX rate as a shock absorber, such that the positive effects of free floating are easily mitigated against (Devereux and Lane 2003). The question of the "fear of floating" (Calvo and Reinhart 2000) may be countered by the question of the "fear of commitment" in an environment involving the future obligation of every EU country (except UK and Sweden) to eventually join the EMU. The question of "either fix or float" and suboptimal intermediary policies has been discussed at great length (Mundell 1961, Friedman and Mundell 2001, Buiter and Grafe 2002). As Friedman and Mundell (2001) concluded, intermediary solutions are suboptimal. With credit and liability euroization constraints present in Croatia, it might have been optimal in the past to have a formal currency board as in Bosnia and Herzegovina and Montenegro, or, fast-forwarding to the present day, beneficial in the short run to make an earlier firm commitment to the Economic and Monetary Union. This may be the principal reason for the change of the FX policy description on the official CNB web site from free float to managed float. The authors' aim was to put this label to comprehensive econometric testing. DATA AND METHODOLOGY The time series of kuna/euro FX rate consists of monthly observations covering the period from January 1997 to April 2014 (CNB 2014). Consumer price index (CPI), foreign investment (FI), and M1 and M4 monetary aggregates data were comprehensively available only on a quarterly basis from Q4 of 2000 to Q4 of 2013 (CNB 2014). To model the FX time series, a Box-Jenkins methodology with a truncated Fourier series approach was used. Let yt be time series with t = 1, ..., N, where N is length of time series. In order to determine the seasonal variations and trend, the time series is divided into two components: where Yt is a stochastic irregular component and 2 cos truncated t is a deterministic periodic functionof the Fourier series form: (1) South European Journal of Economics and Business, Volume 10 (1) 2015 East c1 is the mean, c2 is the linear trend, c3 is the sea 4 sonality amplitude, cis the phase correction, T is the period and t is time in months. To model the stochastic component Yt in the equation (1), ARIMAmodelling of time series is used (Box, Jenkins, Reinsel2008). If Yt 2stationary one can is and a cos model construct p-order autoregressive (AR) (3) and/ or q-order moving average (MA) model (4): cos (2) Granger test is a standard bivariate regression: current In AR models, the value of the process is ex linear aggregate of previous values 1 pressed asa finite, a Vector Autoregression (VAR) approach is considered of the process Yt-i tcos MA models, t and white noise . In 2 Y depends on q of ran- (Johansen 1991). finitenumber . linearly previous Statistical tests and estimation of the model coefdom shocks t-i. When (3) and (4) are both included in onegets a mixed autoregressive-moving ficients was performed by the E-Views 7.2 statistical 2 one model, model. If Yt is 2 cos nonstationary, one can package. average (ARMA) 12 1 construct anautoregressive-integrated moving av erage (ARIMA) model of order (p, d, q), where d is the difference of process after dth 1the which stationarity is RESULTS AND DISCUSSION achieved. ARIMA model fitting was performed with a 2 R =0.885 three-stage Box-Jenkins technique: identification, esThe assessment of the correct label on the FX re . timation and verification (Maddala During the gime of a country requires a careful analysis of its time 2001). identification phase, the main tool was a visual analyseries, and testing whether the stochastic process of 2 sis 1 of the autocorrelation cos (ACF) and partial function the exchange rate values follows a mean reverting 2 12 autocorrelation function (PACF) (Enders 2010). process in response to central bank interventions. For To test the one series have assumption that may this purpose, we use Box-Jenkins time-series analy a delayed response to the other series the cross-corsis, Granger causality and Johansen cointegration (CCF) was analyzed. Autocorrelation relation function tests (Box, Jenkins, and Reinsel 2008, Granger 1969, R2=0.885 and cross-correlation coefficients are considered sigJohansen 1991). To the authors' knowledge the autorenificant within 1 the bounds. gressive (AR) kuna/euro FX time series has for the first time been augmented by a truncated Fourier series. The time series stationarity may be influenced by breaks in the data . structural breaks since structural Stability testing and Box-Jenkins can change the value of its mean, or the vector of its 2 movement. To 2 breaks theZivot- ARIMA modelling identify structural cos 1 12 Andrews test was used (Zivot and Andrews 1992). For the monthly FX series for the period January To test for stationarity in the time series we use the 1997­April 2014 the Zivot-Andrews test found an Augmented Dickey-Fuller test (ADF). endogenous structural break in level and trend in To identify R2=0.885 other transmission channels, the lags September 1998 (Fig. 1). Therefore, the observations and leads between the variables, and to simultanecan be grouped around two separate targets in level ously avoid spurious correlations, the CPI, FI, M1 and st and time: the period before and after Sep 1998. The M4 1 differences (differences assured stationarity) are structural break in level and trend shown in Figure Granger tested and the speed of the pass-through ef1 can be attributed mainly to the change in statistifects is tested with impulse response functions. The cal methodology and the introduction of the Value 9 (3) (4) for all pairs 1 of xt , yt series in the group (Granger 1 1969). The strength of causation is reported according to the .the Wald statistics for the F-statistics based on joint hypothesis: . 2 2 cos testedfor cointegra 1 Nonstationary variables are 2 12 2 are tions. If two nonstationary time series cos cointe- 1 12 grated with some stationary time series, a causal re lationship may be and assumed further tested with a Vector Error Correction (VEC) model (Engle and 2 R =0.885 Granger 1987). If the =0.885 R2 time series are not cointegrated, (5) FIGURES FIGURES -1 -1 -2 -2 -3 -3 -4 -4 -5 -5 1994 1994 1995 1995 1996 1996 1997 1997 1998 1998 1999 1999 2000 2000 2001 2001 2002 2002 2003 2003 Zivot-Andrew Breakpoints Zivot-Andrew Breakpoints Figure 1: Zivot-Andrews structural Data source: CNB 2014, calculation: E-Views 7.2. Figure 1: Zivot-Andrews structural break test. break test. Data source: CNB 2014, calculation: E-Views 7.2. Figure 1: Zivot-Andrews structural break test. Data source: CNB 2014, calculation: E-Views 7.2. Means Season by cos 7.8 7.7 7.8 7.6 7.7 kuna / euro/ euro kuna 7.5 7.6 7.4 7.5 7.3 7.4 7.2 7.3 7.1 7.2 7.1 Jan Jan Feb Feb Mar Mar Apr Apr May May Means by Season Jul Aug Sep Oct 7.8 7.7 7.8 7.6 7.7 kuna / euro/ euro kuna 7.5 7.6 7.4 7.5 7.3 7.4 7.2 7.3 Dec 7.1 7.2 7.1 Jul Aug Sep Oct Nov Dec 1 Figure 2: Kuna/euro FX rate season. Data source: CNB 2014, calculation: E-Views 7.2. Figure 2: Kuna/euro FX rate means by means by season. Data source: CNB 2014, calculation: E-Views 7.2. Jun Jun Nov kuna/euro FX rate is represented by the following Added Tax in 1998. After Sep 1998 the FX rate is fairly 2 equation: constant with a mean of 7.44 and a coefficient of vari 2 cos 1 7.8 12 ation of 1.9%. Further calculations and analysis are 2 7.8 7.6 restricted to the period between October 1998 and 2 12 1 7.6 April 2014. 7.4 2 (6) When analysing seasonal averages of the kuna/ R =0.885 7.4 4% 7.2 euro FX rate it can be seen that they exhibit a regu2% 4% 7.2 Parameter coefficients, standard errors as well as lar behaviour with maximum values in winter months, 7.0 0% 2% values in summer months (Fig. 2). Such standard statistical tests and diagnostic measures are and minimum 7.0 given in Table 1. The linear trend is statistically not sigseasonal behaviour can be explained by large tourism -2% 0% Residual receipts, an important driver for the Croatian econonificant (p> 0.1). Actual -4% -2% Fitted Residual my, peaking in summer months and euro Actual -6% -4% Table 1: Parameter values of the HRK/EUR 2012 Fitted 2014 FX rate time-series model. denominated 1999 2000 2001 2002peaking loan repayments 2003 2004 2005 2006 2007 2008 2009 2010 2011 2013 in winter -6% 1999 2000 2001this seasonal months. Due to 2002 2003 2004 2005 2006 2007 2008 Coefficient Std. Error 2013 2014 2009 2010 2011 2012 Variable t-Statistic Prob. behaviour the3: HRK/EUR FX rate series, model (periodic and AR(1) fit) and residuals. Data source: CNB Figure dynamics of the monthly C(1) 7.468 0.057 0.000 kuna/euro FXcalculation: E-Views 7.2. (1). model (periodic and AR(1) fit) and residuals. Data130.201 CNB 2014, rate can be described by Figure 3: HRK/EUR FX rate series, source: C(3) 0.052 0.009 5.271 0.000 Residuals remaining after the 7.2. 2014, calculation: E-Views removal of the periodic component were tested C(4) -1.385 0.188 -7.365 0.000 for stationarity. The ADF test showed that AR(1) 0.937 0.026 35.685 0.000 the series is stationary in level and trend. R-squared 0.885 Mean dependent var 7.444 The ACF of the residuals has dropped beAdjusted R-squared 0.884 S.D. dependent var 0.14 low the statistically significant level afS.E. of regression 0.048 Akaike info criterion -3.221 ter approximately two years. The series is characterized by a drop in the PACF after Sum squared resid 0.412 Schwarz criterion -3.151 only one month, without any significant Log likelihood 300.283 Hannan-Quinn criter. -3.192 reverse effect. This behaviour implies an F-statistic 463.475 Durbin-Watson stat 1.552 ARIMA(1,0,0) i.e. AR(1) process. The final model of the dynamics of the monthly Data source: CNB 2014, calculation: E-Views 7.2. Kuna Kuna / FX rate rate / euro euro FX Figure 2: Kuna/euro FX rate means by season. Data CNB calculation: E-Views 7.2. source: 2014, . Residuals Residuals Figure 2: Kuna/euro FX rate means by season. Data source: CNB 2014, calculation: E-Views 7.2. 7.8 Kuna / euro FX rate 7.6 7.4 4% Residuals 2% 0% -2% -4% -6% 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Residual Actual Fitted Figure 3: FX rate series, model series, model (periodic and AR(1) fit) and residuals. Data source: CNB Figure 3: HRK/EUR HRK/EUR FX rate (periodic and AR(1) fit) and residuals. Data source: CNB 2014, calculation: E-Views 7.2. 2014, calculation: E-Views 7.2. .15 .10 Quantiles of Normal Quantiles of Normal .05 .00 .20 Distance to points above median Distance to points above median .16 .12 .08 .04 .00 .20 .16 .12 .08 .04 .00 .00 .00 .04 .10 .05 .00 -.05 -.05 -.10 -.10 -.15 -.15 -.3 -.3 -.2 -.2 -.1 -.1 .0 .0 .1 .1 .2 .04 .08 .08 .12 .12 .16 .16 .20 .20 .24 Quantiles of Residuals Quantiles of Residuals Distance to pointspoints below median Distance to below median Figure 4a: Quantiles of residuals. Calculation: E-Views 7.2. Figure 4b: Residuals distances from median. Calculation: E-Views 7.2. Figure 4a: Quantiles of residuals. Calculation: Figure 4b: 4b: Residuals distances from median. Figure 4a: Quantiles of residuals. Calculation: Figure Residuals distances from median. The E-Views relationship between the original time series Calculation: E-Views 7.2. of FX rate and inflation rate that the absolute values 7.2. Calculation: E-Views E-Views 7.2. 7.2. and the model is shown in Fig. 3. The residuals are hoare not correlated. The CCF of FX rate and inflation moscedastic (White test: F-statistic=1.166, p=0.322; chain indices is also not statistically significant (all CCF Harvey test: F-statistic=1.732, p=0.145). coefficients were smaller than 0.15). The theoretical and symmetry quantile-quantile The period of 2001-2008 saw a stable kuna/euro 350,00350,00 1100,00 1100,00 plots in Fig. 4a and 4b show a close to normal distribuFX rate (Fig. 3), and as can be seen in Fig. 5, a fine up325,00325,00 1000,00 1000,00 tion of the residuals. ward slope of continuously compounded quarterly 300,00300,00 900,00900,00 exponential growth of 8% in FI (R2=0.97), cumulatively 275,00275,00 800,00800,00 800% between Q4/2000 and Q4/2007, and a 30% in250,00250,00 700,00700,00 crease in CPI (CBD 2014), coming to a halt after the Testing For Transfer Mechanisms 225,00225,00 600,00600,00 outbreak of the financial crisis in 2008. Looked at from 200,00200,00 500,00500,00 M1 M1 Testing for the FX rate/rate of inflation relationthis perspective, the crisis deprived Croatia of sig175,00175,00 400,00400,00 FI FI ships (correlation, cross-correlation, and Granger caunificant growth in FI and GDP. Since300,00300,00 and Croatia was 150,00150,00 sality) provides insight on the transfer mechanisms still is in an implicit currency peg regime, a stop in FI 125,00125,00 200,00200,00 between the100,00100,00correlation between FX rate and two. The growth rates also Granger caused a stop in growth 100,00100,00 2001 2001 2002 2003 2005 2006 2007 2007 2008 2010 aggregates 2013 5). 2002 2003 2004 conclusion 2008 monetary 2011 2012 (Fig. inflation is R=-0.11 (p>0.01), leading to the2004 2005 2006 rates of2009 2009 2010 2011 2012 2013 Index of net foreign investment (December 2000=100) Index of net foreign investment (December 2000=100) Figure 5: Levels of net foreign investments (FI) and monetary aggregate M1. M1. Data source: CBD 2014, Figure 5: Levels of net foreign investments (FI) and monetary aggregate Data source: CBD 2014, 11 CNBCNB 2014, own calculation. 2014, own calculation. Index of monetary aggregate M1 (December 2000=100) Index of monetary aggregate M1 (December 2000=100) E-Views 7.2. Calculation: E-Views 7.2. 350,00 325,00 300,00 275,00 250,00 225,00 200,00 175,00 150,00 125,00 100,00 M1 FI 1100,00 1000,00 900,00 800,00 700,00 600,00 500,00 400,00 300,00 200,00 100,00 Index of net foreign investment (December 2000=100) Figure 5: Levels of net foreign investments (FI) and monetary aggregate M1. Data source: CBD 2014, CNB 2014, own calculation. Table 2 shows the pairwise Granger causality tests on quarterly data (data was 1st differentiated to achieve stationarity) of several factors conjecturing causal relationships in an inflation pass-through effect. The reason for using quarterly data is the availability of foreign investment time series in quarterly data, as well as because quarterly data assured a stronger Table 2: Granger Causality Tests. Null Hypothesis: D(FX) does not Granger Cause D(CPI) D(CPI) does not Granger Cause D(FX) D(FX) does not Granger Cause D(FX) D(CPI) does not Granger Cause D(FI) D(M1) does not Granger Cause D(CPI) D(CPI) does not Granger Cause D(M1) D(M4) does not Granger Cause D(CPI) D(CPI) does not Granger Cause D(M4) D(FI) does not Granger Cause D(FX) D(FX) does not Granger Cause D(FI) D(M1) does not Granger Cause D(FX) D(FX) does not Granger Cause D(M1) D(M4) does not Granger Cause D(FX) D(FX) does not Granger Cause D(M4) D(M1) does not Granger Cause D(FI) D(FI) does not Granger Cause D(M1) D(M4) does not Granger Cause D(FI) D(FI) does not Granger Cause D(M4) D(M4) does not Granger Cause D(M1) D(M1) does not Granger Cause D(M4) INTERVENTION does not Granger Cause D(FX) D(FX) does not Granger Cause INTERVENTION Figure 5: Levels of net foreign investments (FI) and monetary aggregate M1. Data source: CBD 2014, CNB 2014, own calculation. expression of seasonality in the data, thus creating a better wave signal and improving the measurement of impulses in impulse response functions. The analysis of quarterly time series data confirmed no statistically significant direct transmission channel between the change in the FX rate and inflation (Table 2). Economic theory provides us with different Index of monetary aggregate M1 (December 2000=100) Obs. 50 50 50 50 50 50 50 50 50 50 231 F-Statistic 1.025 0.094 2.646 2.342 20.098 0.357 7.709 1.570 2.603 2.869 1.557 3.111 0.685 8.248 1.486 6.279 6.080 6.712 1.636 1.749 5.438 2.380 Prob. 0.367 0.911 0.082 0.108 0.000 0.702 0.001 0.219 0.085 0.067 0.222 0.054 0.509 0.001 0.237 0.004 0.005 0.003 0.206 0.186 0.000 0.030 FI=Foreign Investment; CPI=Consumer Price Index; FX=FX rate; M1&M4=Monetary Aggregates; Prefix D denotes first difference of data. Data source: CNB 2014, calculation: E-Views 7.2 determinants of inflation sources. Nevertheless "...inflation is always and everywhere a monetary phenomenon" (Friedman 1963). This statement was confirmed for the example of CNB monetary policy by testing the growth of the monetary aggregates M1 and M4 in regard to the change in the CPI (Table 2). There is Granger causation between the highly correlated variables of M4 changes and FI changes, between the changes in M4 and inflation (changes in CPI), as well as between FX rate changes and changes in M4. The strongest Granger causation of inflation comes from changes in the monetary aggregate M1 (F-statistic=20.1, p<0001). Strong Granger causation of changes in monetary aggregates from FI, with the FX rate staying relatively fixed, is a confirmation of an effective currency peg. Is it fixed, or does it float? The Mundell's Impossible trinity theory states that with free capital flows and stable exchange rates, a country cannot keep its monetary policy sovereign (Mundell 1961). The dilemma is between fixing and floating. The question of "fear of floating" (Calvo and Reinhart 2000) is countered by the question of "fear of commitment". Intermediary policies are unstable and for a small open economy lead to untrustworthy monetary policies (Buiter and Grafe 2002) the productive sector cannot commit to with long-term plans. In the case of corner solutions (fix or float) the markets are given unambiguous information about the government's reactions to shocks and about the risks from non-hedged foreign exchange positions (Schadler et al. 2004). A floating rate gives no particular advantage over the hard peg. It gives no real monetary policy independence in an environment of high exchange rate/inflation rate pass-through velocity. It is not an automatic shock absorber in a high liability euroization environment and does not discourage unhedged currency exposures because the financial derivatives markets are undeveloped. The elimination of currency risks enhances policy credibility and stimulates Foreign Investments (FI). The CNB has obviously discarded its own monetary policy sovereignty. The present low inflation rates are more a result of a stop in FI inflows than of FX policy (Fig. 5), and the presence of an intermediary monetary transmission channel between the two (Table 2). Figures 6a and 6b show the correlation between ln(FI) and ln(M4), and between ln(M4) and ln(CPI) in Croatia. Furthermore, they depict a possible transmission mechanism of inflation in Croatia. The mechanism starts with a pegged FX rate. Foreign investments get absorbed by the monetary aggregates M1 and M4 (sterilized or not), which further increase the CPI. There are actually no legal requirements for the CNB to use a double anchor for monetary stability. The CNB has the foremost constitutional obligation to preserve price stability. The CNB claims to "maintain the stability of the kuna/euro FX rate in order to meet its primary objective of maintaining price stability" (CNB 2014). As our econometric analysis shows,24 24 the best way to preserve price stability is by controlling monetary aggregates. Other intermediary goals may be pursued only if they are commensurate with the 6,1 6,1 5,9 5,9 ln of monetary aggregate M4 5,7 5,7 5,5 5,5 5,3 5,3 5,1 5,1 4,9 4,9 4,7 4,7 4,5 4,5 4,5 4,5 4,8 5,1 5,4 5,7 6 5 4,8 5,1 5,4 5,7 5 6 6,3 6,3 6,6 6,9 7,2 6,6 6,9 7,2 ln of monetary aggregate M4 R² = 0,9294 ² R² = 0,9294 ² ln of CPI 6,1 6,1 5,9 5,9 5,7 5,7 5,5 5,5 ln of CPI 5,3 5,3 5,1 5,1 4,9 4,9 4,7 4,7 4,5 4,5 75 75 4,6 4,64,65 4,654,7 4,74,7 4,7 4,8 4,84,85 4,854,9 4,9 ln of monetary ag ag M4 M4 ln of monetary aggregate aggregate R² = 0,9010,901 R² = ln o foreign investm ments (FI) of ln o foreign investm of ments (FI) Figure 6 6 CorrelatibetweenM4. Data and M4 Data guregure 6b: rrelation bet bet and CPI. and CPI. Figure 6a: 6a: Correlati ion and FI FI M4 CBD Figure 6b: Cor Cor ion 4. 4. rrelationtween tween and aCPI. a Figure 6a: Correlation between FI between and source:Data Fig Fig 6b: Correlation between M4 M4 M4 Data source: 2014, CNB 2014, own calculation. CBD 2014, CNB 2014, own calculation. source: C C 2014, C C2014, ownown calculati source: CBD 2014, CNB 2014, ocalculati ion. Dat Dat source: CBD 4, CNB 2014, 14, own CBD CNB o ion. ta source: CBD 2014 4, CNB 20 own ta 2014 calccalc culation. culation. 13 primary goal and this is the case under the absence of strong Balassa-Samuelson and Baumol-Bowen effects. FX rate stability guarantees CPI stability indirectly. There is Granger causality between the changes in FI, and inflation through the monetary M1 and M4 transmission channel. With the FX pegged to the euro, and Croatia's recession eventually coming to an end, the pressure on inflation might increase, so the timing of this change in the proclaimed CNB FX policy actually represents a positive signal towards EMU entry. This evolution comes as no surprise since Croatia has the highest level of euroization of all CEEC's and practically no ability to conduct a sovereign monetary policy. With nothing to lose, a firmer commitment may bring some reduction in risks and interest rates. 6,1 At the core of the question is the problem of heter5,9 odox goals and the effectiveness of the pass-through effect between FX returns and inflation. The pass5,7 through effect measures ² the response of inflation to R² = 0,9294 5,5 exchange rate changes. The smaller, more open and more internationally integrated an economy, the larg5,3 er the pass-through effect should be. The efficiency of the exchange rate as an adjustment mechanism dur5,1 ing the EMU convergence process depends highly on 4,9 the pass-through effect (Devereux and Engel 2003). A high pass-through coefficient means also a high cross4,7 correlation between the FX rate changes and inflation, 4,5 and reduced adjustment policy effectiveness. Such 4,5 4,8 5,1 5,4 5,7 6 5 6,3 6,6 6,9 7,2 high pass-through requires constant FX interventions. ln o foreign investm of ments (FI) Table 2 shows the results of testing the CNB FX interventions and FX moments for mutual Granger Figure 6a: possible between effectiveness in ion 4. causation as6 Correlati sign of CNB FI and M4 Data one its implicit currency peg without an explicit currency source: C CBD 2014, C CNB 2014, own calculati o ion. board. It shall be reminded once again, that the CNB interventions amount to less than 3% of the total kuna/ euro market volume (CNB 2014). The Granger causality test clearly shows that the CNB FX interventions ln of monetary aggregate M4 1,0 euro / kuna FOREIGN EXCHANGE rate of the intervention day relative to the FOREIGN EXCHANGE rate 4 days prior to intervention in % 0,8 0,6 0,4 0,2 0,0 -0,2 -0,4 -0,6 -0,8 -3.000 -2.500 -2.000 are effective since they do Granger cause the FX rate to change (p<0.001). Even more interestingly, there is Granger causation going from FX rate change to intervention (p=0.03), which leads to the conjecture that CNB interventions are based on observed developments in FX markets and are not pre-emptive (Mohanty and Berger 2013). A cross-correlation test of FX rates and FX interventions has shown a disappearance of the effect after two quarters. 24 We test if the FX rate is changed in the right direction. In Fig. 7 a correct slope of the regression can be seen, but there are too many shots missing their mark. Nevertheless, the kuna/euro FX rate is stable in the long run and without a statistically significant trend, showing a successful FX intermediate policy goal. 6,1 Johansen co-integrations tests between FX rate 5,9 (stationary), and combinations of M1, M4, and CPI (nonstationary) give no cointegrating vectors, re5,7 quiring the use of a VAR instead of a VEC model. We tested the VAR Impulse Response Functions of relative 5,5 R² = 0,901 changes in the FX rate, M1, M4, and CPI to test for the 5,3 inflation pass-through. Figure 8 shows the impulse response functions of a dependent variable to a shock 5,1 of a 1 change in the independent variable. All vari4,9 ables are differenced for stationarity and normalized. The time lag is shown in months. 4,7 Fig. 8 shows the expected response result from in4,5 flation to a 1 shock in the FX rate, and an expected 4,7 4,7 75 4,8 4,6 4,65 response result from inflation to a 14,85 shock4,9 the in ln of monetary ag aggregate M4 change of M4. The response from the M4 to a 1 shock in the change of the FX rate is also as expected. It Fig gure 6b: first, rrelation bet tween but a overshoots atCor then backlashes,M4 and CPI. then normalizes after four months. It may be concluded that the Dat source: CBD 2014from the 20 rate, via ta transmission channel goes 4, CNB FX14, own the M4 monetary aggregate to inflation. The response of calc the culation.in the FX rate to a 1 shock in the rate of change inflation representing the central bank intervention is ln of CPI y = -0.0 0002x + 0.0111 R² = 0.292 -1.500 -1.000 0 -500 0 500 1.0 000 Intervention amount in milli n ions of kuna 1.500 2 2.000 2.500 3.000 Figure 7: Euro/kuna FX rate changes as a consequence of FX interventions. Data source: CNB 2014, own representation. 7: anges as a co onsequence of FX interventions. Dat source: CN 2014, o ta NB Figure 7 Euro/kuna FX rate cha own repr resentation. Response of D(CPI) to D(FXRATE) 3% 2% 1% 0% -1% -2% -3% 2 4 6 8 10 12 3% 2% 1% 0% -1% -2% -3% 4% 2% 0% -2% -4% Response of D(M4) to D(FXRATE) 4% 2% 0% -2% -4% Response of D(CPI) to D(M4) 4% 2% 0% -2% -4% 4% 2% 0% -2% -4% 2% 1% 0% -1% -2% -3% Response of D(FXRATE) to D(CPI) 2% 1% 0% -1% -2% 2 4 6 8 10 12 -3% Figure 8: Impulse Response Functions of changes in FX, M1, M4, and CPI Data source: CNB 2014, calculation: E-Views 7.2. Fig. 8: Impulse Response Functions of changes lagged one month behind the inflation rate change, calculation: E-Views 7.2. confirming our hypothesis that the CNB is targeting the monthly changes in the inflation rate, i.e. a moving target. A currency peg gives a stable benchmark for the inflation rate. The surplus inflation over and above the eurozone inflation rate is then attributable to changes in relative prices. If the productivity growth is primarily in the internationally traded goods sector, the prices of tradables have to fall relative to the prices of nontradables. According to the Balassa-Samuelson effect, with fixed exchange rates the prices of non-tradables will have to rise. The CNB classifies its newly officially proclaimed FX policy as a managed float (CNB 2014). Nevertheless, the CNB acknowledges its commitment to pursue the multiple (heterodox) target policy (exchange rate and inflation targeting) as the best way to preserve low inflation rates. It is the CNB's constitutional obligation to pursue primarily the price stability target. Only once this goal has been fulfilled may the CNB pursue additional objectives. According to the CNB, its interventions on the exchange market accrue for no more than 3% of the total transactions volume. The largest amount has been used in the operations of selling kuna for euros in an attempt to prevent an appreciation of the kuna. Only recently has the CNB intervened in FX, M1, M4, and CPI Data source: CNB 2014, several times to buy the kuna in an attempt to prevent depreciation of the kuna (CNB 2013, CNB 2014). Depreciations feed inflation expectations and risk premiums. A higher risk premium reduces the real rate of return on capital. This is detrimental to investment and growth. Croatia would be better off if it could eliminate this negative and partially self-sustaining effect. But even without high exchange rate passthrough, a fixed exchange rate could be an optimal monetary policy in an environment of region-specific shocks (Devereux 2003). Countries with difficulties in controlling their money supply benefit from a currency board/monetary union because the money supply is the primary Granger cause of inflation. This is also the Croatian case where Granger causality test results show best values between growth of M1 and inflation. Nevertheless, confidence in the domestic currency may depend on factors other than the exchange rate regime such as central bank independence (Cukierman, Miller, and Neyapti 2002), the degree of the openness of the economy and the level of debt and its sustainability because of its inherent incentive to monetize (Romer 1993, Romer 1998, Terra 1998). These are some of the issues worth future research. CONCLUSION The goal of the paper was to test three statements made by the Croatian National Bank regarding its long term FX policy commitments, which are also often used by a number of other European central banks. The results of our econometric analysis show that the monthly weighted average kuna/euro FX rate is trend stationary, i.e. without a statistically significant trend, normally distributed and with homoscedastic residuals after modelling. To test the three hypotheses proposed by the CNB, it was also necessary to use the quarterly differentiated data of FI, FX, M1, M4, and CPI. To test the effectiveness of interventions, nominal FX and intervention data were tested for Granger causality. Following the results of time series stability analysis, the statement of FX rate stability cannot be rejected. It seems that the CNB indirectly pursues the goal set in the statement about not determining a priori boundaries or intervention points by directly pursuing the goal set in the statement about maintaining the stability of kuna/euro FX rate changes. The upper and lower intervention points may be generalized as FX rates' first differences. Since the CNB defends no nominal FX rate, but maintains its rate of change at ± 1.9% per month, the second hypothesis of no upper and lower bounds of nominal FX rate returns can be rejected. A Granger causality test confirms the bidirectional causation effects between FX rate and interventions. The FX rate change is an important trigger of intervention. The results show that for Croatia kuna/euro FX rate stability is not a direct mechanism of price stability. The Granger causality test shows that the CNB cannot directly target inflation via FX rate. Johansen co-integration tests do not show any expected cointegrating vectors between the analyzed variables, and the cross-correlation tests do not show any significant direct pass-through effects between the FX rate and inflation as stated by the CNB. We find that there is not enough statistical evidence for the confirmation of at least one out of three statements put forward by the CNB and which are usually considered absolutely correct and often serve as a starting point for fiscal/monetary policy decision-making. Nevertheless, we find a good political reason to change the official policy from free float to managed float: prospective EMU accession. Our FX rate model shows that the kuna/euro FX rate follows a very narrow currency peg of 7.44 kuna/euro with a coefficient of variation at 1.9%, so even this policy change may be assessed as too moderate, as it clearly is a functional currency peg or a quasi-currency board as it is sometimes also called. As already stated, a firmer FX policy commitment toward the euro may be the only policy choice for the future. Small open economies such as Croatia that joined the EU and enjoy full freedom of capital movement are confronted with a classic policy trilemma that becomes the dilemma of whether to fix or to float because of the inability of choosing capital controls. It seems the CNB has anchored, although without a strong commitment. The recent statement change is probably just one more step towards the EMU. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png South East European Journal of Economics and Business de Gruyter

Econometric Analysis of Croatia’s Proclaimed Foreign Exchange Rate

Loading next page...
 
/lp/de-gruyter/econometric-analysis-of-croatia-s-proclaimed-foreign-exchange-rate-Z7XHjkdITL
Publisher
de Gruyter
Copyright
Copyright © 2015 by the
ISSN
2233-1999
eISSN
2233-1999
DOI
10.1515/jeb-2015-0001
Publisher site
See Article on Publisher Site

Abstract

The officially proclaimed foreign exchange policy of the Croatian National Bank (CNB) is a managed float with a discretionary right of intervention on the Croatian kuna/euro foreign exchange (FX) market in order to maintain price stability. This paper examines the validity of three monetary policy hypotheses: the stability of the nominal exchange rate, the stability of exchange rate changes, and the exchange rate to inflation pass-through effect. The CNB claims a direct FX to inflation rate pass-through channel for which we find no evidence, but we find a strong link between FX rate changes and changes in M4, as well as between M4 changes and inflation. Changes in foreign investment Granger cause changes in monetary aggregates that further Granger cause inflation. Changes in FX rate Granger cause a reaction in M4 that indirectly Granger causes a further rise in inflation. Vector Autoregression Impulse Response Functions of changes in FX rate, M1, M4, and CPI confirm the Granger causalities in the established order. Keywords: central bank policies, monetary transmission effects, inflation targeting JEL classification: C22, E52, E58, F42 INTRODUCTION The Croatian National Bank (CNB) has recently changed its official policy from a free floating to a managed floating exchange regime (CNB 2013, CNB 2014). The CNB reserves the right to intervene on the currency markets and it did so more than 200 times in an 18 years period (1997-2014). The CNB has not officially determined an a priori upper or lower boundary or intervention point but it claims to maintain the stability of the kuna/euro foreign exchange (FX) rate in order to meet its primary objective of price stability. A similar approach was recently also taken by the Czech National Bank (CZNB 2014). The aim of the paper is to analyze the Croatian kuna/euro foreign exchange policy (FX policy) using Copyright © 2015 by the School of Economics and Business Sarajevo * Davor Mance, MA Assistant University of Rijeka, Faculty of Economics davor.mance@efri.hr Sasa Zikovi, PhD Associate Professor University of Rijeka, Faculty of Economics szikovic@efri.hr Diana Mance, PhD Senior Assistant University of Rijeka, School of Medicine diana.mance@uniri.hr an econometric approach. The three hypotheses tested in this paper and based upon CNB statements on their web pages are: (1) `the CNB does not predetermine the lower and upper level of the kuna exchange rate it is committed to defend (the upper and lower intervention point)'; (2) `the CNB participates in foreign exchange market transactions in order to prevent excessive exchange rate fluctuations in both directions', and (3) `the CNB maintains the stability of the kuna/ euro exchange rate in order to meet its primary objective of maintaining price stability' (CNB 2014). The stated reason for the proclaimed CNB policy is a presumed fast transmission channel between inflationary expectations and exchange rate changes (CNB 2014). This inflation pass-through effect is important as it prevents the exchange-rate policy from being an effective policy tool for employment and GDP growth. This is the reason why we analyze the statistical relationships and Granger causalities between Foreign Investments (FI), the foreign exchange (FX) rate and the consumer price index (CPI) via transmission monetary aggregates M1 and M4. The paper continues with a literature review with comments, followed by an explanation of our comprehensive data set and methodology. In the results and discussion section, statistical tests are consequently applied and commented on. LITERATURE REVIEW During the last 15 years several empirical studies on monetary transmission channels in Croatia have come to different conclusions. The presumed inability to model the primary FX rate time series because of its non-stationarity, non-normality, heteroscedasticity, and the presence of frequent structural breaks in the time series has motivated several other authors to model the FX returns instead, and to pursue various ARCH (Autoregressive Conditional Heteroscedasticity), TAR (Threshold Autoregression) and VAR (Vector Autoregression) approaches (Posedel 2006, Tica and Posedel 2009, Erjavec et al. 2012). Earlier exchange rate ­ inflation pass-through research did not come to converging results regarding the endogeneity of the exchange and inflation rate (Choudhri and Hakura 2001, Cukierman, Miller, and Neyapti 2002, Devereux and Engel 2003, Gagnon and Ihrig 2004, Mihaljek and Klau 2008). In regard to Croatia, Stucka (2004) found statistically significant J-curve effects with subsequent implications on investment, production, and international trade, with the latter having an influence on inflation. These are indirect effects that, according to the envelopment theory, should be disregarded in the direct assessment of FX - inflation pass-through effects. Subsequent research shows that the passthrough effect declines with increasing monetary stability and decreasing inflation (Mihaljek and Klau 2008). Monetary stability has, in this regard, a psychological memory effect and a non-linear relationship. For a small open economy such as Croatia, credit and liability euroization reduces the efficiency of the FX rate as a shock absorber, such that the positive effects of free floating are easily mitigated against (Devereux and Lane 2003). The question of the "fear of floating" (Calvo and Reinhart 2000) may be countered by the question of the "fear of commitment" in an environment involving the future obligation of every EU country (except UK and Sweden) to eventually join the EMU. The question of "either fix or float" and suboptimal intermediary policies has been discussed at great length (Mundell 1961, Friedman and Mundell 2001, Buiter and Grafe 2002). As Friedman and Mundell (2001) concluded, intermediary solutions are suboptimal. With credit and liability euroization constraints present in Croatia, it might have been optimal in the past to have a formal currency board as in Bosnia and Herzegovina and Montenegro, or, fast-forwarding to the present day, beneficial in the short run to make an earlier firm commitment to the Economic and Monetary Union. This may be the principal reason for the change of the FX policy description on the official CNB web site from free float to managed float. The authors' aim was to put this label to comprehensive econometric testing. DATA AND METHODOLOGY The time series of kuna/euro FX rate consists of monthly observations covering the period from January 1997 to April 2014 (CNB 2014). Consumer price index (CPI), foreign investment (FI), and M1 and M4 monetary aggregates data were comprehensively available only on a quarterly basis from Q4 of 2000 to Q4 of 2013 (CNB 2014). To model the FX time series, a Box-Jenkins methodology with a truncated Fourier series approach was used. Let yt be time series with t = 1, ..., N, where N is length of time series. In order to determine the seasonal variations and trend, the time series is divided into two components: where Yt is a stochastic irregular component and 2 cos truncated t is a deterministic periodic functionof the Fourier series form: (1) South European Journal of Economics and Business, Volume 10 (1) 2015 East c1 is the mean, c2 is the linear trend, c3 is the sea 4 sonality amplitude, cis the phase correction, T is the period and t is time in months. To model the stochastic component Yt in the equation (1), ARIMAmodelling of time series is used (Box, Jenkins, Reinsel2008). If Yt 2stationary one can is and a cos model construct p-order autoregressive (AR) (3) and/ or q-order moving average (MA) model (4): cos (2) Granger test is a standard bivariate regression: current In AR models, the value of the process is ex linear aggregate of previous values 1 pressed asa finite, a Vector Autoregression (VAR) approach is considered of the process Yt-i tcos MA models, t and white noise . In 2 Y depends on q of ran- (Johansen 1991). finitenumber . linearly previous Statistical tests and estimation of the model coefdom shocks t-i. When (3) and (4) are both included in onegets a mixed autoregressive-moving ficients was performed by the E-Views 7.2 statistical 2 one model, model. If Yt is 2 cos nonstationary, one can package. average (ARMA) 12 1 construct anautoregressive-integrated moving av erage (ARIMA) model of order (p, d, q), where d is the difference of process after dth 1the which stationarity is RESULTS AND DISCUSSION achieved. ARIMA model fitting was performed with a 2 R =0.885 three-stage Box-Jenkins technique: identification, esThe assessment of the correct label on the FX re . timation and verification (Maddala During the gime of a country requires a careful analysis of its time 2001). identification phase, the main tool was a visual analyseries, and testing whether the stochastic process of 2 sis 1 of the autocorrelation cos (ACF) and partial function the exchange rate values follows a mean reverting 2 12 autocorrelation function (PACF) (Enders 2010). process in response to central bank interventions. For To test the one series have assumption that may this purpose, we use Box-Jenkins time-series analy a delayed response to the other series the cross-corsis, Granger causality and Johansen cointegration (CCF) was analyzed. Autocorrelation relation function tests (Box, Jenkins, and Reinsel 2008, Granger 1969, R2=0.885 and cross-correlation coefficients are considered sigJohansen 1991). To the authors' knowledge the autorenificant within 1 the bounds. gressive (AR) kuna/euro FX time series has for the first time been augmented by a truncated Fourier series. The time series stationarity may be influenced by breaks in the data . structural breaks since structural Stability testing and Box-Jenkins can change the value of its mean, or the vector of its 2 movement. To 2 breaks theZivot- ARIMA modelling identify structural cos 1 12 Andrews test was used (Zivot and Andrews 1992). For the monthly FX series for the period January To test for stationarity in the time series we use the 1997­April 2014 the Zivot-Andrews test found an Augmented Dickey-Fuller test (ADF). endogenous structural break in level and trend in To identify R2=0.885 other transmission channels, the lags September 1998 (Fig. 1). Therefore, the observations and leads between the variables, and to simultanecan be grouped around two separate targets in level ously avoid spurious correlations, the CPI, FI, M1 and st and time: the period before and after Sep 1998. The M4 1 differences (differences assured stationarity) are structural break in level and trend shown in Figure Granger tested and the speed of the pass-through ef1 can be attributed mainly to the change in statistifects is tested with impulse response functions. The cal methodology and the introduction of the Value 9 (3) (4) for all pairs 1 of xt , yt series in the group (Granger 1 1969). The strength of causation is reported according to the .the Wald statistics for the F-statistics based on joint hypothesis: . 2 2 cos testedfor cointegra 1 Nonstationary variables are 2 12 2 are tions. If two nonstationary time series cos cointe- 1 12 grated with some stationary time series, a causal re lationship may be and assumed further tested with a Vector Error Correction (VEC) model (Engle and 2 R =0.885 Granger 1987). If the =0.885 R2 time series are not cointegrated, (5) FIGURES FIGURES -1 -1 -2 -2 -3 -3 -4 -4 -5 -5 1994 1994 1995 1995 1996 1996 1997 1997 1998 1998 1999 1999 2000 2000 2001 2001 2002 2002 2003 2003 Zivot-Andrew Breakpoints Zivot-Andrew Breakpoints Figure 1: Zivot-Andrews structural Data source: CNB 2014, calculation: E-Views 7.2. Figure 1: Zivot-Andrews structural break test. break test. Data source: CNB 2014, calculation: E-Views 7.2. Figure 1: Zivot-Andrews structural break test. Data source: CNB 2014, calculation: E-Views 7.2. Means Season by cos 7.8 7.7 7.8 7.6 7.7 kuna / euro/ euro kuna 7.5 7.6 7.4 7.5 7.3 7.4 7.2 7.3 7.1 7.2 7.1 Jan Jan Feb Feb Mar Mar Apr Apr May May Means by Season Jul Aug Sep Oct 7.8 7.7 7.8 7.6 7.7 kuna / euro/ euro kuna 7.5 7.6 7.4 7.5 7.3 7.4 7.2 7.3 Dec 7.1 7.2 7.1 Jul Aug Sep Oct Nov Dec 1 Figure 2: Kuna/euro FX rate season. Data source: CNB 2014, calculation: E-Views 7.2. Figure 2: Kuna/euro FX rate means by means by season. Data source: CNB 2014, calculation: E-Views 7.2. Jun Jun Nov kuna/euro FX rate is represented by the following Added Tax in 1998. After Sep 1998 the FX rate is fairly 2 equation: constant with a mean of 7.44 and a coefficient of vari 2 cos 1 7.8 12 ation of 1.9%. Further calculations and analysis are 2 7.8 7.6 restricted to the period between October 1998 and 2 12 1 7.6 April 2014. 7.4 2 (6) When analysing seasonal averages of the kuna/ R =0.885 7.4 4% 7.2 euro FX rate it can be seen that they exhibit a regu2% 4% 7.2 Parameter coefficients, standard errors as well as lar behaviour with maximum values in winter months, 7.0 0% 2% values in summer months (Fig. 2). Such standard statistical tests and diagnostic measures are and minimum 7.0 given in Table 1. The linear trend is statistically not sigseasonal behaviour can be explained by large tourism -2% 0% Residual receipts, an important driver for the Croatian econonificant (p> 0.1). Actual -4% -2% Fitted Residual my, peaking in summer months and euro Actual -6% -4% Table 1: Parameter values of the HRK/EUR 2012 Fitted 2014 FX rate time-series model. denominated 1999 2000 2001 2002peaking loan repayments 2003 2004 2005 2006 2007 2008 2009 2010 2011 2013 in winter -6% 1999 2000 2001this seasonal months. Due to 2002 2003 2004 2005 2006 2007 2008 Coefficient Std. Error 2013 2014 2009 2010 2011 2012 Variable t-Statistic Prob. behaviour the3: HRK/EUR FX rate series, model (periodic and AR(1) fit) and residuals. Data source: CNB Figure dynamics of the monthly C(1) 7.468 0.057 0.000 kuna/euro FXcalculation: E-Views 7.2. (1). model (periodic and AR(1) fit) and residuals. Data130.201 CNB 2014, rate can be described by Figure 3: HRK/EUR FX rate series, source: C(3) 0.052 0.009 5.271 0.000 Residuals remaining after the 7.2. 2014, calculation: E-Views removal of the periodic component were tested C(4) -1.385 0.188 -7.365 0.000 for stationarity. The ADF test showed that AR(1) 0.937 0.026 35.685 0.000 the series is stationary in level and trend. R-squared 0.885 Mean dependent var 7.444 The ACF of the residuals has dropped beAdjusted R-squared 0.884 S.D. dependent var 0.14 low the statistically significant level afS.E. of regression 0.048 Akaike info criterion -3.221 ter approximately two years. The series is characterized by a drop in the PACF after Sum squared resid 0.412 Schwarz criterion -3.151 only one month, without any significant Log likelihood 300.283 Hannan-Quinn criter. -3.192 reverse effect. This behaviour implies an F-statistic 463.475 Durbin-Watson stat 1.552 ARIMA(1,0,0) i.e. AR(1) process. The final model of the dynamics of the monthly Data source: CNB 2014, calculation: E-Views 7.2. Kuna Kuna / FX rate rate / euro euro FX Figure 2: Kuna/euro FX rate means by season. Data CNB calculation: E-Views 7.2. source: 2014, . Residuals Residuals Figure 2: Kuna/euro FX rate means by season. Data source: CNB 2014, calculation: E-Views 7.2. 7.8 Kuna / euro FX rate 7.6 7.4 4% Residuals 2% 0% -2% -4% -6% 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Residual Actual Fitted Figure 3: FX rate series, model series, model (periodic and AR(1) fit) and residuals. Data source: CNB Figure 3: HRK/EUR HRK/EUR FX rate (periodic and AR(1) fit) and residuals. Data source: CNB 2014, calculation: E-Views 7.2. 2014, calculation: E-Views 7.2. .15 .10 Quantiles of Normal Quantiles of Normal .05 .00 .20 Distance to points above median Distance to points above median .16 .12 .08 .04 .00 .20 .16 .12 .08 .04 .00 .00 .00 .04 .10 .05 .00 -.05 -.05 -.10 -.10 -.15 -.15 -.3 -.3 -.2 -.2 -.1 -.1 .0 .0 .1 .1 .2 .04 .08 .08 .12 .12 .16 .16 .20 .20 .24 Quantiles of Residuals Quantiles of Residuals Distance to pointspoints below median Distance to below median Figure 4a: Quantiles of residuals. Calculation: E-Views 7.2. Figure 4b: Residuals distances from median. Calculation: E-Views 7.2. Figure 4a: Quantiles of residuals. Calculation: Figure 4b: 4b: Residuals distances from median. Figure 4a: Quantiles of residuals. Calculation: Figure Residuals distances from median. The E-Views relationship between the original time series Calculation: E-Views 7.2. of FX rate and inflation rate that the absolute values 7.2. Calculation: E-Views E-Views 7.2. 7.2. and the model is shown in Fig. 3. The residuals are hoare not correlated. The CCF of FX rate and inflation moscedastic (White test: F-statistic=1.166, p=0.322; chain indices is also not statistically significant (all CCF Harvey test: F-statistic=1.732, p=0.145). coefficients were smaller than 0.15). The theoretical and symmetry quantile-quantile The period of 2001-2008 saw a stable kuna/euro 350,00350,00 1100,00 1100,00 plots in Fig. 4a and 4b show a close to normal distribuFX rate (Fig. 3), and as can be seen in Fig. 5, a fine up325,00325,00 1000,00 1000,00 tion of the residuals. ward slope of continuously compounded quarterly 300,00300,00 900,00900,00 exponential growth of 8% in FI (R2=0.97), cumulatively 275,00275,00 800,00800,00 800% between Q4/2000 and Q4/2007, and a 30% in250,00250,00 700,00700,00 crease in CPI (CBD 2014), coming to a halt after the Testing For Transfer Mechanisms 225,00225,00 600,00600,00 outbreak of the financial crisis in 2008. Looked at from 200,00200,00 500,00500,00 M1 M1 Testing for the FX rate/rate of inflation relationthis perspective, the crisis deprived Croatia of sig175,00175,00 400,00400,00 FI FI ships (correlation, cross-correlation, and Granger caunificant growth in FI and GDP. Since300,00300,00 and Croatia was 150,00150,00 sality) provides insight on the transfer mechanisms still is in an implicit currency peg regime, a stop in FI 125,00125,00 200,00200,00 between the100,00100,00correlation between FX rate and two. The growth rates also Granger caused a stop in growth 100,00100,00 2001 2001 2002 2003 2005 2006 2007 2007 2008 2010 aggregates 2013 5). 2002 2003 2004 conclusion 2008 monetary 2011 2012 (Fig. inflation is R=-0.11 (p>0.01), leading to the2004 2005 2006 rates of2009 2009 2010 2011 2012 2013 Index of net foreign investment (December 2000=100) Index of net foreign investment (December 2000=100) Figure 5: Levels of net foreign investments (FI) and monetary aggregate M1. M1. Data source: CBD 2014, Figure 5: Levels of net foreign investments (FI) and monetary aggregate Data source: CBD 2014, 11 CNBCNB 2014, own calculation. 2014, own calculation. Index of monetary aggregate M1 (December 2000=100) Index of monetary aggregate M1 (December 2000=100) E-Views 7.2. Calculation: E-Views 7.2. 350,00 325,00 300,00 275,00 250,00 225,00 200,00 175,00 150,00 125,00 100,00 M1 FI 1100,00 1000,00 900,00 800,00 700,00 600,00 500,00 400,00 300,00 200,00 100,00 Index of net foreign investment (December 2000=100) Figure 5: Levels of net foreign investments (FI) and monetary aggregate M1. Data source: CBD 2014, CNB 2014, own calculation. Table 2 shows the pairwise Granger causality tests on quarterly data (data was 1st differentiated to achieve stationarity) of several factors conjecturing causal relationships in an inflation pass-through effect. The reason for using quarterly data is the availability of foreign investment time series in quarterly data, as well as because quarterly data assured a stronger Table 2: Granger Causality Tests. Null Hypothesis: D(FX) does not Granger Cause D(CPI) D(CPI) does not Granger Cause D(FX) D(FX) does not Granger Cause D(FX) D(CPI) does not Granger Cause D(FI) D(M1) does not Granger Cause D(CPI) D(CPI) does not Granger Cause D(M1) D(M4) does not Granger Cause D(CPI) D(CPI) does not Granger Cause D(M4) D(FI) does not Granger Cause D(FX) D(FX) does not Granger Cause D(FI) D(M1) does not Granger Cause D(FX) D(FX) does not Granger Cause D(M1) D(M4) does not Granger Cause D(FX) D(FX) does not Granger Cause D(M4) D(M1) does not Granger Cause D(FI) D(FI) does not Granger Cause D(M1) D(M4) does not Granger Cause D(FI) D(FI) does not Granger Cause D(M4) D(M4) does not Granger Cause D(M1) D(M1) does not Granger Cause D(M4) INTERVENTION does not Granger Cause D(FX) D(FX) does not Granger Cause INTERVENTION Figure 5: Levels of net foreign investments (FI) and monetary aggregate M1. Data source: CBD 2014, CNB 2014, own calculation. expression of seasonality in the data, thus creating a better wave signal and improving the measurement of impulses in impulse response functions. The analysis of quarterly time series data confirmed no statistically significant direct transmission channel between the change in the FX rate and inflation (Table 2). Economic theory provides us with different Index of monetary aggregate M1 (December 2000=100) Obs. 50 50 50 50 50 50 50 50 50 50 231 F-Statistic 1.025 0.094 2.646 2.342 20.098 0.357 7.709 1.570 2.603 2.869 1.557 3.111 0.685 8.248 1.486 6.279 6.080 6.712 1.636 1.749 5.438 2.380 Prob. 0.367 0.911 0.082 0.108 0.000 0.702 0.001 0.219 0.085 0.067 0.222 0.054 0.509 0.001 0.237 0.004 0.005 0.003 0.206 0.186 0.000 0.030 FI=Foreign Investment; CPI=Consumer Price Index; FX=FX rate; M1&M4=Monetary Aggregates; Prefix D denotes first difference of data. Data source: CNB 2014, calculation: E-Views 7.2 determinants of inflation sources. Nevertheless "...inflation is always and everywhere a monetary phenomenon" (Friedman 1963). This statement was confirmed for the example of CNB monetary policy by testing the growth of the monetary aggregates M1 and M4 in regard to the change in the CPI (Table 2). There is Granger causation between the highly correlated variables of M4 changes and FI changes, between the changes in M4 and inflation (changes in CPI), as well as between FX rate changes and changes in M4. The strongest Granger causation of inflation comes from changes in the monetary aggregate M1 (F-statistic=20.1, p<0001). Strong Granger causation of changes in monetary aggregates from FI, with the FX rate staying relatively fixed, is a confirmation of an effective currency peg. Is it fixed, or does it float? The Mundell's Impossible trinity theory states that with free capital flows and stable exchange rates, a country cannot keep its monetary policy sovereign (Mundell 1961). The dilemma is between fixing and floating. The question of "fear of floating" (Calvo and Reinhart 2000) is countered by the question of "fear of commitment". Intermediary policies are unstable and for a small open economy lead to untrustworthy monetary policies (Buiter and Grafe 2002) the productive sector cannot commit to with long-term plans. In the case of corner solutions (fix or float) the markets are given unambiguous information about the government's reactions to shocks and about the risks from non-hedged foreign exchange positions (Schadler et al. 2004). A floating rate gives no particular advantage over the hard peg. It gives no real monetary policy independence in an environment of high exchange rate/inflation rate pass-through velocity. It is not an automatic shock absorber in a high liability euroization environment and does not discourage unhedged currency exposures because the financial derivatives markets are undeveloped. The elimination of currency risks enhances policy credibility and stimulates Foreign Investments (FI). The CNB has obviously discarded its own monetary policy sovereignty. The present low inflation rates are more a result of a stop in FI inflows than of FX policy (Fig. 5), and the presence of an intermediary monetary transmission channel between the two (Table 2). Figures 6a and 6b show the correlation between ln(FI) and ln(M4), and between ln(M4) and ln(CPI) in Croatia. Furthermore, they depict a possible transmission mechanism of inflation in Croatia. The mechanism starts with a pegged FX rate. Foreign investments get absorbed by the monetary aggregates M1 and M4 (sterilized or not), which further increase the CPI. There are actually no legal requirements for the CNB to use a double anchor for monetary stability. The CNB has the foremost constitutional obligation to preserve price stability. The CNB claims to "maintain the stability of the kuna/euro FX rate in order to meet its primary objective of maintaining price stability" (CNB 2014). As our econometric analysis shows,24 24 the best way to preserve price stability is by controlling monetary aggregates. Other intermediary goals may be pursued only if they are commensurate with the 6,1 6,1 5,9 5,9 ln of monetary aggregate M4 5,7 5,7 5,5 5,5 5,3 5,3 5,1 5,1 4,9 4,9 4,7 4,7 4,5 4,5 4,5 4,5 4,8 5,1 5,4 5,7 6 5 4,8 5,1 5,4 5,7 5 6 6,3 6,3 6,6 6,9 7,2 6,6 6,9 7,2 ln of monetary aggregate M4 R² = 0,9294 ² R² = 0,9294 ² ln of CPI 6,1 6,1 5,9 5,9 5,7 5,7 5,5 5,5 ln of CPI 5,3 5,3 5,1 5,1 4,9 4,9 4,7 4,7 4,5 4,5 75 75 4,6 4,64,65 4,654,7 4,74,7 4,7 4,8 4,84,85 4,854,9 4,9 ln of monetary ag ag M4 M4 ln of monetary aggregate aggregate R² = 0,9010,901 R² = ln o foreign investm ments (FI) of ln o foreign investm of ments (FI) Figure 6 6 CorrelatibetweenM4. Data and M4 Data guregure 6b: rrelation bet bet and CPI. and CPI. Figure 6a: 6a: Correlati ion and FI FI M4 CBD Figure 6b: Cor Cor ion 4. 4. rrelationtween tween and aCPI. a Figure 6a: Correlation between FI between and source:Data Fig Fig 6b: Correlation between M4 M4 M4 Data source: 2014, CNB 2014, own calculation. CBD 2014, CNB 2014, own calculation. source: C C 2014, C C2014, ownown calculati source: CBD 2014, CNB 2014, ocalculati ion. Dat Dat source: CBD 4, CNB 2014, 14, own CBD CNB o ion. ta source: CBD 2014 4, CNB 20 own ta 2014 calccalc culation. culation. 13 primary goal and this is the case under the absence of strong Balassa-Samuelson and Baumol-Bowen effects. FX rate stability guarantees CPI stability indirectly. There is Granger causality between the changes in FI, and inflation through the monetary M1 and M4 transmission channel. With the FX pegged to the euro, and Croatia's recession eventually coming to an end, the pressure on inflation might increase, so the timing of this change in the proclaimed CNB FX policy actually represents a positive signal towards EMU entry. This evolution comes as no surprise since Croatia has the highest level of euroization of all CEEC's and practically no ability to conduct a sovereign monetary policy. With nothing to lose, a firmer commitment may bring some reduction in risks and interest rates. 6,1 At the core of the question is the problem of heter5,9 odox goals and the effectiveness of the pass-through effect between FX returns and inflation. The pass5,7 through effect measures ² the response of inflation to R² = 0,9294 5,5 exchange rate changes. The smaller, more open and more internationally integrated an economy, the larg5,3 er the pass-through effect should be. The efficiency of the exchange rate as an adjustment mechanism dur5,1 ing the EMU convergence process depends highly on 4,9 the pass-through effect (Devereux and Engel 2003). A high pass-through coefficient means also a high cross4,7 correlation between the FX rate changes and inflation, 4,5 and reduced adjustment policy effectiveness. Such 4,5 4,8 5,1 5,4 5,7 6 5 6,3 6,6 6,9 7,2 high pass-through requires constant FX interventions. ln o foreign investm of ments (FI) Table 2 shows the results of testing the CNB FX interventions and FX moments for mutual Granger Figure 6a: possible between effectiveness in ion 4. causation as6 Correlati sign of CNB FI and M4 Data one its implicit currency peg without an explicit currency source: C CBD 2014, C CNB 2014, own calculati o ion. board. It shall be reminded once again, that the CNB interventions amount to less than 3% of the total kuna/ euro market volume (CNB 2014). The Granger causality test clearly shows that the CNB FX interventions ln of monetary aggregate M4 1,0 euro / kuna FOREIGN EXCHANGE rate of the intervention day relative to the FOREIGN EXCHANGE rate 4 days prior to intervention in % 0,8 0,6 0,4 0,2 0,0 -0,2 -0,4 -0,6 -0,8 -3.000 -2.500 -2.000 are effective since they do Granger cause the FX rate to change (p<0.001). Even more interestingly, there is Granger causation going from FX rate change to intervention (p=0.03), which leads to the conjecture that CNB interventions are based on observed developments in FX markets and are not pre-emptive (Mohanty and Berger 2013). A cross-correlation test of FX rates and FX interventions has shown a disappearance of the effect after two quarters. 24 We test if the FX rate is changed in the right direction. In Fig. 7 a correct slope of the regression can be seen, but there are too many shots missing their mark. Nevertheless, the kuna/euro FX rate is stable in the long run and without a statistically significant trend, showing a successful FX intermediate policy goal. 6,1 Johansen co-integrations tests between FX rate 5,9 (stationary), and combinations of M1, M4, and CPI (nonstationary) give no cointegrating vectors, re5,7 quiring the use of a VAR instead of a VEC model. We tested the VAR Impulse Response Functions of relative 5,5 R² = 0,901 changes in the FX rate, M1, M4, and CPI to test for the 5,3 inflation pass-through. Figure 8 shows the impulse response functions of a dependent variable to a shock 5,1 of a 1 change in the independent variable. All vari4,9 ables are differenced for stationarity and normalized. The time lag is shown in months. 4,7 Fig. 8 shows the expected response result from in4,5 flation to a 1 shock in the FX rate, and an expected 4,7 4,7 75 4,8 4,6 4,65 response result from inflation to a 14,85 shock4,9 the in ln of monetary ag aggregate M4 change of M4. The response from the M4 to a 1 shock in the change of the FX rate is also as expected. It Fig gure 6b: first, rrelation bet tween but a overshoots atCor then backlashes,M4 and CPI. then normalizes after four months. It may be concluded that the Dat source: CBD 2014from the 20 rate, via ta transmission channel goes 4, CNB FX14, own the M4 monetary aggregate to inflation. The response of calc the culation.in the FX rate to a 1 shock in the rate of change inflation representing the central bank intervention is ln of CPI y = -0.0 0002x + 0.0111 R² = 0.292 -1.500 -1.000 0 -500 0 500 1.0 000 Intervention amount in milli n ions of kuna 1.500 2 2.000 2.500 3.000 Figure 7: Euro/kuna FX rate changes as a consequence of FX interventions. Data source: CNB 2014, own representation. 7: anges as a co onsequence of FX interventions. Dat source: CN 2014, o ta NB Figure 7 Euro/kuna FX rate cha own repr resentation. Response of D(CPI) to D(FXRATE) 3% 2% 1% 0% -1% -2% -3% 2 4 6 8 10 12 3% 2% 1% 0% -1% -2% -3% 4% 2% 0% -2% -4% Response of D(M4) to D(FXRATE) 4% 2% 0% -2% -4% Response of D(CPI) to D(M4) 4% 2% 0% -2% -4% 4% 2% 0% -2% -4% 2% 1% 0% -1% -2% -3% Response of D(FXRATE) to D(CPI) 2% 1% 0% -1% -2% 2 4 6 8 10 12 -3% Figure 8: Impulse Response Functions of changes in FX, M1, M4, and CPI Data source: CNB 2014, calculation: E-Views 7.2. Fig. 8: Impulse Response Functions of changes lagged one month behind the inflation rate change, calculation: E-Views 7.2. confirming our hypothesis that the CNB is targeting the monthly changes in the inflation rate, i.e. a moving target. A currency peg gives a stable benchmark for the inflation rate. The surplus inflation over and above the eurozone inflation rate is then attributable to changes in relative prices. If the productivity growth is primarily in the internationally traded goods sector, the prices of tradables have to fall relative to the prices of nontradables. According to the Balassa-Samuelson effect, with fixed exchange rates the prices of non-tradables will have to rise. The CNB classifies its newly officially proclaimed FX policy as a managed float (CNB 2014). Nevertheless, the CNB acknowledges its commitment to pursue the multiple (heterodox) target policy (exchange rate and inflation targeting) as the best way to preserve low inflation rates. It is the CNB's constitutional obligation to pursue primarily the price stability target. Only once this goal has been fulfilled may the CNB pursue additional objectives. According to the CNB, its interventions on the exchange market accrue for no more than 3% of the total transactions volume. The largest amount has been used in the operations of selling kuna for euros in an attempt to prevent an appreciation of the kuna. Only recently has the CNB intervened in FX, M1, M4, and CPI Data source: CNB 2014, several times to buy the kuna in an attempt to prevent depreciation of the kuna (CNB 2013, CNB 2014). Depreciations feed inflation expectations and risk premiums. A higher risk premium reduces the real rate of return on capital. This is detrimental to investment and growth. Croatia would be better off if it could eliminate this negative and partially self-sustaining effect. But even without high exchange rate passthrough, a fixed exchange rate could be an optimal monetary policy in an environment of region-specific shocks (Devereux 2003). Countries with difficulties in controlling their money supply benefit from a currency board/monetary union because the money supply is the primary Granger cause of inflation. This is also the Croatian case where Granger causality test results show best values between growth of M1 and inflation. Nevertheless, confidence in the domestic currency may depend on factors other than the exchange rate regime such as central bank independence (Cukierman, Miller, and Neyapti 2002), the degree of the openness of the economy and the level of debt and its sustainability because of its inherent incentive to monetize (Romer 1993, Romer 1998, Terra 1998). These are some of the issues worth future research. CONCLUSION The goal of the paper was to test three statements made by the Croatian National Bank regarding its long term FX policy commitments, which are also often used by a number of other European central banks. The results of our econometric analysis show that the monthly weighted average kuna/euro FX rate is trend stationary, i.e. without a statistically significant trend, normally distributed and with homoscedastic residuals after modelling. To test the three hypotheses proposed by the CNB, it was also necessary to use the quarterly differentiated data of FI, FX, M1, M4, and CPI. To test the effectiveness of interventions, nominal FX and intervention data were tested for Granger causality. Following the results of time series stability analysis, the statement of FX rate stability cannot be rejected. It seems that the CNB indirectly pursues the goal set in the statement about not determining a priori boundaries or intervention points by directly pursuing the goal set in the statement about maintaining the stability of kuna/euro FX rate changes. The upper and lower intervention points may be generalized as FX rates' first differences. Since the CNB defends no nominal FX rate, but maintains its rate of change at ± 1.9% per month, the second hypothesis of no upper and lower bounds of nominal FX rate returns can be rejected. A Granger causality test confirms the bidirectional causation effects between FX rate and interventions. The FX rate change is an important trigger of intervention. The results show that for Croatia kuna/euro FX rate stability is not a direct mechanism of price stability. The Granger causality test shows that the CNB cannot directly target inflation via FX rate. Johansen co-integration tests do not show any expected cointegrating vectors between the analyzed variables, and the cross-correlation tests do not show any significant direct pass-through effects between the FX rate and inflation as stated by the CNB. We find that there is not enough statistical evidence for the confirmation of at least one out of three statements put forward by the CNB and which are usually considered absolutely correct and often serve as a starting point for fiscal/monetary policy decision-making. Nevertheless, we find a good political reason to change the official policy from free float to managed float: prospective EMU accession. Our FX rate model shows that the kuna/euro FX rate follows a very narrow currency peg of 7.44 kuna/euro with a coefficient of variation at 1.9%, so even this policy change may be assessed as too moderate, as it clearly is a functional currency peg or a quasi-currency board as it is sometimes also called. As already stated, a firmer FX policy commitment toward the euro may be the only policy choice for the future. Small open economies such as Croatia that joined the EU and enjoy full freedom of capital movement are confronted with a classic policy trilemma that becomes the dilemma of whether to fix or to float because of the inability of choosing capital controls. It seems the CNB has anchored, although without a strong commitment. The recent statement change is probably just one more step towards the EMU.

Journal

South East European Journal of Economics and Businessde Gruyter

Published: Apr 1, 2015

References