6.1. Ex ante credibility of the exchange rate band
We use the simple method presented in Svensson (1990) to test for the ex ante credibility of the exchange rate band operated by the National Bank of Hungary from May 4, 2001 – or, more precisely, from October 1, 2001, the date of abolition of the crawling peg – till February 25, 2008. This simple test of credibility uses the concept of uncovered interest rate parity to construct artificial rate of return bands to test whether actual interest rates were within the band in the observed period. To calculate these rate-of-return bands, we first need to decide which maturities of the market interest rates we use to build the rate of return bands. We base our decision on the forecasting ability of the money et, that is, how far ahead can the Hungarian money es or, in other words, when the forward interest rates lose their predictive power.
Therefore, in figure 12, we first present the dot charts that show the relationship between the forward mark
market predict interest rat
month interest rates21 and the actually realized 3-month Bubor interest rates.
Figure 12 Expected and actual 3-month Bubor interest rates
Expected (FRA1×4) vs. actual interest rates
y = 0,9368x + 0,0044 R2 = 0,8183
0%
2%
4%
6%
8%
10%
12%
14%
h later
0% 5% 10% 15%
Predicted 3-month rate
Actual 3-month rate (1 mont)
Expected (FRA2×5) vs. actual interest rates
y = 0,8778x + 0,0095 R2 = 0,5823
0%
2%
4%
6%
8%
10%
12%
14%
er
0% 5% 10% 15%
Predicted 3-month rate
Actual 3-month rate (2 months lat)
21 The forward time series of 1×4, 2×5, 3×6 and 12×15 forward rate agreements (FRA) from April 12, 2005 to April 2, 2009 was kindly provided by MNB.
Expected (FRA3×6) vs. actual interest rates
y = 0,8438x + 0,0145 R2 = 0,5637
0%
2%
4%
6%
8%
10%
12%
14%
0% 5% 10% 15%
Predicted 3-month rate
Actual 3-month rate (3 months later)
Expected (FRA12×15) vs. actual interest rates
y = -0,0629x + 0,0878 R2 = 0,002
0%
2%
4%
6%
8%
10%
12%
14%
0% 5% 10% 15%
Predicted 3-month rate
Ac3-month rate (1 year later)tual
Source: MNB.
These charts tell us that the Hungarian money market loses its predictive power at the maturity of 12 months, which means that this should be the longes maturity for which to construct the rate-of-return bands for the simple tests of credibility of the exchange rate band. In figure 13, we present the charts of these constructed rate-of-return bands up to the maturity of 12 months together with the movement of the respective actual Bubor and Euribor interest rates. Right next to each rate-of-return band chart, we inserted the chart of the annualized rates of forint depreciation based on the interest rate discrepancies between the respective Bubor and Euribor interest rates.
t
Figure 13 Rate-of-return bands and expected forint depreciation
Rate of return band and actual interest rate (3m)
-50%
Annualized rates of HUF depreciation (3m)
-60%
lower bound of depr. (3m) upper boud of depr. (3m)
Rate of return band and actual interest rate (6m)
-30%
Annualized rates of forint depreciation (6m)
-30%
lower bound of depr. (6m) upper boud of depr. (6m)
Rate of return band and actual interest rate (12m)
-20%
Annualized rates of forint depreciation (12m)
-20%
Source: own calculations based on data from and www.mnb.hu.
ths, based on the strong bound of the official exchange rate band – and the expected annualized rate of depreciation – +3.56%, 3.38% and 3.13%, respectively, based on the interest rate differential of Bubor and Euribor – was relatively small. Thus, on this day, the actual Bubor rate was very close to the lower bound of the rate-of-return band. Another day when this gap was relatively small – 2.98%, 3.5% and 4.08%, respectively – was Apparently, the credibility of the exchange rate band as of Svensson (1990) was intact throughout the period. However, there were periods when the sustainability of the exchange rate band was tested. On January 16, 2003, when the EURHUF exchange rate tested the strong bound, the gap between the lower bound for depreciation – -0.01%, 0.00% and 0.00%, respectively, for 3, 6 and 12 mon
March 9, 2005 main within the band throughout the observed period.
Figure 14 EURHUF exchange rate expectations within the official floatation band . However, as figure 14 testifies, the exchange rate was at all times expected to re
240 250 260 270 280 290 300 310 320 330
Parity
Expected EURUHF in 3 months Expected EURHUF in 6 months Expected EURHUF in 12 months Actual EURHUF
230
Jul-01 Jan-02
Jul-02 Jan-03
Jul-03 Jan-04
Jul-04 Jan-05
Jul-05 Jan-06
Jul-06 Jan-07
Jul-07 Jan-08 Source: own calculations based on data from www.mnb.hu and www.euribor.org.
6.2. Ex ante credibility of the IT regime 6.2.1. Credibility as of Svensson (2009)
In this subsection, we present two charts, which show the evolution of the central bank’s inflation forecasts and the market’s inflation expectations compared with the actual official inflation target. First, in figure 15, we show the gap between MNB’s CPI forecasts and the respective CPI targets. Then, in figure 16, we illustrate the same gap between the market’s CPI forecasts and the respective CPI targets.
Figure 15 The gap between MNB’s CPI forecast and the respective CPI target22
Source: own calculations based on data from MNB’s Inflation Reports available at www.mnb.hu.
Figure 16 The gap between the market’s CPI forecast and the respective CPI target23
-1%
Source: own calculations based on data from MNB’ Inflation Reports available at www.mnb.hu.
22
represents th GAP1 represents the difference between MNB’s CPI forecast and the respective CPI target that is closer in time, while GAP2 e gap between MNB’ forecast and the respective CPI target that is further away in time. Note that, in August 2005, AP2 discontinues in Market expectations are discerned from analyst CPI surveys by Reuters as they appeared in MNB’s Inflation reports on www.mnb.hu.
MNB announced the switch to a medium term CPI target of 3% as of 2007, which is why the curve of the G August 2005.
23
The c B’s forecasts nor market expectations were anchored to the official CPI
targ sts and target very rarely came under 1%. This portrays the
inef t enting market expectations. This is, however, – as suggested by historical and narrative evidence – more likely to have been a result of the lack of coordination between the two a ghting inflation and not the low level of credibility of the central bank self. In fact, in another chart in figure 17, we show that the gap between MNB forecasts and the inflation target and that between market expectations and the inflation target moved very closely in the observed
period ts out,
“the degree of correspondence between inflation expectations and the central bank’s inflation forecasts then becomes a measure of how credible the central bank’s inflation forecasts and analyses are” (Svensson, 2009, p. 16.). So, in short, the picture seems controversial: the official CPI target had little to no effect on the markets CPI expectations, while market expectations were close to MNB’s CPI forecasts.
Figure 17 The correspondence between the market’s and MNB’s CPI forecasts as compared to the target se harts suggest that neither MN
et, and the gap between foreca fec iveness of the CPI target itself, in ori
br nches of economic policy in fi it
– their correlation coefficient being 0.8824. This is important because, as Svensson (2009) poin
-2%
-1%
0%
1%
2%
3%
4%
Jan-01 Jul-01
Jan-02 Jul-02
Jan-03 Jul-03
Jan-04 Jul-04
Jan-05 Jul-05
Jan-06 Jul-06
Jan-07 Jul-07 GAP1 (Market-target)
GAP1 (MNB-target)
Source: own calculations based on data from MNB’s Inflation Reports available at www.mnb.hu.
24 The correlation coefficient of MNB’s and the market’s forecasts – not that of the gap between the forecasts and the target – is 0.76, somewhat lower.
.2.2. Credibility measure of Laxton and N’Diaye (2002)
The credibility measure for Hungarian monetary policy applied by Laxton and N’Diaye (2002) is presented e same measure for Germany. For Hungary, we use the daily fixing of the 6
below in figure 18 together with th
10-year benchmark government bond yield retrieved from www.akk.hu, while, for Germany, we use 10-year Bund yields calculated from bond prices fixed every day at 16:00 on Bloomberg, kindly provided by MNB.
Both for Hungary and Germany, we used 5% as the minimum level of government bond yields, but having tested other alternatives we can add that the choice of this value does not alter the end result.
As opposed to the German curve, which is mostly stable and above 50% – evidence of a relatively high and lity –, the Hungarian curve is that of a very volatile, non-mean-reverting variable with
November 2006 – January 2008,
• November 2003 – November 2004,
• July 2006 – October 2006,
• February 2008 – April 2009.
stable level of credibi
relatively big upswings and downturns with a range of almost 100%. We ran an augmented Dickey-Fuller test for the Hungarian curve with the null hypothesis of the existence of a unit root. The test, the results of which are presented in Appendix 4, shows that the null hypothesis cannot be rejected, thus, the time series of the variable is non-stationary. On the other hand, the German curve is visibly that of a stationary time series. The Hungarian curve starts with a value of 21.1% on July 13, 2001 and terminates with a value of 3%
on April 30, 2009, reaching its trough, 0%, on December 2, 2003, and its peak, 99.2%, on September 2, 2005. Periods of relatively high credibility (>50%) based on interest rates are the followings:
• November 2001 – May 2002,
• October 2002 – October 2003,
• December 2004 – June 2006,
•
while the rest of the observed period is characterized by relatively low credibility (<50%):
• July 2001 – October 2001,
• June 2002 – September 2002,
Figure 18 Credibility of monetary policy as discerned from the level of domestic interest rates
0%
80%
90%
100%
10%
20%
30%
40%
50%
60%
01 01 02 02 03 03 04 04 04
70%
Jul -De
c-May - Oc
t-Mar
-Aug- Jan- Ju n-No
v-Apr-05 Sep-05
Feb-06 Jul-06
Dec-06 May-07
Oct-07 Mar-08
Aug-08 Jan-09 Credibility_HUN
Credibility_GER
Source: own calculations based on data from www.akk.hu and Bloomberg.
After the simple mption that
credibilit ving characteristi
along with the Müller-Petalas-method of parameter time pa map the evolutio
y of the Hungarian IT regime of Goldberg and Klein
e s t price and inflation ].
ependent variables ( from our analysis, identically with the method ein (2005). In the ification, we use as dependent variable the d between the benchmark yields of the ten-year government bond and the
three-by the Government Debt Management Agency (ÁKK) of Hungary. We decided to
s for the observed term spread th ause
the meaning of short and long term bonds
nchmark y preceding the announcement of the inflation data to its value registered at 14:15 on the day of the inflation announcement26. We observe two periods using these 7. Dynamic tests of credibility
methods, we implement a more advanced test, which is built on the assu y is a dynamically evol c of monetary policy. We appl
th estimation to
y the Elliott-Müller test n of the market perception of credibilit . We follow the specifications
(2005) and w use Hungarian a se data in [2 7.1. The variables
We use two d qt+ −qt− [1]) in
implemented in Goldberg and Kl first spec change in the term sprea
month treasury bill issued
use shorter term an the terms used by Goldberg and Klein (2005) bec was different in Hungary in the observed period25. The change in the term spread is, thus, calculated as the change in the difference of the 10-year and 3-month be
yields from its value registered at 14:15 on the da
dependent variables: one runs from July 2001 to February 2008, the last month when the currency exchange rate floatation band was intact, and the other runs from July 2001 till April 2009, the end of our entire observed period.
In the other specification, we use as dependent variable the high-frequency change in the euro-forint exchange rate (EURHUF). In this case, qt+ −qt−becomes the change in the log value of the foreign exchange rate from its value registered half an hour before the publication of the monthly inflation data to its value registered half an hour after the data release. The change is negative when the forint appreciates
25 Goldberg and Klein (2005) used the yield on the two-year government paper for the short term, which would have been considered medium term in Hungary.
fixing of bond yields because we found these the most suitable and relevant for our analysis.
Considering the relatively low turnover and volatility of the market for Hungarian fixed income assets compared to the core US and euro zone markets analyzed by Goldberg and Klein (2005), this fact should not deter results significantly. The dates of the inflation data releases did not coincide with any other major economic news releases or MC/Governor comments so the effect on the slope of the curve can always evidently be connected with the CPI announcement.
26 We use the daily benchmark
aga i s).
The foreign exchange rate data we use are the transaction data registered in Reuters spot matching trading system such that we determine the closing price for a given period as the price at which the last deal of that period was struck.27 Exchange rate data were only provided until February 2008, which is why the observed period in this specification runs from July 2001 to February 2008 (80 observations).
The independent variable ( ) is the same in the two specifications and also identical with what is used in Goldberg and Klein (2005), that is, the surprise component of the monthly inflation announcement.
We approximate this surprise as the difference between the consensus inflation estimate of Hungary-based macro analysts surveyed by the online journal portfolio.hu28 and the actual Hungarian monthly headline consumer price index (CPI) in percentage points, published by KSH. The difference is then normalized by its standard deviation. As it is apparent from Appendix 3, one is unable to reject the unbiased nature of the CPI survey of portfolio.hu as a good predictor of the actual inflation data. The inflation data are usually ion schedule, while the consensus is usua
In the following section, we perform the three above mentioned Elliot-Müller tests for time variation in the slope and, jointly, in the slope and intercept of the specifications described above, then we implement the Müller-Petalas procedure for the dependent variables if we find some evidence of time variation. In the Elliot-Müller test, a value of the qLL statistic smaller – more negative – than its critical value signals a change in the market perception of MNB’s policy stance over time in the observed period. Both for performing the Elliot-Müller test and for calculating the estimated time paths using the method of Müller and Petalas, we used a Stata program which was kindly made available to us by the authors of Goldberg and Klein (2005). As suggested by Müller and Petalas (2009), we used c = 10 throughout the calculations.
7.2. Results of the Elliot-Müller tests
In table 13, we present the results of the Elliot-Müller tests we performed in the indicated periods. The the longer period suggest a inst the euro (log of EURHUF decreases) and pos tive when it depreciates (log of EURHUF increase
+
−
+ − t t
t E x
x
published at 9:00 on the day previously announced in KSH’s publicat lly one day before the announcement by KSH.
results of the test using the term spread specification both in the shorter and in
conclusion to reject the null hypothesis of the stability of the regression coefficient. This result means that some persistent time variation is prevalent in the reaction of the term spread to the CPI releases, meaning
27 These high frequency data were kindly made available to us by Reuters Hungary Ltd.
28 We used the Reuters consensus before November 2004 – the publication of the first portfolio.hu estimate –, but we opted for portfolio.hu’s consensus estimates afterwards, as they were more reliable than those of Reuters.
that the bond market’s perception of MNB’s credibility was unstable throughout the period of July 2001 – February 2008. Such a conclusion cannot be made in case of the EURHUF exchange rate specification, where the null hypothesis of parameter stability in time cannot be rejected at the 90% significance level.
Table 13 The results of Elliot-Müller tests for persistent time variation
Test of time variation of the change in the term spread the change in EURHUF
Slope (γi) only -7.3949* -5.473
Joint slope (γi) and intercept (α) -12.448 -11.398
Number of observations 78 80
(July 2001 – February 2008)
Slope (γi) only -4.7767
Joint slope (γi) and intercept (α) -12.920*
Number of observations (July 2001 – April 2009)
91
-
Critical values Slope: -11.05*** (1%), -8.36** (5%), -7.14* (10%) Slope and intercept: -17.57*** (1%), -14.32** (5%), -12.8* (10%) Source: own calculations from data by www.akk.hu , www.ksh.hu , www.portfolio.hu and Reuters Hungary.
7.3. Results of the Müller-Petalas procedure
We now perform the Müller-Petalas procedure to illustrate the evolution in time of the credibility of the central bank as perceived by the bond market. In figure 19, we present the smoothed estimated parameter path of γiusing the Müller-Petalas procedure for the best specification based on the data set of July 2001 to February 2008, where the change in the term spread is the dependent variable. We back the smoothed estimated parameter path by the time-path of the actual base rate and the annual rate of headline CPI flation in the observed period so that one can put the parameter values into historical context. For the sake in
of curiosity, we also visualize in figure 20 the curve of the estimated parameter path of the specification where the change of the log value of the high-frequency EURHUF exchange rate is the dependent variable.
Although the results from the Elliot-Müller test of the later specification were not significant, it is interesting to see whether this estimated EURHUF parameter time path shows a similar picture to that shown by the term spread specification.
Figure 19 Estimated smoothed parameter path of γi of the term spread change and the MNB base rate Y-o-y CPI inflation (left scale)
MNB base rate (left scale)
Term spread coefficient (right scale)
2%
Source: own calculations based on data from www.akk.hu and www.mnb.hu.
Figure 20 Estimated smoothed parameter path of γi of the EURHUF change and the MNB base rate
6%
Y-o-y CPI inflation (left scale) MNB base rate (left scale) EURHUF coefficient (right scale)
Source: own calculations based on data from Reuters Hungary Ltd. and www.mnb.hu.
Apparently, the estimated time paths of both γi parameters shows the same declining trend in the observed period, and even the timing of the upswings and downturns of the two curves appears to be very similar, almost identical if one focuses on the big picture. In a short first phase, an upward trend is visible on both parameter paths, followed by a long-going period of decline – both curves turn downwards on July 11, 2002
ward spiral breaks this upward trend on July 11, 2008.
part from the cycles in the curves, if one looks at the values of the parameters at the beginning and the end
ne these same specifications for the euro zone and the Unites States in the aftermath of the establishment of the European Central Bank in 1999. They conclude that the market perception of ECB was evolving after its inception in a way that reflected a gradual improvement in the market’s perception of ECB’s anti-inflation stance, while, on the other hand, the market’s perception of the Federal Reserve’s reputation as an inflation-fighter remained relatively stable in the very same period. Goldberg and Klein (2005) argue that the evolution of perceived credibility should come as no surprise in the case of a newly established central bank, such as the ECB was in 1999, whereas one would expect perceived credibility to remain stable at the same time in countries with relatively older and more established central banks, such as the Federal Reserve in the US.
Applying the above logic to Hungary and the MNB, whose mandate to fight inflation was granted in July
Applying the above logic to Hungary and the MNB, whose mandate to fight inflation was granted in July