Inflation differentials in the Euro area: did the ECB care?

Abstract

Compared to inflation differentials among regions in the United States, European Monetary Union (EMU) inflation differentials are larger and more persistent. Based on augmented monetary policy reactions functions, this article addresses the question whether the presence of pronounced inflation differentials in combination with low average inflation rates has influenced monetary policy decisions of the ECB. The article finds statistical evidence that the ECB took inflation differentials into account which may reflect the fear of deflation in low inflation countries like Germany.

Acknowledgements

We would like to thank the anonymous referee for valuable remarks and suggestions. However, all remaining errors are due to us.

Notes

¹ Alternative measures for the inflation dispersion are the weighted cross-country standard deviation, the maximum span of inflation rates and the coefficient of variation. All these measures, however, display the same qualitative picture. See ECB (2003, Annex 1) for a discussion of possible divergence measures.

² The ECB (2003, p. 46) also expresses the view that there is no comovement in inflation rate levels and its dispersion.

³ An alternative available measure for the US inflation dispersion is the unweighted standard deviation of CPI inflation for 14 metropolitan areas. This measure is somewhat higher than for the four Census Regions (see ECB, 2005, Chart 1). However, since these metropolitan areas are much smaller than most euro area countries, the four Census Regions are probably the more appropriate base for a comparison.

⁴ See also ECB (2003, p. 6).

⁵ See, for example, the annual report of the ‘Sachverständigenrat’ – the German Council of Economic Advisors – for 2001. See additionally ECB (2003, p. 32, Table 5) for a summary of available studies. Those studies, however, consider periods prior to the EMU and tend to overstate to actual contribution of the BSE, since productivity convergence was considerably stronger prior to the EMU. On average, the studies imply that the BSE results in a standard deviation of 0.6.

⁶ For example, the category ‘meat’ has a weight of 3% in the Austrian HICP and 5.5% in the Spanish HICP.

⁷ Employing pass-through regressions Campa and Gonzalez Minguez (2006) find that differences in the degree of transmission of a common exchange rate movement into consumer prices among the euro area countries do exist. For example, they calculate an average pass-through rate to consumer prices for the euro area to be about 0.5 with country-specific rates that range from 0.2 for Italy to 1.5 for Ireland. Furthermore, they demonstrate that most of the differences among member countries are due to the distinct degree of openness rather than to the heterogeneity in the structure of imports. In an applied analysis Cunningham and Haldane (2002) use calibrated simulations to demonstrate the impact of alternative specifications of import-price pass-through on the monetary policy transmission mechanism.

⁸ In a semi-structural modelling approach, Clausen and Hayo (2006) find asymmetries in the effects of output gaps on inflation within EMU member countries. This can also be seen as an indication for the existence of structural differences in the EMU.

⁹ Altavilla and Landolfo (2005) use a Markov-switching approach in order to analyse whether the ECB reacts asymmetrically. The authors do indeed find that the phase of the business cycle is an important matter in the ECB's monetary policy decisions. Furthermore, Fendel and Frenkel (2005) employ augmented monetary policy reaction functions and show that the ECB also takes into account information from the term structure of interest rates.

¹⁰ We find that that monetary policy reaction functions for the ECB are robust against changes in detrending method. This confirms the result of Fendel and Frenkel (2005).

¹¹ Huang and Lin (2006) propose an alternative estimation approach based on a dynamic ordered probit model with time-varying parameters carried out via recent advances in Bayesian simulation approach, namely, the Markov chain Monte Carlo.

¹² More specifically, we use the first six lags as well as the ninth and the twelfth lag of the output gap and the first, the third, the sixth, the ninth and the twelth lag of the short-term interest rate and the inflation gap. This instrument set displays the core set of instruments that is present throughout all subsequent regressions. In the following we only report the instruments that we add to this core set of instruments in particular regressions. The J-statistics which are not reported here indicate that we chose a valid set of instruments for all subsequent regressions.

¹³ The Japanese experience points to adverse effects of deflation so that a ‘fear of deflation’ seems justified. Among others, Morana (2005) argues that monetary policy played a pivotal role for Japanese deflation and that a more expansionary monetary policy stance of the Bank of Japan could have stopped the deflation in Japan.

¹⁴ We also re-estimated the regressions based on formulation (4) and forced the smoothing parameter ρ to be zero. As the R ² values of the estimations (using the three alternative inflation measures) are significantly lower than for the specifications in Table 3, we do not present the results here. However, also in these cases, an improvement in the R ² when employing the modified measures for the inflation gap was also present.

¹⁵ On top of the core set of instruments, we instrument the expected SD of inflation rates and the expected coefficient of variation rates by their first six lags, respectively. For the squared forms of the two divergence measures we add a one before squaring them because they contain values smaller than unity. We also tried the contemporaneous divergence measures in some additional regressions but the effects were not significant in those cases and we, therefore, abstain from reporting them.

¹⁶ We also performed the regressions based on the unweighted standard variation but results turned out to be quite identical. For the sake of brevity, these results are not presented.

¹⁷ Again the first six lags of the divergence measure were employed as additional instruments in regressions (10) and (11).

¹⁸ We also tested more explicitly for the presence of structural breaks over the whole sample period but did not get statistically robust results of a regime shift. The same results were obtained when we examined the specification in the alternative form with interest rate differences on the left-hand side.