Changes in euro area monetary transmission?

Abstract

Empirical evidence on whether euro area monetary transmission has changed is, at best, mixed. We argue that this inconclusiveness is likely to be due to the fact that existing empirical studies concentrate on the effects of particular developments on specific transmission channels. Such analyses typically require strong assumptions. Moreover, specific changes could have off-setting effects regarding the overall effectiveness of monetary policy. In order to shed light on this issue, we investigate whether there has been a significant change in the overall transmission of monetary policy to inflation and output by estimating a standard Vector Autoregression (VAR) for the euro area and by endogenously searching for possible break dates. We find a significant break point around 1996 and some evidence for a second one around 1999. We compare the effects of monetary policy shocks for these episodes and find that the well-known ‘stylized facts’ of monetary policy transmission remain valid. Therefore, we argue that the general guiding principles of the Eurosystem monetary policy remain adequate. Moreover, it seems that monetary transmission after 1998 is not very different from before 1996, but probably very different compared to the interim period.

Acknowledgements

We appreciate helpful comments and suggestions by Jörg Breitung, Christina Gerberding, Stefan Gerlach, Felix Hammermann, Wolfgang Lemke, Manfred J. M. Neumann and Ralph Setzer. We benefited from insightful discussions on an earlier version of this article at the 7th BIS Annual Conference, in particular we are grateful to Arminio Fraga Neto, Ben Friedman, Marvin Goodfriend, Charles Goodhart, Otmar Issing, Mervyn King, Ulrich Kohli, Christopher Sims, Christian Upper, Ignazio Visco and Michael Woodford. We thank seminar participants at the ROME workshop in Frankfurt, at the Erich Schneider seminar in Kiel and at the conference on ‘Monetary policy transmission mechanism in the euro area in its first 10 years’ at the European Central Bank in Frankfurt, especially our discussant Stefano Siviero.

Notes

¹ This list is probably not exhaustive as there might be other channels at work. Some of these, for example the risk-taking channel of monetary policy (Borio and Zhu, 2008) have up to now not been examined in greater detail. For a more detailed literature review, see Cournède et al. (2008) or Weber et al. (2009).

² For instance, in the case of Germany, the apparent weakness of the bank lending channel can be traced back to the institutional structure of the German banking system and the long-term relationships between banks and customers which tend to entail an implicit insurance of the credit customer against adverse shocks, such as a restrictive monetary policy (Ehrmann and Worms, 2004). If financial development increases competition between banks but also between banks and other financial market segments, then this so-called ‘housebank principle’ could lose importance, thereby ceteris paribus strengthening the bank lending channel.

³ Note, Sbordone (2007) and Woodford (2007) argue that such changes are not likely to be large.

⁴ We disregard the Akaike Criterion (AC) as it asymptotically overestimates the order with some probability, whereas HQ estimates the order consistently and SC is even strongly consistent (see, for e.g. Lütkepohl, 2005). For the sub-samples we use the same lag order.

⁵ Note our sample ends in 2006:04. See the Appendix for a description of the data. As most time series included in the VAR exhibits a stochastic trend, we do not include a deterministic trend. A possible linear trend is picked up in the vector of constants.

⁶ Specifically, we fitted a local quadratic polynomial for each observation of the original series and used this function to estimate the values needed for a quarterly series. The quadratic polynomial is formed by taking sets of three adjacent points from the original series and fitting a quadratic, so that the average of the quarterly frequency points matches the original data. This method is especially well-suited in the case in which relatively few data points are interpolated and the source data is fairly smooth. The conversion was done in EViews 6.0.

⁷ Yet, including real housing wealth improves the overall fit of the model and proves to be an important explanatory variable for the euro area in other instances as well (see, e.g. the money demand analysis of Greiber and Setzer, 2007).

⁸ As noted by Candelon and Lütkepohl (2001) it turns out that in samples of common size the χ ² and F approximations of the actual distributions may be poor even if a single break point is tested. The actual rejection probabilities may be much larger than the desired type I error. For completeness, we also apply a system 1-step Chow test as implemented in Doornik and Hendry (2007). This test indicates parameter instability for the mid-1990s, although somewhat earlier as indicated by the bootstrap versions.

⁹ We used Anders Warne's program Structural VAR 0.40.

¹⁰ Note, to the extent that the true model of the economy is appropriately described by a linear model, potentially omitted variables do not generate spurious instability. A possible omission might bias the parameter estimates of the systematic component, but would not imply structural changes across samples (see also Boivin and Giannoni, 2002).

¹¹ This interpretation, notably a break around the mid-1990s is consistent with the results in Breitung and Eickmeier (2009).

¹² The figure shows the effects of a one SD monetary policy shock. If we assume a tightening of 100 bp, GDP falls about 0.5% after six quarters in the first sample and about 0.9% after six quarters in the second one. GDP deflator falls about 0.2% and 0.4% after 4 years, respectively.

¹³ For the sake of comparability, we impose in Fig. 7 the monetary shock to be of the same size in both sub-samples. Technically, the impulse responses have been obtained by estimating an extended version of the baseline VAR that includes a vector of dummy variables. These dummy variables take the value zero for the period 1980:01 to 1996:01 and the value one for the period 1996:02 to 2006:04. In addition, we applied a ‘general-to-specific’ procedure to get a more parsimonious model. However, in comparison to an unrestricted version of the baseline model this procedure does not seem to have a notable effect on the IRFs.

¹⁴ In Weber et al. (2009), we show that these differences are indeed statistically different from zero.

¹⁵ There is some evidence that the mean of consumer price inflation has changed in the mid-1980s (Altissimo et al., 2006). Note that our results do not depend on the aggregation method used for the euro area data. Specifically, if we construct euro area GDP and the euro area GDP deflator with flexible exchange rates does not have a notable effect.

¹⁶ Yet, the use of factor-augmented VARs is not without problems as it might produce ‘spurious results’, see Uhlig (2009).

¹⁷ Such a break point is reasonable from an economic point of view. For instance, in the euro area, the exchange rate became unavailable as a monetary policy tool already around 1995. If one takes it for granted that we have witnessed at least one such notable change it is not too surprising that our estimates for the whole sample do not appear fully satisfactory.

¹⁸ The documented change in the relative strength of monetary policy is unlikely to have happened at a specific date. For instance, as regards EMU the fact that monetary regimes in Europe would change on 1 January 1999 was well-known before. Hence, agents likely started to prepare quite some time before that event. Further, it is reasonable to assume that these preparations for adjustment were stretched over several years.

¹⁹ We get very similar results if we estimate our benchmark VAR for the first and third periods, respectively, and assume that the monetary policy shock in both periods is of equal size. The identification of the structural monetary policy shock in Equation 5 is therefore empirically correct.

²⁰ The bootstrap procedure implicitly assumes that the SDs of the VAR residuals do not change over time.

²¹ Note, under the null hypothesis of parameter constancy, the respective impulse response functions should differ only randomly.

²² This shortcoming is very much related to a general shortcoming of the VAR approach. Specifically, the VAR identifies only the effects of monetary policy shocks, although most variation in the policy rate is due to an endogenous response to the state of the economy. For a critical review of the VAR approach, see Walsh (2010).

²³ The same is true if we start our sample in 1984 and compare 1984–1996 with 1999–2006.

²⁴ In a certain sense, this specific interim period is similar to other periods that have been scrutinized from a monetary policy point of view. Specifically, the interim regime 1996–1998 witnessed similar patterns of disinflation as the time spans after the appointment of Paul Volcker in the US and the period following the decision of Deutsche Bundesbank (2004) to switch to monetary targeting.