The role of ECB communication in guiding markets

In a survey conducted by Blinder et al. (2017), more than half of the responding central bank governors say that new communication measures have been adopted at their institution since the crisis. Moreover, the overwhelming majority (more than 80%) of central bankers and academic economists (more than 90%) stated that the role of central bank communication has intensified. Both the majority of central bankers (60%) and academic economists (75%) expect communication policy changes to remain in place, or to go even further (Blinder et al. 2017).

As stated by Hartmann and Smets (2018, p. 4), “the need for additional communication in a complex (non-standard) policy environment rose and forward guidance became an essential tool for easing policy in a low interest rate context”.

See, for instance, Jones (2014). For a critical view on unconventional policies, see Borio and Gambacorta (2017).

For a comprehensive treatment see, Hartmann and Smets (2018).

Regular longer-term refinancing operations (LTRO) have maturities of three months. The VLTROS had a maturity of 36 months and were conducted as fixed rate full allotment procedures. They were accompanied by a reduction in the required reserve ratio from 2 to 1% and an increase collateral availability by broadening the definition of eligible assets.

The idea dates back to Black (1995), who argued that when interest rates reach the zero lower bound, people would prefer to hold cash rather than financial instruments generating negative interest.

The data produced by Leo Krippner can be accessed through the website of the Reserve Bank of New Zealand: https://​www.​rbnz.​govt.​nz/​research-and-publications/​research-programme/​additional-research/​measures-of-the-stance-of-united-states-monetary-policy/​comparison-of-international-monetary-policy-measures. In order to cope with the non-linearity introduced by the option effect, Krippner applies an Iterated Extended Kalman Filter for the estimation. He uses German and French zero government bond rates and overnight indexed swap rates (once they become available for the euro area in 2008) with maturities of 0.25, 0.5, 1, 2, 3, 5, 10 and 30 years to approximate the euro yield curve. For comparison, Krippner sets the ELB to 12.5 basis points across all estimated currencies. McCoy and Clemens (2017) analyze the effect of the choice of the effective lower bound in Krippner’s shadow rate model estimation and give an illustrative example. They point out that, the higher the ELB is set, the more pronounced is the option effect, implying that the shadow yield curve (and, thus, the shadow rate) is shifted downwards. Hence, a trade-off exists between a meaningful monetary policy stance indicator and an economically relevant policy rate estimate. In our analysis, we use the shadow rate as a proxy for the stance of monetary policy and, therefore, are not concerned with the relatively high lower bound of 12.5 basis points.

The weight for the forecast of the current year is \( m/12 \) and the weight for the forecast of the following year is \( \left( {12 - m} \right)/12 \), where \( m \) denotes the number of remaining months in the current year. As the survey usually is published in the first half of the month, we assign the current month to the remaining months of the year.

Using nominal GDP weighted averages from the euro area member states to construct expectations that go back to January 1999 does not change the conclusions. However, as money growth expectations are not available at the member state level, we would lose information by going back that far.

The database can be accessed through https://​sdw.​ecb.​europa.​eu/​browse.​do?​node=​9689716. The first vintages of the for-us relevant variables published in that real-time database date from January 2001.

The benchmark revisions took place early March 2002, June 2002, March 2003, May 2005, March 2011 and March 2016.

The way in which our indicators are constructed very much follows the procedure underlying the so-called KOF Monetary Policy Communicator (MPC), as published by the KOF Swiss Economic Institute and used, e.g., by Sturm and de Haan (2011), Conrad and Lamla (2010), Lamla and Sturm (2013), Bulíř et al. (2013) and Neuenkirch (2013). The key difference is that the MPC is a leading indicator for monetary policy and therefore considers only forward-looking statements regarding prices.

While most of the data we use are proprietary and cannot be shared publicly, we are happy to provide our communication data upon request.

The Euribor 3-month futures contracts have expiry dates on the 3rd Wednesdays of each of the following six months and at quarterly frequencies for longer horizons. Each contract can be traded until two days before the expiry date, meaning we do not have a natural series of future prices. We use an end-of-day series that represents the future contract that expires next and can still be traded. That series may contain price jumps once a month, exactly two days prior to the expiry dates. Owing to the timing of the ECB Governing Council meeting days, those jumps do not affect our market surprise measure.

We have also experimented with the US dollar-euro exchange rate. The conclusions are not affected by that alternative.

The change in the shadow rate and the change in the main refinancing rate are not correlated in this table because it takes up to 6 days for the main refinancing rate effectively to be changed after having been announced publicly by the ECB. By that time, the shadow rate already has adapted.

The effective change in the MRR usually takes place six days after it was announced publicly. Since financial markets incorporate the announced change in the MRR immediately, we use the decided change in MRR in the subsequent analyses.

Including each of these communication variables separately does not change the conclusion: only communication on the economy is significant in explaining the change in the shadow rate. Given the low correlation between those indicators, that was to be expected (see Table 2).

All results in Table 3 are based upon those days in which a Governing Council meeting has taken place and a monetary policy press release has been published. If we move to a daily frequency, increasing the size of the sample to 5724 observations, the significance of the variables increases, while no significant changes in signs are recorded.

A general-to-specific methodology means dropping insignificant variables one at a time until all variables remaining in the model are significant.

We have furthermore carried out a placebo test in which the dependent variable has been moved to another arbitrary date and find that in general all variables become insignificant.

Standardized beta coefficients are shown in Table A.1 in the online appendix.

Hence, we use phase 2 as our baseline. As noted, phase 1, owing to data limitations, is not included in our sample. Similar analyses have been carried out for the other columns in Table 3. The conclusions are not affected.

This time, however, we are working with what boils down to a monthly frequency that did not change over time.

An illustration of the timing of the model can be found in Figure A.2 in the online appendix.

Note that it did not make much sense to include this variable in our first model when we looked at the changes between the day before and the day after the Governing Council meeting. In such a short time frame, including it always resulted in insignificant coefficient estimates.

Since the Great Financial Crisis, the variable takes on non-zero values. Hence, it covers only the last two monetary policy phases of our sample, i.e., 2008m9-2014m5 and 2014m6-2018m12.

The correlation between the residuals of the inflation and GDP growth equations is the highest and equals 0.27.

Table A.2 in the online appendix reports standardized coefficients of the results presented in Table 7, allowing for a direct comparison of the quantitative importances of each of the explanatory variables.