Country-specific euro area government bond yield reactions to ECB’s non-standard monetary policy program announcements

Initial events only

The alternative choice of only the 7 initial events of each program tests for the hypothesis of diminishing effects. These pivotal events should induce the strongest effects. This approach assumes repeating announcements do not provide new information. In consequence, investors do not amend their choices because these announcements do not have a surprising character. An advantage of this approach is that every program is weighted equally using its first announcement only.

The only change is a weakly significant and negative coefficient of Spain in the immediate case. Hence, even focusing on the introduction of the programs, no noticeable influence emerges. For the delayed case, the negative impact is only significant for core countries while periphery countries are unaffected by initial announcements (except for Spain). The issue that periphery countries do not react to the announcements is puzzling, especially considering a negative impact on yields for both groups in the panel regressions. One explanation could be that investors need additional confirmation to change their perception on periphery countries. On the initial announcement day they might be still skeptical about future development. After a confirming announcement they trust the policy change and adopt their expectations accordingly. Consequently, the results highlight that repeated announcements do matter.

Excluding (T)LTRO

One might argue the (T)LTROs belong to conventional monetary policy because they are close to the conventional LTROs. Therefore, all regressions are repeated without the announcements denoted with (T)LTRO in Table A1. Hence, 21 events remain and the time span starts from 2009 because no other types of events happen before.

In fact, while dropping these 5 (pure) (T)LTRO events in question the results for the immediate case persist. In contrast, a striking difference can be found for the delayed case for which the yields of Germany, the Netherlands, Ireland and Greece become insignificant. This implies (T)LTRO announcements are crucial for the yield reduction in those economies. Hence, we confidently keep (T)LTRO events in the analysis. We observe that countries of both groups are still affected. The panel analysis is unresponsive to this modification for both same-day and delayed effects.

Including events of technical details

The ECB regularly announces technical details of the asset purchase programs. Investors should not react to these announcements as they do not change the situation on financial markets substantially. To test for this hypothesis, events regarding details of the programs are added. Table A1 lists all relevant 71 events; (T)LTRO is the dominating program.

The results for the immediate case persist. Assuming delayed effects significantly negative estimators exist for most countries (Austria, France, Belgium, Spain, Italy, Ireland, Portugal). This is in line with the previous findings stating little immediate influence but substantial delayed negative effects. The program-specific effects attenuate for core countries when considering the full event set. The significance levels of (T)LTRO decrease and this program becomes even insignificant for Germany and Finland. Similarly, CBPP announcements do not affect French, Austrian and Finnish bonds any more and the SMP looses its impact on the bonds of the Netherlands and Austria. The general picture of the other programs is robust to the inclusion of additional events, though. Most importantly, the effect of daAt and the qualitative difference between core countries and periphery countries persist in the panel, albeit the estimators become smaller when accounting for technical details.

Overall, adding events that provide technical details does not change the main results. Put differently, they are not essential and can be omitted. However, when considering all 71 events, core countries seem to be less affected.

Regime shift

Afonso et al. (2018) detect a new bond-pricing regime for sovereign yields after the OMT announcement. The speech by Mario Draghi on July 26, 2012 might effectively have changed the perception of monetary policy announcements in general. One could split up the data sample into a pre-period and into a post-period with respect to this speech. However, the amount of announcements would reduce accordingly to 9 events before and 17 events after the speech, which impairs the program-specific analysis and the idea of assessing the average impact of announcements. Moreover, comparing two time series with varying observation lengths worsens their comparability. To keep the amount of events stable, an alternative specification adds another dummy taking the value of 1 for the time after the regime shift, and 0 otherwise. The results are robust to the inclusion of this regime dummy implying that announcement effects are not specific to a certain point of time but hold for the whole examination period. The dummy variable itself is significant mostly for periphery countries.

Forward guidance receives increasing attention and is applied by several central bank as an additional tool of their monetary policy. Most importantly, forward guidance is not a non-standard monetary policy itself but a way a central bank commits to such measures. A combination of both FOMC’s forward guidance and large-scale asset purchases continue to move medium- and longer-term interest rates even when short-term rates were stuck at zero for the US economy (Swanson and Williams (2014)). Forward guidance is inherent in statements regarding non-standard monetary policy but it is not a measure by itself, for example on December 3, 2015, the Governing Council commits to extend the EAPP ‘at least until March 2017’. Hence, since the introduction of forward guidance, any ECB communication could be classified as forward guidance because of the commitment for the future. Related to the announcement of unconventional measures one cannot disentangle the effect that is due to the measure and the effect that is due to its committing character. In the euro area, forward guidance was first implemented on the interest rate on July 4, 2013. This date arguably establishes a decisive change that alters expectations substantially for the subsequent announcements. Therefore, a regime dummy is implemented taking the value of 1 since July 4, 2013, and 0 before this date. As for the OMT regime, the regime dummy is insignificant in the country-specific analysis. The results remain qualitatively unchanged in all specifications when accounting for the forward guidance period.

The role of policy rate announcements

In unconventional times, ECB undertakes multiple measures at the same time. The communication of such tools and the explanation of their complementarities is essential (ECB (2017)). Not a single measure but the combination of both standard and non-standard announcements affects financial markets. It is in the nature of EBC’s communication strategy to announce regular monetary policy decisions as the adjustment of the policy rate concomitantly with statements on non-standard measures such as asset purchase programs. For that reason we are unable to disentangle the events when there was a policy rate reduction and a non-standard measure announced at the same time.

Nevertheless, to control for effects specific to the policy rate, we isolate the corresponding nine events (denoted with an asterisk in Table A1), six of them coincidence with the chosen 26 key announcements. This goes beyond related literature that is not concerned about parallel policy rate change announcements.^[30] A robustness check considers the remaining 20 key announcements. The coefficients of daAt are weaker and those of Germany and the Netherlands become insignificant. However, the overall yield-reducing effect still persists. In consequence, independent of policy rate changes, non-standard monetary policy announcements affect most government bond yields. This is supported by the panel regressions which remain qualitatively unchanged when omitting the six overlapping events.

Furthermore, we test for the distinct announcement effect of a policy rate change itself. For the period under consideration, such changes were announced on 21 dates (including the six overlapping events with seminal non-standard measure announcements). Specifically, the dummy dapolicyt (‘day after policy rate change’) replaces daAt taking the value of 1 only if there is a policy rate change announced the previous day. Interestingly, neither an immediate nor a delayed impact appears in any country. A further separation of the event set into 9 overlapping events and 12 pure policy rate change events does not alter the results. The panel shows a very weak impact of policy rate change announcements on government bond yields for core countries only. A potential explanation for these puzzling findings could be that the 21 changes were anticipated or at least not surprising enough to change market behavior. It is arguable whether a slight decrease in the policy rate matters given the interest rate is already close to the zero lower bound on most announcement dates. We believe the novel (non-standard) measure should outweigh the ‘old’ policy instrument during the days when there is an overlap. Hence, our study supposes the non-standard measures are so striking that they dominate other information released on the date under consideration, even a change in the policy rate.

Overall, when comparing both types of announcement, non-standard monetary policy announcements do have a significant yield-reducing effect while announcements regarding the policy rate do not seem to affect government bond yields.

Random selection

The significant impact of the 26 chosen events could merely be a statistical coincidence. Therefore, iteratively 26 dates are randomly drawn from the data sample and employed in the analysis. Even after 30 iterations the results do not indicate an impact of any randomly chosen event set on government bond yields. Similarly, to control for reactions to monetary policy announcements independent of their content, 26 dates are randomly drawn from the 132 monetary policy press releases made by the ECB during the observation period. Another draw of 26 events is made from the 71 ECB announcements listed in Table A1. Both tests do not produce any significant results, either. Hence, a monetary press release per se does not affect government bond yields. This underlines the appropriateness of the chosen events.

In contrast, when considering each announcement of a random event set individually, all dates suppose a highly significant impact on all yields. This is probably due to the utilization of Newey-West standard errors. Fomby and Murfin (2005) explain this issue with econometric terms. Arbitrarily selected event dates all seem to be highly significant even without any specific event happening on the chosen date because heteroscedastic and autocorrelation robust standard errors’ t-statistics are spuriously identified. Ford et al. (2010) illustrate this problem in a financial event study. Consequently, the focus on the aggregate announcements is less vulnerable to such econometric biases than the view on single a date.

In sum, the findings are robust to regime shifts, policy rate announcements and randomly drawn events. However, the number of events is crucial to the results. Taking few events (initial events, excluding (T)LTRO) makes the estimators of some countries insignificant. In contrast, employing many events (technical details) does not modify the estimators substantially. One should bear in mind this minor sensitivity when comparing the present findings with other studies. After all, the best way is to find economic arguments why to include the events ignoring the total number of chosen events. It has been decided to keep the baseline scenario of 26 events outlined in Section 2.2 as a middle way. It generates an economically meaningful result showing that both country groups are impacted by asset purchase program announcements – just the extent differs.

Choice of control variables

Instead of CESI, several other indices are implemented to control for macroeconomic surprises. On the one hand, the Morgan Stanley Capital International (MSCI) Europe Index as suggested by Jäger and Grigoriadis (2017) controls for European-wide events. In line with Haitsma et al. (2016) the MSCI World excluding Europe index is added to control for macroeconomic events outside Europe. On the other hand, country-specific MSCI indices replace the global variables to see whether the estimates improve. While the MSCI Europe index gives a significant estimator in most countries, the other two indices are unsuitable control variables showing significant estimators only for 5 countries: For MSCI World, the negative impact of such surprises to the yields of core countries is counterintuitive. The country-specific MSCI indices, in turn, suggest a mixed influence on yields with 6 insignificant estimators, 3 significantly positive ones (Austria, Ireland, Greece), and 2 significantly negative ones (France, Finland). Accordingly, the national indices seem to cancel out each other in the panel framework. Applying these MSCI indices, the Dutch and German yields become insignificant.^[31] This implies that a European surprise index as the CESI or the MSCI Europe are most suitable control variables for global surprises. We follow Georgiadis and Gräb (2016) and use ‘Citigroup Economic Surprise Indices [because they] are objective and quantitative measures of data surprises’ (p. 258). As a result, the main results are qualitatively robust to these variations.

We come to the same conclusions when employing the surprise and uncertainty indices developed by Scotti (2016) and the iTraxx Europe index to depict the investors’ preference for risk. Furthermore, the V2TX index is based on EURO STOXX 50 realtime option prices and reflects the market sentiment in Europe as in Falagiarda and Reitz (2015). It replaces the VIX. Similar to using the country-specific MSCI indices, the coefficient of daAt becomes insignificant for Germany and the Netherlands when substituting CESI or adding VIX as an additional control. These minor variations do not change the findings of the panel analysis. Since the estimators are still higher for periphery countries than for core countries we can confidently keep the CESI as main variable to control for macroeconomic surprises.

Effect on control variables

One might argue that monetary policy announcements directly affect the stock market. Haitsma et al. (2016) find that unconventional monetary policy surprises move European stocks while Fausch and Sigonius (2018) detect significant reactions for German stock returns. If this holds for the country-specific stocks, the application of both daAt and Δstockt as independent variables is impossible. To test for it, we take the country-specific stock market indices as dependent variable. Hence, Δyt is replaced by Δstockt in Equations (1), (2), and (7). Likewise, Δyi,t is replaced by Δstocki,t in Equations (3), (4), (5), (6), and (8).

The results demonstrate that ECB’s announcements neither have an immediate nor a delayed direct effect on the stock market indices. Hence, the choice of Δstocki,t as independent variable is appropriate. The announcements only have an indirect effect on the stock markets in the medium-term as higher liquidity induces rising asset prices.

Similarly, if the announcements directly impact global indicators such as the CESI, it is not feasible to include them as control variables. Therefore, ΔCESIt analogously replaces Δyt as dependent variable to control for interaction effects with daAt. Showing insignificant estimators for each specification, ECB’s announcements neither have an immediate nor a delayed direct effect on the CESI.

Yield spreads as dependent variable

Using the spread defined in Section 3.4 as dependent variable instead of the level, the results and hence the conclusions remain qualitatively unchanged. Taking the German yield as numeraire instead to calculate the spread gives similar results except there is no output for the German yields by definition. Put differently, the study confirms the spread-reducing effects worked out in related literature.

Alternative monetary policy surprise

The daily change in 3-month Euribor futures might be insufficient to control for monetary policy news. In particular in a low interest environment monetary policy news are reflected in the medium-term interest rates and not necessarily visible at a short horizon. Therefore, we follow Hanson and Stein (2015) and include the change in 2-year nominal sovereign yields to approximate monetary policy news. Specifically, for each country we add Δyieldt2y=yt+12y−yt−12y as an additional control in all regressions. All findings are robust to this modification. The results are also qualitatively unchanged when replacing Δfuturet by Δyieldt2y.

Variables in growth rates

Growth rates might be more suitable to compare yield dynamics of the various countries. First-differences only take the absolute differences into account independent of the level in the respective country. For instance, periphery countries typically state higher absolute changes in yields than core countries due to its higher yield level. In contrast, if one considers growth rates instead one corrects for this shortcoming by dividing by the absolute level. This might weaken the observed differences among the countries.

When using growth rates variables the estimators of daAt for Germany and the Netherlands become insignificant. In consequence, the estimator of daAt becomes insignificant for core countries in the panel specification. In addition, the coefficients of daAt resemble in size for all bonds. Hence, the utilization of growth rates instead of first-differences relativizes the previously found quantitative differences between both country groups. Nevertheless, the first-difference analysis suits better for policy analysis because it provides the absolute changes in yields. These are more relevant for the countries because they correspond to the overall short-term costs/benefits of a euro area government’s refinancing conditions in response to an ECB’s asset purchase announcement. Moreover, investors are presumably more interested in the absolute yield changes that represent actual profits/losses than in an abstract growth number.

In sum, when employing different data as control variables, the results remain robust for a given event set. Thus, we are quite confident with the results.