Qualitative and quantitative central bank communication and inflation expectations

3.1 The ECB qualitative communication index

Speeches, interviews and testimonies related to monetary policy made by the individual committee members are measured by a monthly index in the vein of the one by Ehrmann and Fratzscher (2007). This index covers communications by the six Executive Board members of the ECB and the Governors of the national central banks of the Eurosystem.^[7] It starts in June 2004 and ends in June 2011 to match the publication of ECB inflation projections. For this time period, Reuters News, a standard newswire service accessible through Factiva, is used to gather all reports about forward-looking policy statements. The focus is specifically on the future monetary policy inclination and explanations or clarifications of past decisions are not taken into account. Following Kohn and Sack (2004) and Ehrmann and Fratzscher (2007), the classification is kept as straightforward and general as possible. The search commands used are Governing Council, Governor, member, President, or Vice President along with interest rate or monetary policy. Moreover, only the first report in Reuters News, which directly follows the statement and is rather descriptive, is considered, the updates or analyses are not included. Each statement is classified into three categories: those that give an inclination of tighter monetary policy, no change, or lower interest rates:

MPt={+1tighter inclination0no change−1easing inclination

For each month between June 2004 and June 2011, three variables are computed. MP_ST provides the average inclination of all statements during a given month, so the policy stance of Governing council members’ communication. It is comprised between [–1; 1]. MP_INT displays the intensity of the communication so the number of statements times the stance MP_ST, and is used for robustness purposes. Finally, MP_DISP measures the dispersion of statements and is computed as the standard deviation of the stance of all statements during a given month. In a Bayesian updating model, the weight given to the signal (MP_ST) should depend on the precision of the signal. We assume that MP_DISP captures the precision of the signal conveyed in ECB speeches. Figure 1 plots the main ECB communication variables along with the ECB interest rate, while Table 1 provides some descriptive statistics.

Figure 1:

Main ECB variables and the KOF index.

Some interesting facts appear from the preceding figure and table. There are much more statements that have a tightening inclination than neutral or easing ones. Fifty-six percentage of the statements made between June 2004 and June 2011 have a hawkish inclination and this makes sense as interest rates were increasing or high for half of the sample period. It is nevertheless interesting to note that the ECB signals much more interest rate hikes than decreases. This is in line with Jansen and De Haan (2009) for the ECB or Hayo and Neuenkirch (2010) for the Federal Reserve who find that these central banks seem cautious about mentioning rate cuts too much. It can also be noted from Figure 1 that ECB projections, the ECB rate and ECB qualitative communication, either MP_ST or KOF (see below), are consistent with each other.

This classification methodology is usually referred to as “content analysis” because of the systematic analysis of the content of a message (Holsti 1969) and it is worth noting that this work is by nature judgmental and subjective. In particular, the choice has been made to focus on forward-looking conventional and unconventional monetary policy announcements with a possible effect on price stability and inflation, and not on policies providing liquidity to money markets and banks. We believe that it is important to differentiate policies aiming at price stability and financial stability following the usual segmentation of monetary policy mandates. Another possible caveat is that Reuters News may have not reported or misinterpreted some statements.

The present index is therefore compared to the KOF monetary policy communicator for the euro area which provides a quantitative measure of the ECB communication with a special focus on forward-looking statements concerning price stability (see Conrad and Lamla (2010) or the KOF website^[8] for more details) and is available on the same time span than the present index. It enables to assess the robustness and relevance of the latter. However, the KOF index translates the ECB President’s statement concerning risks to price stability as made during the monthly press conference (and only this specific Governing Council day) as well as ECB projections into a unique common index. In contrast, the index constructed in this paper encompasses all qualitative communication of each month and focuses specifically on speeches and statements in contrast with ECB projections.^[9] Table 1 shows the correlation matrix between the KOF index and all indices of this study.^[10]

3.2 ECB projections

The ECB/Eurosystem staff macroeconomic projections^[11] for the euro area are produced biannually since December 2000, and quarterly since June 2004 with a special emphasis on their disclosure to the public. They are published during the first week of March, June, September and December and are presented as ranges for both the harmonized index for consumer prices (HICP) and real GDP. The ranges are based on twice the mean absolute projection error of historical projection errors to reflect uncertainty. As common for the Federal Open Market Committee (FOMC) at the Federal Reserve forecasts, the midpoint of the range is used to represent ECB projections. Until 2006Q1, the underlying scenarios for interest rates and commodity prices were that these variables remain constant over the projection horizon; since 2006Q2 they are based on market expectations derived from future rates.^[12] These projections are published as average annual percentage changes and target current and next years, so are fixed-event projections. They might have seasonal effects as the forecasting horizon decreases quarter after quarter: one might suppose that the effects of ECB inflation projections on private ones are stronger in the beginning of each year and smaller at the end when more information is known on actual variables.^[13] Finally, the sample considered here starts in June 2004 when ECB projections became quarterly so as to combine the need for high-frequency data to measure qualitative communication and the relative low-frequency of publication of ECB projections. In addition, we interpolate quarterly ECB projections to monthly frequency by filling the gaps of the 2 months following their disclosure to the public with the value of the last projection published.^[14] This assumption seems reasonable as it respects the information structure and corresponds to the information set available to: private agents in the following 2 months of each quarter. However, this assumption introduces a bias against ECB projections which remain constant during 2 months whatever the macroeconomic or policy developments.

3.3 Private forecasts

The private inflation forecasts come from Consensus Economics Inc. The consensus forecasts (CF) is a monthly survey of quantitative predictions of private professional forecasters, with an average of 30 institutional respondents, for about fifteen macroeconomic variables including the overall index for HICP for the euro area (with changing composition), as measured by Eurostat and forecasted by the ECB, and calculated as average annual percentage change for current and next years. Because of its fixed-event nature, the forecasting horizon decreases every month, and we introduce month fixed-effects in our estimations to control for that and propose two robustness tests (see footnote 11). Surveys are collected at the end of the first week or beginning of the second week of each month.

We also consider a financial market-based measure of inflation expectations derived from inflation swaps 5-years 5-years-forward. The instruments are financial market contracts to transfer in 5 years the inflation risk over the 5-years-forward from one counterparty to another. We consider the average of all the working day observations in each month to be consistent with CF data. In general, the advantage of financial market expectations over survey measures of expectations is that they are directly related to payoff decisions, so there is no strategic response bias or no difference between stated and actual beliefs. Although one disadvantage is that financial market expectations do not provide a direct measure of inflation expectations as they may be affected by credit risk, liquidity and inflation risk premia. Swaps tend to be a better market measure for deriving inflation expectations than inflation-indexed bonds because they are generally less sensitive to liquidity and risk premia.

3.4 Other variables

The ECB interest rate considered is the main refinancing operations interest rate. It enables to check whether ECB communication may be a proxy for ECB decisions or whether ECB communication adds some specific information to private inflation expectations’ formation. Indeed, the ECB qualitative communication variable may measure the “procyclical” effect of the speech (e.g. when central bankers say they expect high future inflation, then private inflation expectations should increase) and in the meantime may capture the “countercyclical” effect of the same speech (e.g. when central bankers say they will increase interest rates, private inflation expectations should decrease). With the implementation of non-standard policy measures after 2008, one may argue that the interest rate is not necessarily capturing all aspects of the monetary policy stance. Therefore, we also use the ECB shadow rate estimated by Wu and Xia (2016).

We use various controls for the macroeconomic environment that should in theory impact the private expectations formation. Core the HICP considering all-items excluding energy and unprocessed food, the output gap (the HP-filtered monthly interpolated real GDP), the credit growth rate (to euro area residents other than governments), the oil price growth rate (based on the Brent crude oil spot price) and a proxy for global uncertainty and liquidity risks capturing the financial instability originating in the United States during the 2008–2009 crisis (the VIX index) are included in the empirical model as control variables.^[15] These variables, especially the output gap, credit growth and the VIX, also enable to control for the financial turmoil and the great recession that take place during the sample period. The overall sample starts in June 2004, ends in June 2011 and is constituted of 85 monthly observations.

4.1 Empirical model

Based on the framework described in Section 2, we estimate the effects of ECB qualitative communication, projections and interest rate on private inflation forecasts beyond the effects of past private inflation forecasts – which explain approximately 85% of the variance of private inflation forecasts – and macroeconomic controls. Alternatively, one can view the expectations formation process as an AR(1) process complemented with the relevant information set used by private agents to predict inflation. Because private inflation forecasts are formed, collected and published at the early beginning of each month and in order to comply with the information timing structure, we assume private inflation forecasts dated t are formed based on the information set dated t–1 (see Figure 2). This is the only information set available to private agents and respecting the data generating process of macroeconomic variables.^[16] Based on this information structure constraint, private forecasts at date t are therefore regressed on all variables at date t–1.

Figure 2:

Timeline of information structure.

This information set includes variables likely to help predict inflation, in addition to ECB qualitative and quantitative communication and the ECB rate: core HICP, the output gap, credit growth, the oil price and a measure of uncertainty. Core HICP is supposed to affect positively inflation as its underlying fundamental driving force, the output gap similarly through a Phillips curve, credit growth is also supposed to have a positive effect via the quantity theory of money, the oil price is supposed to capture the positive impact of commodity and volatile prices on inflation while the effect of uncertainty may have different effect on inflation expectations: an inflation risk premium would increase inflation, while uncertainty may be associated to recessions and falling inflation. One would also want to include a macroeconomic news variable, but this seems impractical because of the format of private forecasts. Usually the news variable encompassing the information set released between t–1 and t is computed as the difference between the forecast of a given variable (inflation) in t–1 and the actual value of the given variable in t, and this is not possible with CF forecasts as the monthly forecasts are not for the next month horizon. One may nevertheless argue that the news component is small as the ECB qualitative communication variable encompasses all speeches during a given month and all macroeconomic variables are generated at the end of this given month, while private forecasts are formed at the early beginning of the following month. We therefore implicitly assume that price and monetary policy news, which affect expected future inflation (Beechey and Wright 2009), are comprised in t–1 variables and that no news is published between the end of a month and the early beginning of the following month when private agents form their forecasts.

Equation (2) can be rewritten by decomposing the vector X_t–1 in two vectors, one comprising our variables of interest and the other the macroeconomic controls. Because we include a lag of the left-hand side variable in our model, this is equivalent to looking at how inflation expectations change (so having the first difference in inflation expectations as the left-hand side variable) in response to different right-hand side variables, but without an implicit assumption about the inertia coefficient. The estimated equation where y_t,h is the private forecast made in t for a given event date h is therefore:

where Π_t contains the three policy variables: ECB projections, the ECB qualitative communication (either MP_ST, MP_INT or MP_DISP) and the ECB rate, and Ω_t encompasses the macroeconomic controls: core HICP, the output gap, credit growth, the oil price, the proxy for uncertainty and month fixed-effects. The model is estimated with robust standard errors using heteroskedastic and autocorrelation-consistent (HAC) robust variance estimates.^[17]

With forward-looking behavior and intertemporal smoothing, random shocks that affect private forecasts are likely to also affect ECB and macro variables, and all those variables are likely to be endogenous to private inflation forecasts (said differently, their correlation with the error term ε is not equal to zero). In order to solve the identification issue, we assume that only our three variables of interest – the vector Π_t – are endogenous and that the macro controls are exogenous. The equation (3) is estimated with IV using two-stages-least-squares (2SLS) to identify the causal effects of the three ECB variables.

Another issue arises. The IV estimator requires additional variables that are correlated with these endogenous regressors but not with the error term ε, what Stock, Wright and Yogo (2002) call the “weak identification” problem as instruments are only weakly correlated with the included endogenous variables. Bai and Ng (2010) and Kapetanios and Marcellino (2010) propose to use factor analysis to overcome weak identification in IV estimations as they show that the estimated factors can be more efficient IV than observed variables. Using a principal component analysis, we estimate and generate three factors as linear combinations maximizing the common variance of ECB current- and next-year projections, MP_ST, MP_INT, the ECB rate, the ECB shadow rate, core HICP, the output gap, credit growth, oil prices.^[18] We believe this set of variables is relevant because it will generate instruments that encompass the information set of policymakers, their communications and actions, and the most likely economic indicators used in their reaction function. The Kaiser-Meyer-Olkin measure of sampling adequacy provides a simple way to assess the relevance of applying principal component analysis on the selected variables by comparing the partial correlations and correlations between variables and is provided in Table 2 which summarizes the estimation and characteristics of these factors. The first three components capture 87% of the cumulative variance of the underlying series, and we use them as instruments of the vector Π_t gathering the three endogenous policy variables.^[19]

Tests of weak identification are reported in Table 2 and at the bottom of Tables 3–8.^[20] First, we assess the predictive power for the endogenous regressors of our three instruments compared to a simple AR(1) model (the use of lags as instruments being the benchmark in the macroeconomic literature) and provide the corresponding adjusted R² in Table 2. Second, the Cragg-Donald Wald test statistic assesses whether the excluded instruments are correlated with the endogenous regressors. A rejection of the null should be treated with caution however, because weak instrument problems may still be present. Third, the Anderson canonical correlations LM statistic informs whether the equation is identified – that the excluded instruments are correlated with the endogenous regressors. Under the null, the statistic is distributed as χ² with 1 degree of freedom here. A rejection of the null indicates that the model is identified. All tests confirm that the instrument set is relevant for identifying the causal effect of the three variables of interest.^[21]

4.2 Linear estimates

Table 3 displays estimates of the effect of the ECB rate, the qualitative and the quantitative ECB communication variables on current-year private forecasts. The first column presents the baseline model tested. ECB current-year inflation projections have a positive effect (0.25) – consistent with Hubert (2015) for a group of inflation targeting central banks and Hubert and Maule (2016) on UK data – and this suggests according to the hypothesis 2 that the signal about the future path of inflation dominates the signal about policy. ECB qualitative communication, measured by MP_ST and capturing the stance of overall communications, and the ECB rate are not significant as put forward by the hypotheses 1 and 3. Lagged private inflation forecasts have positive and significant effects together with the output gap and oil prices.

The second estimation replaces the ECB rate by the ECB shadow rate estimated by Wu and Xia (2016) in order to take into account the non-standard policy measures implemented after 2008.^[22] In the third and fourth columns, ECB qualitative communication and ECB projections are respectively removed from the estimated model to test for the potential substitutability of the ECB communication types. In column 5, the GMM generalization of the LIML estimator to the case of possibly heteroskedastic and autocorrelated disturbances – the continuously updated GMM estimator or CUE – is used. In column 6, the ECB qualitative communication variable MP_ST is replaced by the intensity of the qualitative communication MP_INT. In columns 7 and 8, ECB projections are replaced by the quarter-over-quarter difference in ECB projections and the difference between ECB projections and CF forecasts, respectively. In column 9, the sample starts in 2006m4 after the ECB changed the way its projections are formed. The ECB used a constant interest rate scenario before 2006m4 and has used a market interest rate assumption since then. In column 10, the ECB projections are decomposed into two variables for ECB projections published during the first semester and the second one. Because ECB projections are fixed-event forecasts, we expect projections in the first semester to have more effect than those published in the second semester.^[23] In column 11, ECB next-year projections replace the ECB current-year ones in order to assess whether longer-horizon projections have a different effect on current-year private forecasts. In column 12, the frequency of the dataset is quarterly as ECB projections are published on a quarterly basis but interpolated to monthly frequency in the baseline model. We assess the impact of this data transformation on our results by estimating our empirical model at quarterly frequency. We regress CF forecasts of March, June, September and December on the ECB rate and ECB projections of March, June, September and December, available to private forecasts since the first week of these months, and lagged CF forecasts and controls of February, May, August and November.^[24] In column 13, the dispersion of the ECB qualitative communication is added to the vector Π_t of endogenous policy variables. MP_ST is replaced by PR_ST and NPR_ST, qualitative communication indices for the President and the other members of the Governing Council, in columns 14 and 15, respectively. In column 16, an alternative monthly communication index MP_ST2 is computed with only the communications made after the policy meeting.^[25]

All robustness tests confirm that current-year ECB inflation projections have a positive effect on current-year private inflation forecasts (except in the case of the variation in ECB current-year projections) whereas ECB qualitative communication and the ECB rate are not significant. Although increasing ECB projections have no effect on private forecasts, it is interesting to note that ECB projections higher than private forecasts positively affect private inflation forecasts. Two interpretations of the main result and the positive coefficient associated to ECB projections are possible: A. the ECB somewhat creates inflationary pressures by publishing its forecasts as private forecasters now expect inflation to increase, whereas B. by communicating on an 1% inflation increase that is going to happen (the inflation outlook signal), and therefore signaling its intention to counter it – under the assumption of an inflation adverse and credible central bank – (the policy path signal), private agents only increase their inflation expectations by 0.25% and the ECB has succeeded to dampen in part inflationary pressures.

Neither the ECB qualitative communication nor the ECB rate appears significant. One may argue for the latter that the forecasting horizon being inferior to the delays of transmission of monetary policy, it is consistent that the ECB rate has no effect on current-year inflation forecasts. Concerning the effect of qualitative communication, the same argument may apply to this variable which captures the future orientation of monetary policy by focusing on all forward-looking statements referring to policymaking. Finally, it is worth stressing that the two communication types are not substitutes (see columns 4 and 5).

Table 4 follows the same pattern than Table 3 and displays estimates of the effects on private next-year inflation forecasts. ECB next-year projections and the ECB qualitative communication MP_ST are not significant, whereas the ECB rate has a significant negative effect (–0.09). Moreover, lagged private forecasts, the output gap, credit growth and oil prices appear to be additional determinants of private next-year inflation forecasts. A potential reason for which ECB projections should be less important in determining private next year forecasts is that the ECB rate is supposed to impact next year inflation and so, private agents are more prone to focus on the policy instrument when forming their expectations at this horizon, as confirmed by the significant effect of the ECB rate. This suggests that actions speak louder than communication at the horizon when monetary policy is supposed to have its maximum impact on the economy, consistent with Gürkaynak, Sack, and Swanson (2005). It is also interesting to note that, when considering the first difference of ECB qualitative communication (column 7), the associated coefficient becomes positive and significant. The fact that the change in the tone of ECB speeches has a positive effect on inflation expectations suggests that the change in the ECB qualitative communication is interpreted by private agents as signaling the change in risks to price stability as perceived by policymakers and therefore signals inflationary pressures to private agents rather than the future policy path.

Table 5 follows the same pattern than Tables 3 and 4 and presents estimates of the effects on private 1-year-ahead inflation forecasts. We construct 1-year-ahead fixed-horizon forecasts as a weighted average of fixed-event forecasts based on the number of quarters forecasted in both the current and next years following Dovern, Fritsche, and Slacalek (2012).^[26] These results are to be compared with Table 4 for next-year forecasts since for each month of a given year, the 1-year-ahead forecast falls on the next calendar year. This transformation does not alter the estimated results shown in Table 4: ECB projections and the ECB qualitative communication variable are not significant whereas the ECB rate has a significant negative effect and the change in the tone of ECB speeches is confirmed to have a positive effect.

Table 6 shows estimates when the dependent variable is 5-years 5-years-forward inflation expectations obtained from inflation swaps. The results should be analysed differently from those of Tables 3–5. In the previous tables, we look at how central bank qualitative and quantitative communication and action affect short-term inflation expectations, so focusing on the monetary policy transmission mechanism. In this set-up, we look at the response of long-term inflation expectations, so focusing on how private agents perceive steady-state inflation to evolve and central bank preferences and credibility to maintain inflation in line with its target. The ECB rate still has a negative effect but twice smaller. ECB next-year projections and the qualitative communication variable have negative effects too, but only if they are considered alone. Their negative effect suggests that both types of communication signal the policymakers’ intention to counter inflationary pressures (the policy path signal) more than the inflationary pressures themselves, which is consistent with the horizon studied. The fact that their effects are not significant simultaneously suggests that the two types of communications may be substitutes at long-term horizons.

It appears from these linear estimates that ECB projections are rather a very short-term communication tool to affect private current-year inflation forecasts which may help taming private inflation expectations by signaling policymakers are aware of forthcoming inflationary pressures (the policy path signal reduces the inflation outlook signal) while only the change in the ECB qualitative communication affects private next-year or 1-year-ahead inflation forecasts and seems to signal more inflationary pressures as perceived by policymakers than the future policy path. Short-term estimates therefore suggest that both types of communication are not substitutes and that actions speak louder than communication at the horizon when monetary policy has traction on the economy. At the opposite, long-term estimates suggest that the two types of communications may be substitutes and signal the endogenous reaction of policymakers (the policy path signal). The differentiated outcomes across horizons is consistent with Neuenkirch (2013) who find that shocks to the KOF index have a positive effect on inflation expectations on the short run and a negative effect on the long run.

4.3 Interacted effects

After having estimated the simultaneous linear effects of both types of communication and of the ECB rate on private inflation forecasts, we investigate to what extent central bank communications and action interact. One could for instance expect that qualitative communication affect the way ECB projections are interpreted and that hawkish speeches conveying a signal about the future inflation outlook reinforce the positive effect of ECB projections on private current year inflation forecasts.

Starting from the baseline model described by the equation (3); an interaction term is added to the vector Π_t and also considered as endogenous. Since all interacted variables are continuous, the interaction is analysed in terms of the multiplication of the predictor and the moderator variable to simplify the interpretation of the results. The additional instrument is generated as the product of the multiplication of the predictor variable with the component most correlated to the moderator variable. In addition to the value of each coefficient, Table 6 presents the effect of the predictor variable on private forecasts for low and high values of the moderator variable, defined as the mean –1 SD and mean +1 SD of the relevant variable (Table 1 provides these descriptive statistics for all variables).

The left panel of Table 7, focusing on the effect on private current-year forecasts, provides evidence of non-linear effects. First, the effect of ECB projections positively depends on the level of ECB projections. The same is also true for ECB qualitative communication (MP_ST) whereas we find that this variable has no significant linear effect. Second, ECB projections have more effect when interacted with ECB qualitative communication: the effect of ECB projections on private inflation forecasts is stronger with high rather than low values of MP_ST. More hawkish speeches convey stronger signals about inflationary pressures and increase the relative strength of the inflation outlook signal of ECB projections. Third, ECB qualitative communication (MP_ST) has a more positive effect on private inflation forecasts (the inflation outlook signal dominates) when the dispersion of speeches (MP_DISP) is high whereas it has a negative effect on private inflation forecasts (the policy path signal dominates) when the dispersion is low, so there is a consensus among policymakers. Fourth, ECB projections have more effect when interacted with the ECB rate. The pattern here is similar to the interaction with ECB qualitative communication: ECB projections associated to a high interest rate convey stronger signals about inflationary pressures and increase the relative strength of the inflation outlook signal of ECB projections.

The right panel of Table 7, focusing on the effect on private next-year forecasts, provides no evidence of non-linear effects, while the ECB rate remains the main determinant (among ECB variables) of private expectations. This confirms the result from the preceding subsection that actions speak louder than communication at the horizon when monetary policy has traction on the economy, consistent with Gürkaynak, Sack, and Swanson (2005).

Table 8 shows estimates for 5-years 5-years-forward inflation expectations, and provides evidence of non-linear effects. Whereas the linear effects of ECB projections are positive on current-year inflation forecasts, null on next-year and 1-year-ahead inflation forecasts, and negative (when substituted with qualitative communication) on 5-years 5-years-forward inflation forecasts, ECB projections have more negative effects on private inflation forecasts when they are lower rather than higher. Similarly, the effect of ECB projections is also more negative when the ECB rate is low. It suggests that high ECB projections or a high ECB rate reinforces the inflation outlook signal of ECB projections.

The policy implications are the following. First, policymakers may want to pay attention to the interacted effects of their communication types with each other and with the ECB rate. At very short- and long-horizons, ECB speeches or the ECB rate have a tendency to reinforce the effect of ECB projections when they are consistent and convey the same signal about inflationary pressures. Last, the dispersion of speeches makes it more difficult for people to take a signal about the future path of policy and reinforces the inflation outlook signal and is therefore damaging to the management of inflation expectations.