Did the 2022 energy inflation surge dent European Central Bank and Fed credibility? Evidence across horizons

Together with Leonardo Cadamuro

This post reproduces the main text of Bruegel Working Paper 22/2026, “Did the 2022 energy inflation surge dent European Central Bank and Fed credibility? Evidence across horizons” (DOI 10.64153/RXBL1266). Appendix A, on alternative US expectations series, and Appendix B, on the regression methodology, are in the working paper. We gratefully acknowledge input from Andreas Billmeier, Rebecca Christie, Gene Frieda, Hans Geeroms and Guntram Wolff.

Abstract

The ultimate test of central bank credibility is whether it maintains price stability. However, inflation is affected by shocks. For these to wash out, long averaging is needed, delaying assessment of the central bank’s price-stability performance. It may be necessary to wait years to conclude whether a central bank did its job.

We propose a complementary credibility test that uses empirical proxies of inflation expectations instead of actual inflation as the variable to be stabilised. We apply our test to the energy inflation shock caused by Russia’s invasion of Ukraine and ask whether this significantly impacted European Central Bank and Federal Reserve credibility, with perfect credibility meaning no impact of current inflation on expected inflation over the appropriate horizon, while zero credibility implies full pass-through, or even a magnifying effect. We test credibility at various horizons and distinguish between speed credibility (the ability and willingness to act promptly, visible at short horizons) and destination credibility (the perceived commitment to the long-run target, visible at long horizons).

We find the ECB lost more speed credibility than the Fed but preserved its destination credibility. The Fed preserved speed credibility but showed more long-run drift. Current inflation significantly affected forecasts in both jurisdictions at the one-year horizon, but the effect was significantly larger in the euro area. At the two-year horizon, both central banks broadly maintained medium-term anchoring. At the longer horizon, the ECB avoided any effect of current inflation on inflation forecasts, while US expectations drifted somewhat. Overall, the energy inflation shock did not dent significantly the inflation credibility of either central bank, though we set out some nuances.

1. Introduction

When HICP inflation broke above five percent in the euro area in January 2022, the European Central Bank qualified the surge as transitory. Inflation then peaked at 10.6 percent in October 2022 (the highest since the euro’s creation). The ECB raised interest rates for the first time in over a decade only in July 2022. The Federal Reserve moved in March 2022, four months earlier, having publicly signalled a pivot since November 2021.

The 2021–2022 inflation episode provides ideal material to test the inflation credibility of the two central banks, with results depending critically on which horizon is used. Examining different horizons of the inflation expectation curve is critical for our test, which checks the credibility of the central bank’s engagement in stabilising inflation around its objective. Over the shortest horizon we consider (12 months), credibility does not require inflation to return to two percent; that could require massive, damaging and thus unrealistic monetary tightening. Over the medium term – two years in our case – expectations should be brought back toward price stability, even if the path is uncertain. In the long run – five or ten years in our case – the two percent target must be reconfirmed as a permanent feature of the monetary regime. These three horizons can move independently and the 2021–2022 episode was an unusually clear illustration of this.

Two clarifications are needed before proceeding. First, we use one narrow definition of credibility: the degree to which professional forecasters’ inflation expectations1 stay anchored to the two percent target, over the relevant horizon. This is the critical, but not the only, component of central bank credibility; reputational, political and broader market dimensions lie outside our measure. Second, the use of the expectations term structure, rather than a single point on that curve, brings out the distinction between speed credibility (the perceived willingness to act promptly, identified at the short horizon) and destination credibility (the perceived commitment to the long-run target). The 2021–2022 episode separated the two: the ECB lost more speed credibility than the Fed, but preserved its destination credibility, whereas the Fed preserved speed credibility but experienced some long-run drift.

Our core result is a clear difference across horizons that single-horizon analyses would miss entirely. Using quarterly ECB Survey of Professional Forecasters (SPF) data at the one-year, two-year and five-year horizons for the euro area, and the Philadelphia Federal Reserve Bank SPF at one-year, two-year and ten-year horizons for the United States, we estimate pass-through coefficients between actual and expected inflation, separately in four sub-periods. The ECB did not prevent one-year-ahead expectations moving higher during the inflation surge, signalling a risk to its credibility. However, medium-term anchoring held and long-run institutional credibility was essentially unimpaired and, on our measures, better maintained than the Fed’s. Visual inspection of the relationship between current and forecast inflation is consistent with these findings, providing a more precise diagnosis than treating the episode as a wholesale loss of ECB or Fed credibility.

The ECB’s problem in 2021–2022 was thus speed, not destination: professional forecasters kept believing that euro-area inflation would return to two percent but they did not believe the ECB would move fast enough to get there within one year. That delay was largely the by-product of a sequencing commitment not to raise rates before net asset purchases had ended, which delayed by months the pivot signal the Fed had already sent. The priority before the next shock is therefore not to reaffirm a commitment to the two percent target, which never wavered, but to reform how the ECB signals the speed of its reaction function.

This paper addresses two robustness challenges head-on. First, low pass-through over long horizons might be a mechanical artefact: professional forecasters routinely converge their long-run forecasts toward the two-percent target regardless of conditions. We test this with a long-short sensitivity test following Gürkaynak et al (2010): if anchoring is genuine, longer-term expectations should not respond to revisions in one-year expectations, even during the surge. The post-peak data indeed shows near-zero sensitivity in all sub-periods. Second, the euro area is a major energy importer, while the US has been a net petroleum exporter since 2019. Higher euro-area pass-through over the one-year horizon might simply reflect the asymmetric energy composition of the shock. We address this by repeating the one-year horizon pass-through regressions for the ECB with core inflation as the regressor, and our result is reinforced: core pass-through during the surge (≈ 0.92) is larger than headline inflation pass-through (0.45), ruling out energy as the driver.

The remainder of this paper is organised as follows. Section 2 reviews the relevant literature. Section 3 presents the conceptual framework. Section 4 describes data and methodology. Section 5 reports the main empirical results. Section 6 presents robustness checks. Section 7 interprets the findings and discusses policy implications. Section 8 concludes. Appendix A provides results with alternative US expectations series.

2. Literature review

sec:lit

The theoretical case for treating credibility as a distinct, accumulating asset goes back at least to Blinder (1997), who argued that it depreciates slowly through inaction and can be partially restored by decisive communication even before formal policy changes. Erceg and Levin (2003) showed formally that reputational damage accumulates when inflation persistently exceeds target: agents revise upwards their estimates of the central bank’s implicit price objective, with lasting effects that depend on the duration of the deviation. Orphanides and Williams (2005) formalised this in a learning framework in which short-term expectations initially drift higher and medium-term ones follow if the commitment signal is weak or delayed. A direct implication of these models is that the horizon profile of pass-through should reveal where in the credibility timeline the signal was unclear. Afrouzi et al (2024) added a contemporary dimension, arguing that the post-pandemic episode should be understood against a backdrop of rising fiscal pressures and eroded central bank independence that has made credibility more fragile.

The financial-markets literature provides both evidence and the main methodological tool we adopt. Gürkaynak et al (2005) showed that distant-horizon nominal forward interest rates in the US respond to macroeconomic surprises, a sign that the long-run inflation expectations embedded in them are not perfectly anchored, even in low-inflation regimes. Gürkaynak et al (2010) extended this to a cross-country comparison, finding that explicit inflation-targeting countries show lower sensitivity of long-run forward rates to news, but also that this property can erode when the target is missed persistently. Their methodological contribution was to test the sensitivity of long-horizon expectations to short-horizon revisions rather than to realised inflation directly, which sharpens identification by separating the question of whether the long-run expectations move, from the question of whether short-run forecasts affect longer-term expectations. We apply this logic to survey expectations, for which the absence of risk and liquidity premia makes the test less ambiguous than it is in financial markets. Levin et al (2004) showed that inflation-targeting regimes reduce both the persistence of inflation and the co-movement of expectations across horizons.

Survey-based evidence on euro-area inflation anchoring before the surge paints a nuanced picture. Łyziak and Paloviita (2017) documented that long-term euro-area professional expectations were reasonably anchored before 2015. Natoli and Sigalotti (2018) found early upward tail sensitivity, an increase in the probability mass assigned to inflation outcomes above two percent at long horizons, visible in the upper percentiles of the cross-sectional distribution of ECB SPF five-year forecasts. Corsello et al (2021) used ECB SPF microdata to show that long-term expectations had already de-anchored downward from the two percent target by mid-2019, a below-target drift later documented by Paloviita et al (2021), who estimated an effective implicit ECB target of approximately 1.6 percent to 1.7 percent.

As we will show, this below-target baseline matters when interpreting the inflation surge: part of the upward movement in short-term ECB expectations reflected a normalisation toward target, rather than genuine de-anchoring above it. Dovern and Kenny (2020) documented rising cross-sectional dispersion of medium-run expectations after the Global Financial Crisis. Neri et al (2022) confirmed that all major ECB anchoring measures agreed on medium-term stability, while diverging at shorter horizons during the surge, a finding consistent with our horizon differences.

The macroeconomic determinants of the 2021–2022 inflation surge are well documented. bernanke2023 (bernanke2023, bernanke2024) decomposed the global surge into commodity shocks, supply-chain disruptions and demand-composition shifts, with the relative importance varying across countries and over time. Shapiro (2022) decomposed PCE (Personal Consumption Expenditures) inflation into supply- and demand-driven components, finding the early 2021 US surge predominantly supply-side. Bianchi et al (2023) offered a fiscal interpretation, suggesting that comparing ECB and Fed credibility requires controlling for differential fiscal contexts (larger euro-area fiscal heterogeneity, sovereign spread risk) that are orthogonal to the monetary policy credibility question we study. The supply-versus-demand composition of the surge also differed across the two jurisdictions, with a larger demand and fiscal component in the US; this asymmetry is relevant for interpreting the cross-jurisdiction comparison. We return to it in the discussion.

The measurement of inflation expectations and the interpretation of pass-through regressions have received careful attention. Coibion and Gorodnichenko (2015) documented information rigidities that distinguish professional from household forecasters, making the two populations respond differently to the same inflation news. D’Acunto et al (2022) provided a comprehensive taxonomy of expectation formation. Reis (2023) catalogued four pitfalls in using expected inflation, most relevantly that professional forecasters are least appropriate for measuring policy transmission to price-setting behaviour.

We take this latter critique seriously: our focus throughout is on institutional credibility as assessed by informed agents, not on the Phillips curve expectations channel. The papers most directly related to our focus are Baumann et al (2025) and Coibion and Gorodnichenko (2025), who confirmed that SPF-based expectations appeared anchored at long horizons during the inflation surge, while household surveys did not (a divergence we confirm for the Michigan Survey vs Philadelphia SPF comparison; see Appendix A). Ahn and Farmer (2024) showed that the dispersion of long-run expectations rose above historical norms even when the mean remained anchored. Gennaioli et al (2024) offer a memory-based micro-foundation for expectation de-anchoring that is consistent with the asymmetric horizon response we document. Burban et al (2024) found via a time-varying parameter model that the long-term ECB anchor remained stable throughout the surge, while the one-to-two-year path showed drift, broadly corroborating our sub-period OLS results.

3. Conceptual framework

sec:framework

We measure inflation credibility at the relevant horizon h by the pass-through slope β from regressing h-year-ahead inflation expectations on contemporaneous actual inflation:

Et[πt+h] = α + β · πt + εt (1)

Under full credibility β = 0, current inflation is uninformative about future inflation because the anchor is completely binding. Under zero credibility β ≥ 1, agents extrapolate current inflation fully or more than fully, with β = 1 corresponding to a random-walk view of inflation and β > 1 to an extrapolation beyond the current level. We estimate Equation (1) separately by sub-period and horizon, yielding a (4 × 3) matrix of pass-through coefficients that maps the horizon gradient over the cycle.

Sub-period OLS is preferred to continuous time-varying alternatives (rolling regressions, state-space models) because our central question is whether pass-through changed at clearly identified dates, not how it continuously evolved. Sub-period OLS asks a direct question: did the slope change between the pre-crisis baseline and the surge? Each sub-period contains 18–60 monthly observations, sufficient to identify β̂ reliably given the serial correlation structure of the data. To assess whether any apparent change is statistically significant, we checked our dates with Chow F-tests and data-driven Bai-Perron break detection (Section 6).

A concern is that β ≈ 0 at long horizons may be a mechanical artefact: professional forecasters are aware of the two percent target and routinely mean-revert their long-run forecasts toward it regardless of conditions, making low pass-through trivial. To distinguish genuine from mechanical anchoring, we implement the long-short sensitivity test of Gürkaynak et al (2010):

ΔEt[πt+H] = α + δ · ΔEt[πt+1] + ut (2)

Under genuine long-horizon anchoring, δ = 0: long-horizon expectations do not respond to revisions at short horizons. The test is sharper than it might appear. A purely mechanical mean-reversion story predicts some positive co-movement, however small, because forecasters who revise upward at one year must do so at five years too (only more slowly) if they have consistent probabilistic models. Only if five-year expectations are governed by a genuinely separate process (the target itself, rather than the path to it) would we observe δ = 0 during a surge when short-term expectations rise sharply. The survey nature of our data is important here: unlike financial market forward rates, SPF five-year expectations contain no inflation risk premium that could co-move with short-horizon news for reasons unrelated to the inflation outlook. This makes the zero null in Equation (2) cleaner to interpret.

As a complementary check, we estimate forecast-error pass-through:

ΔEt[πt+H] = α + γ · (πt – Et−12[πt]) + vt, where the bracketed term is the 1yr forecast error (3)

where the regressor is the one-year forecast error (realised minus expected inflation lagged twelve months). This test asks a different question from Equation (2): not whether long-horizon expectations co-move with short-horizon revisions, but whether they respond to inflation surprises. A near-zero or negative γ̂ during the surge would mean that above-forecast actual inflation led forecasters to lower their long-horizon expectations, consistent with interpreting the surge as transitory rather than as information about the long-run regime. A positive γ̂ would be evidence of genuine de-anchoring at the five-year level.

4. Data and methodology

sec:data

Sample. January 2015 to June 2025, spanning four episodes: pre-crisis (2015–2019), COVID-19 (2020–mid 2021), inflation surge (mid 2021–2022) and post-peak (2023–2025). The sample begins in 2015 to avoid conflating the credibility question with the immediate post-global financial crisis (GFC) normalisation period, during which ECB expectations were depressed for reasons unrelated to monetary policy. The COVID-19 episode is retained as a distinct sub-period because the inflation expectations dynamics in 2020–2021 (sharp deflationary fears followed by rapid normalisation) form a natural counterfactual to the surge.

Expectations. For the United States, we use three series from the Philadelphia Federal Reserve Bank Survey of Professional Forecasters microdata (SPFmicrodata.xlsx). The one-year and two-year measures are the cross-sectional means of the annual average CPI forecasts for the current year and the following year (columns CPIB and CPIC in the microdata). The ten-year measure is the cross-sectional mean of the CPI10 question, available from 1991Q4. For the euro area, we use ECB SPF mean point forecasts from the quarterly microdata files at one-year, two-year and longer-term horizons; the longer-term question targets four calendar years after the survey year and we refer to it as the `five-year’ horizon following standard SPF convention. The ECB SPF panel comprises roughly 40 to 60 respondents per question and the Philadelphia SPF fewer (about 33 at the most recent rounds). We use cross-sectional means throughout and examine forecaster disagreement separately in Section 6.

For the US ten-year Philly SPF and the ECB SPF five-year, the horizons are not the same (ten years versus approximately four to five years). The consequences of this difference for the interpretation of our results are discussed alongside Table 6. What matters most is the within-jurisdiction comparison across horizons: does pass-through fall sharply as we move from one year to the longer horizons? As we will show, it does so for both central banks, but more so for the ECB.

To get a sense of the differences between inflation expectations obtained with different methodologies, Figure 1 places the survey-based long-horizon measures used for the ECB and the Fed alongside, for the US, the FRED T5YIFR (US five-year forward five-year TIPS breakeven) and the NY Fed Survey of Consumer Expectations (SCE) median one-year-ahead household expectation, constructed from the SCE microdata (monthly, 2015–2025), and, for the euro area, the five-year/five-year forward inflation swap. The SCE series is not used in the main regressions (it measures household, not professional, expectations) but provides a useful visual reference point for the spread between informed and uninformed agents during the surge.

Long-horizon inflation expectations, selected measures
Figure 1: Long-horizon inflation expectations, selected measures

Source: Bruegel based on survey measures (ECB SPF 5yr, Philadelphia Fed SPF 10yr) versus market-based measures (FRED T5YIFR for the US and the 5y5y inflation swap for the euro area) and the NY Fed SCE median household one-year expectation. Note: the SCE series rose sharply during the surge while professional long-horizon measures reacted less, illustrating the informed-versus-uninformed gap documented by Coibion and Gorodnichenko (2015).

Somewhat surprisingly, in the euro area, the swap-based measure also showed much greater variations than the survey-based measure.

Actual inflation. US CPI-U year-on-year (FRED: CPIAUCSL); euro-area HICP all-items year-on-year (ECB SDW). Core inflation (for Section 6): US CPI less food and energy (FRED: CPILFESL); HICP excluding energy, food, alcohol and tobacco (ECB SDW: ICP.M.U2.N.XEF000.4.ANR), i.e. the ECB’s HICPX core measure used in its projections and SPF.

The specific regression methodology, including the estimation of the dynamic local-projection responses, is presented in Appendix B.

5. Main results

sec:results

A first look

Our central finding is that the sensitivity of forecast to actual inflation across horizons falls sharply as the horizon lengthens, but it does so differently in the two jurisdictions. We take the horizons in turn, beginning with the one-year horizon, before turning to the two- and five-year horizons. Table 1 reports HAC (heteroscedasticity- and autocorrelation-consistent) pass-through coefficients between actual and expected inflation at the one-year horizon.

Table 1: Pass-through coefficient β̂: one-year-ahead expectations (HAC SE)

PeriodUnited States (Philadelphia Fed SPF)Euro area (ECB SPF)N
Pre-crisis (2015–2019)0.091***
(0.012)
0.210***
(0.032)
60
COVID (2020–mid 2021)0.138***
(0.018)
0.182***
(0.018)
18
Inflation surge (mid 2021–2022)0.229***
(0.031)
0.453***
(0.080)
18
Post-peak (2023–2025)0.024
(0.024)
0.138***
(0.021)
30

Source: Bruegel. Note: β̂ from OLS (Equation 1). Newey-West HAC standard errors (12 lags) in parentheses. ***p<0.01, **p<0.05, *p<0.10.

At one year, euro-area pass-through more than doubled during the surge (0.210 → 0.453), and the US coefficient also rose markedly (0.091 → 0.229). The ECB’s increase is larger in absolute terms (0.243 percentage points versus 0.138 pp), and, more to the point, the surge-period gap between the two coefficients is itself statistically significant: pooling the two jurisdictions in a single regression with a jurisdiction-by-inflation interaction, euro-area pass-through exceeds the US by 0.224 (HAC SE 0.090, p=0.018). The euro-area coefficient already stood above the US before the surge (a gap of 0.119, p<0.01), but the gap widened during the surge. Economically, a 10 percentage-point rise in actual HICP was associated with a 4.5 pp upward revision to ECB one-year SPF expectations during the surge, compared with a 2.3 pp revision to Philly SPF expectations for the Fed. ECB one-year anchoring deteriorated more than that of the Fed.

The euro-area surge estimate is partly concentrated in the sharp Q4-2022 observation: excluding it lowers the coefficient from 0.453 to 0.384, still highly significant, whereas the US estimate is essentially unchanged (0.229 to 0.226). The euro-area increase is therefore larger but partly spike-driven, while the US increase, though smaller, is more robust across quarters.

Two readings of the larger euro-area response can be separated. To test whether it reflects genuine de-anchoring above target rather than a normalisation from the pre-existing below-target undershoot, we let the one-year pass-through differ by regime, interacting current inflation with an above-target indicator Dt = 1\πt ≥ 2%\: Et[πt+1] = α + αD Dt + β πt + γ (Dt πt) + εt. Intuitively, β is the one-year pass-through when inflation runs below the two percent target and β + γ is the pass-through when it runs above; γ thus measures whether forecasters extrapolate current inflation more once it exceeds target. A positive γ is the signature of de-anchoring above two percent, whereas γ ≈ 0 means the response is symmetric and the surge-period rise is better read as a normalisation from the prior undershoot. For the euro area, the differential is small and statistically insignificant (γ = 0.04, p=0.51): one-year pass-through is no larger when inflation runs above target than below it, which is what a normalisation interpretation predicts and is hard to reconcile with de-anchoring to above-target inflation. For the US, the differential is positive and marginally significant (γ = 0.07, p=0.06). The euro-area one-year result is therefore in part a correction of the prior below-target undershooting, rather than unambiguous de-anchoring above two percent, a qualification we carry into the discussion.

Table 2 confirms the break statistically. At one year, the ECB pre-crisis versus surge break is highly significant (F=22.7, p<0.001); the US break is also significant (F=7.6, p=0.001), but roughly three times smaller, consistently with a milder short-run effect for the Fed than for the ECB.

Table 2: Chow F-tests for structural break in β̂ — one-year horizon

BreakUnited StatesEuro area
Pre-crisis vs surge7.56***22.65***
Pre-crisis vs COVID24.83***13.21***
Surge vs post-peak7.74***21.39***

Source: Bruegel. Note: F-statistic (H0: stable β̂ across sub-samples). ***p<0.01, **p<0.05, *p<0.10.

Figure 2 shows the time series and makes the impact of actual on expected inflation visually clear: the one-year ECB SPF line tracks actual HICP upward from mid-2021, while the two-year line remains near two percent throughout. Figure 3 plots the same information as scatters by sub-period, and Figure 4 traces the dynamics via rolling 24-month correlations.

Actual versus expected inflation at the one-year and two-year horizons, 2015–2025
Figure 2: Actual versus expected inflation at the one-year and two-year horizons, 2015–2025

Source: Bruegel. Note: dashed vertical lines mark: COVID-19 lockdowns (March 2020), Fed pivot signal (November 2021), Ukraine invasion (February 2022), first Fed rate hike (March 2022), first ECB rate hike (July 2022). Note the near-flat ECB SPF two-year line throughout the surge (bottom right panel). Each event is labelled on the chart.

In addition to illustrating how the relationship between actual and expected inflation changes with the horizon, Figure 2 shows, in both jurisdictions, the steep rise in inflation, which peaked in 2022, and its steep fall thereafter.

Figure 3 plots the same information as scatters by sub-period. If actual inflation affected expected inflation with a unitary coefficient, the observations would lie along the 45-degree line; if there were no impact, they would scatter around the horizontal two percent line. The four panels show a weak effect of actual on expected inflation over the entire period, albeit stronger at the one-year horizon. The red dots, corresponding to observations during the inflation surge, move up from the interpolating line at the one-year horizon, especially for the ECB, but not at the two-year horizon, confirming our main result about a transitory effect of actual inflation on expectations.

Scatter plots of actual versus expected inflation by sub-period
Figure 3: Scatter plots of actual versus expected inflation by sub-period

Source: Bruegel. Note: surge observations in the euro-area one-year panel (red) move strongly toward the 45-degree line; surge observations in two-year panels remain compressed near two percent. The grey line in each panel is the pre-crisis regression fit, not a fit to the sub-period plotted.

Figure 4 traces the dynamics via rolling 24-month correlations. The chart reveals an important asymmetry in the speed of re-anchoring: US one-year rolling correlations returned to pre-crisis levels by early 2023, whereas ECB one-year correlations remained elevated through 2024, two years after the inflation peak.

Rolling 24-month Pearson correlations between actual and expected inflation
Figure 4: Rolling 24-month Pearson correlations between actual and expected inflation

Source: Bruegel. Note: the slower ECB one-year recovery (top right) is notable. The US correlations fall towards zero at the end of the sample because both actual and expected US inflation were nearly flat after 2023, leaving little common variation inside the 24-month window, while euro-area inflation was still falling.

Two features of the pre-surge baseline matter when interpreting the surge-period estimates. First, Table 3 shows that ECB SPF expectations were persistently below target before the surge at all horizons: one-year at 1.40 percent (-0.60 pp versus target), five-year at 1.81 percent (-0.19 pp). Part of the upward swing in one-year expectations from 1.40 percent to 2.91 percent during the surge was therefore a normalisation toward target rather than de-anchoring above it. The effective pass-through from the below-target baseline is even larger than the headline coefficient suggests. Second, US Philly one-year expectations were slightly above target pre-crisis (2.20 percent), implying a smaller absolute swing. The post-peak data confirm a durable improvement for the ECB: five-year expectations now stand at 2.06 percent (2023–2025), above the pre-crisis below-target of 1.81 percent, consistent with the July 2021 symmetric target revision having been internalised by professional forecasters.

Table 3: Mean inflation expectations by sub-period (deviation from 2% target)

PeriodECB SPF 1yrECB SPF 2yrECB SPF 5yrUS Philly 1yr
Pre-crisis1.40% (-0.60)1.61% (-0.39)1.81% (-0.19)2.20% (+0.20)
COVID1.20% (-0.80)1.41% (-0.59)1.68% (-0.32)2.09% (+0.09)
Inflation surge2.91% (+0.91)1.97% (-0.03)2.00% (≈ 0)3.07% (+1.07)
Post-peak2.24% (+0.24)2.05% (+0.05)2.06% (+0.06)2.50% (+0.50)

Source: Bruegel. Note: data from the monthly interpolated series. Deviations from the 2 percent target in parentheses.

Mean inflation expectations by sub-period
Figure 5: Mean inflation expectations by sub-period

Source: Bruegel. Note: black dashed line marks the two percent target. The large swing in ECB SPF one-year (from 1.40 percent to 2.91 percent) contrasts sharply with the near-zero movement in ECB SPF five-year (from 1.81 percent to 2.00 percent).

Regression results for the two- and five-year horizons: medium-term and destination credibility

The regression results for the two-year horizon confirm the significantly different impact of actual on expected inflation discussed in the previous subsection. Both central banks show similarly modest pass-throughs throughout, around 0.08–0.10 during the surge and never above 0.15 in any sub-period (Table 4). The ECB’s two-year coefficient during the surge (0.102) is essentially indistinguishable from the Fed’s (0.080). Forecasters who assumed the ECB would not act within 12 months remained confident it would restore price stability within 24. Table 5 confirms that, while the two-year coefficients do shift across sub-periods, the direction is toward further anchoring (falling coefficients) rather than drift.

Table 4: Pass-through coefficient β̂: two-year-ahead expectations (HAC SE)

PeriodUnited States (Philadelphia Fed SPF)Euro area (ECB SPF)
Pre-crisis (2015–2019)0.026
(0.019)
0.084***
(0.016)
COVID (2020–mid 2021)0.076***
(0.004)
0.095***
(0.019)
Inflation surge (mid 2021–2022)0.080***
(0.014)
0.102***
(0.005)
Post-peak (2023–2025)0.000
(0.008)
0.038***
(0.007)

Source: Bruegel. Note: see Table 1.

Table 5: Chow F-tests for structural break in β̂: two-year horizon

BreakUnited StatesEuro area
Pre-crisis vs surge4.80**17.37***
Pre-crisis vs COVID15.76***40.62***
Surge vs post-peak12.66***131.34***

Source: Bruegel. Note: see Table 2.

Table 6 extends the comparison to the longer horizon, and it is here that the cross-jurisdiction comparison reverses. For the euro area, we use the ECB SPF longer-term (five-year) series; for the US we must use, given the absence of a five-year series, the Philadelphia Fed SPF ten-year CPI forecast, available since 1991Q4. The horizon mismatch (ten years versus approximately four to five years) calls for caution in level comparisons between the US and European results, but the within-jurisdiction variation across sub-periods is directly informative.

Table 6: Pass-through coefficient β̂: longer-horizon expectations (HAC SE)

PeriodEuro area (ECB SPF 5yr)United States (Philly SPF 10yr)
Pre-crisis (2015–2019)0.012
(0.009)
0.040***
(0.011)
COVID (2020–mid 2021)0.011
(0.008)
0.068***
(0.003)
Inflation surge (mid 2021–2022)0.040***
(0.003)
0.105***
(0.015)
Post-peak (2023–2025)0.018***
(0.002)
0.026**
(0.010)

Source: Bruegel. Note: Newey-West HAC standard errors (12 lags) in parentheses. ECB SPF longer-term targets approximately four calendar years ahead. Philadelphia Fed SPF ten-year CPI forecast available from 1991Q4. ***p<0.01, **p<0.05, *p<0.10.

Three observations stand out:

During the inflation surge ECB five-year pass-through was 0.040 (statistically significant but economically very small), producing a ratio of approximately 11:1 between one-year and five-year pass-through for the ECB: a 10 pp rise in actual HICP was associated with a 4.5 pp revision to one-year ECB expectations, but only a 0.4 pp revision to five-year expectations.

US ten-year pass-through during the surge (0.105) was nearly three times the ECB five-year estimate. This result must be interpreted with caution: the comparison between the Fed and the ECB is affected by the different horizons of the long-term expectations, five against ten years. For instance, ten-year inflation forecasts in the US may be impacted by the fear of pressure on the Fed to ease monetary policy in order to favour the funding of loose fiscal policy. The horizon asymmetry, however, does not affect our main result: within a given jurisdiction, the pass-through falls sharply as we move from one year to the longer horizons. Subject to this caveat, US professional forecasters revised their long-run inflation outlooks more than their ECB counterparts. This reversal of the short-horizon pattern (for which the ECB showed higher pass-through) is a central contribution of this paper and gives the speed-versus-destination credibility distinction its empirical content: the ECB risked more short-term credibility than the Fed but maintained better long-run credibility.

The coefficients for both central banks converge to similar low values in the post-peak period (0.018 and 0.026), consistent with genuine re-anchoring at long horizons once the inflation shock had passed.

Figure 6 shows the five-year ECB SPF series in the left panel: it barely moved from its pre-crisis level of 1.81 percent even as HICP reached double digits. The right panel shows the US counterpart, the Philadelphia SPF ten-year series, which rose visibly during the surge before settling above its pre-crisis level, the visual counterpart of the higher US pass-through in Table 6. Figure 1 in Section 4 places all long-horizon survey series alongside market-based expectations for reference.

Long-horizon expectations versus actual inflation: ECB SPF five-year (left) and Philadelphia Fed SPF ten-year (right), 2015–2025
Figure 6: Long-horizon expectations versus actual inflation: ECB SPF five-year (left) and Philadelphia Fed SPF ten-year (right), 2015–2025

Source: Bruegel. Note: the euro-area five-year series (left) remained essentially flat near two percent throughout the surge, providing the primary evidence for maintained ECB institutional credibility. The US ten-year series (right) rose during the surge and settled above its pre-crisis level.

6. Robustness checks

sec:robust

Energy composition and core inflation pass-through

There is a natural alternative hypothesis to our interpretation that the ECB’s greater one-year pass-through during the surge reflected a difference in attitude between the ECB and the Fed. Instead, the higher pass-through could have been caused by the euro area’s more acute exposure to energy-price shocks. The euro area imported about 40 percent of its natural gas from Russia before 2022, whereas the US had been a net oil exporter since 2019. If professional forecasters anticipated euro-area energy prices remaining elevated while US prices normalised quickly, a higher ECB pass-through could be a rational response to energy-market fundamentals, rather than a signal of a difference in the ECB’s reaction function.

We test this by replacing headline inflation with core inflation (HICP/CPI excluding food, energy, alcohol and tobacco) in Equation (1). If energy drove the headline result, core pass-through should be materially smaller. The opposite is true. During the inflation surge, euro-area core pass-through was approximately β̂core ≈ 0.92, roughly double the headline estimate of 0.45. This finding is economically intuitive: the `transitory’ narrative was harder to maintain when underlying (non-energy) prices were rising, because rising core inflation is consistent with second-round effects rather than a pure commodity shock. Professional forecasters revised their one-year ECB expectations upward much more aggressively when core inflation rose than when headline inflation rose, suggesting they attributed headline inflation partly to temporary energy factors but treated core inflation as a signal of the ECB’s policy stance. Energy-import exposure is thus not the driver of the different impact between the Fed and the ECB. Energy nonetheless enters core inflation indirectly, through energy-intensive components such as transport services. Neri et al (2023) put this indirect contribution at around 150 basis points in the fourth quarter of 2022 on their baseline VAR, and at up to 250 basis points on their Phillips curve specification. This does not overturn the comparison, since the same indirect channel operates in both jurisdictions, but it qualifies any literal reading of core inflation as energy-free. Figure 7 shows the headline-versus-core comparison across sub-periods.

Pass-through β̂ for headline versus core inflation, one-year-ahead expectations
Figure 7: Pass-through β̂ for headline versus core inflation, one-year-ahead expectations

Source: Bruegel. Note: during the surge (third bar group), euro-area core pass-through (≈ 0.92, red) exceeds headline (0.45, blue), ruling out energy import exposure as the driver of the ECB credibility gap.

Structural stability: data-driven break detection and dynamic transmission

The Chow statistic in Tables 2–5 tests pre-specified narrative dates chosen based on the ECB’s and Fed’s observable communications. To complement this with a fully endogenous approach, we apply Bai-Perron Dynp (exact solution, k=3) to the one-year expected inflation series for each jurisdiction, allowing the data to select break dates without prior conditioning.

For the euro area, Dynp detects breaks in March 2021, June 2022 and September 2023. For the US, the dates are March 2016 (a minor baseline realignment), August 2021 and November 2022. The key finding is the date of the first surge-related break: the ECB one-year impact began shifting in March 2021, five months before the Fed’s equivalent break in August 2021. Forecasters in the euro area started revising their expectations well before their US counterparts. This timing is important because both economies faced the same global inflationary shock (commodity and supply-chain pressures were simultaneous). The difference in break dates therefore reflects a difference in how quickly central bank communications reduced uncertainty about the policy response. The Fed’s November 2021 pivot signals, which the ECB did not replicate until later (see Section 7), were precisely what better anchored US professional short-term expectations.

We use two algorithms as a deliberate cross-check rather than for redundancy: PELT, which selects the number of breaks endogenously through a BIC-type penalty, recovers essentially the same dates as Dynp, confirming that the March 2021 euro-area break was not an artefact of fixing k=3. Two checks support the narrative windows. First, the data-driven breaks fall close to the boundaries we impose, though not exactly on them: the first surge-related break is dated March 2021 for the euro area, four months before the surge window opens, and August 2021 for the US, one month after it opens. Second, the surge coefficient does not depend on which quarters make up the window: dropping any single quarter from the surge period leaves it highly significant, between 0.38 and 0.57 for the euro area and between 0.18 and 0.27 for the US. Figure 8 overlays narrative and data-driven break dates on the series.

Structural break detection: narrative dates versus Bai-Perron <em>Dynp</em> and PELT” class=”wp-image-3213″/><figcaption class=Figure 8: Structural break detection: narrative dates versus Bai-Perron Dynp and PELT

Source: Bruegel. Note: narrative dates = orange dashed lines; Bai-Perron Dynp = green dash-dot lines; PELT = pink dotted lines. The ECB’s first surge-related break was in March 2021, five months before the Fed’s in August 2021.

Local projections (Jordà, 2005) estimate, through a separate regression at each horizon, the average response of expectations in the months following a one-standard-deviation inflation shock. Unlike a VAR, they require no model of the full joint dynamics. The local projections, specified in Appendix B of the working paper, trace the full dynamic profile of the expectation response to an inflation shock, providing a continuous complement to the sub-period Chow tests. Figure 9 reports the impulse response functions with 90 percent HAC confidence bands. The US one-year peak response (β̂7=0.112 at h=7 months) was reached earlier and was smaller than the euro-area peak (β̂13=0.173 at h=13 months). Not only was the ECB’s response larger, it was also slower to materialise and broader in duration: the US one-year response decayed toward zero by h=18 months while the euro-area response remained elevated through h=20. At the two-year horizon (bottom panels of Figure 9), both responses were smaller and faster-decaying, confirming that medium-term anchoring holds in the local projections framework as it does in the sub-period OLS. The six-month delay in the euro-area peak is consistent with the Bai-Perron evidence: break dates differ by five months and local projections show peak responses differing by six months, giving two independent and numerically similar estimates of the ECB’s communication lag.

Local projections impulse response functions: change in expectations h months after a 1 s.d. inflation shock (90% HAC CI bands)
Figure 9: Local projections impulse response functions: change in expectations h months after a 1 s.d. inflation shock (90% HAC CI bands)

Source: Bruegel. Note: euro-area one-year peak at h=13 (top right); US peak at h=7 (top left). Two-year panels (bottom) show fast-decaying, small responses for both jurisdictions.

Is long-horizon anchoring genuine? The long-short sensitivity test

Table 7 reports the long-short sensitivity coefficient δ̂ from Equation (2). Intuitively, the test asks whether revisions to one-year expectations spill over into five-year expectations. If they do not (δ = 0), the long-run anchor is governed by a process separate from the short-run path, a signature of genuine anchoring rather than of forecasters mechanically mean-reverting their long-run forecasts towards two percent.

We estimate it for the euro area using matched ECB SPF series (five-year on one-year). For the US, no professional-forecaster survey publishes a five-year forward rate at monthly frequency. The Philadelphia SPF does report a ten-year CPI outlook, which we use in Section 5 for the pass-through comparison, but it is not the appropriate instrument for the sensitivity test, which requires a pure long-forward measure responding symmetrically to short-run revisions. We therefore use the FRED T5YIFR (five-year forward five-year TIPS breakeven) as the US long-horizon proxy and regress it on Philadelphia SPF one-year expectations. The instrument mismatch, market-based for the US against survey-based for the euro area, has a predictable bias direction: TIPS break-evens incorporate an inflation risk premium that co-moves positively with short-run inflation news (Gürkaynak et al, 2005), mechanically inflating δ̂ for the US relative to a matched-survey estimate. This affects our ECB–Fed comparison, which should be interpreted with care. The ECB result, based on fully matched ECB SPF instruments, is unaffected by this limitation.

Table 7: Long-short sensitivity (Gürkaynak-Levin-Swanson test): ΔE[π5yr] ΔE[π1yr]

PeriodEuro area (ECB SPF)United States (T5YIFR Philly 1yr)
Pre-crisis (2015–2019)0.040
(0.032)
0.643*
(0.324)
COVID (2020–mid 2021)0.020
(0.029)
0.745***
(0.202)
Inflation surge (mid 2021–2022)-0.002
(0.006)
0.196**
(0.086)
Post-peak (2023–2025)-0.006
(0.006)
-0.028
(0.091)

Source: Bruegel. Note: δ̂ from OLS with HAC SE (12 lags). Euro area: ECB SPF five-year on ECB SPF one-year (matched survey instruments). US: FRED T5YIFR (five-year forward five-year breakeven) on Philadelphia Fed SPF one-year (mixed instruments; breakeven includes risk premia). ***p<0.01, **p<0.05, *p<0.10.

The euro-area results are unambiguous. In every sub-period, including the inflation surge, δ̂ is statistically indistinguishable from zero. During the surge, the point estimate is -0.002 with a standard error of 0.006: as ECB one-year SPF forecasts rose from 1.4 percent to 2.9 percent, euro-area five-year SPF forecasts showed literally zero co-movement. This is not consistent with a mechanical mean-reversion story. A forecaster who systematically revises one-year expectations upward by 1.5 percentage points while keeping five-year expectations unchanged is expressing a genuine belief in the transitory nature of inflation, not just reporting the known target. The precision of the zero estimate (the upper bound of the 95 percent confidence interval is approximately 0.010) strengthens this conclusion: we can rule out economically meaningful long-horizon sensitivity with high confidence.

The US results tell a different story. The FRED T5YIFR is significantly and positively sensitive to Philly one-year revisions in all non-post-peak sub-periods (surge: δ̂=0.196, significant at five percent). As noted above, part of this positive signal reflects the inflation risk premium embedded in TIPS breakevens. But the Philadelphia SPF ten-year pass-through result from Table 6 (0.105 during the surge versus 0.040 for the ECB five-year) provides independent survey-based corroboration that the US long-horizon did in fact experience greater drift. The two pieces of evidence thus point in the same direction: the T5YIFR’s δ̂ is inflated by risk premia, but the underlying movement in actual US professional forecasters’ long-run expectations was also larger than the ECB’s, consistent with greater long-run drift, rather than purely a measurement artefact.

The forecast-error test (Equation 3) provides a final piece of evidence. During the ECB surge, γ̂=-0.005 (p<0.01, negative). When actual HICP inflation exceeded the SPF forecasters’ own one-year-ahead forecast made twelve months earlier, five-year ECB SPF expectations fell slightly. Forecasters who observed that inflation had overshot their own projections did not revise long-run expectations upward; if anything, they interpreted the overshoot as transitory and revised long-run expectations marginally downward. This rules out the interpretation that the zero long-short sensitivity was driven by the inertia of forecasters who simply did not update: those same forecasters were actively revising in response to new data, but in a direction inconsistent with a permanent regime shift. Figures 10 and 11 complete the picture. Figure 11 measures how far individual forecasters disagreed with one another, using the cross-sectional interquartile range (IQR) of their point forecasts.

Long-short sensitivity (top panels) and forecast-error pass-through (bottom panels) for the euro area (left) and US (right)
Figure 10: Long-short sensitivity (top panels) and forecast-error pass-through (bottom panels) for the euro area (left) and US (right)

Source: Bruegel. Note: 90% HAC confidence intervals. Near-zero euro-area coefficients in all sub-periods confirm genuine long-horizon anchoring; negative forecast-error coefficients during the surge indicate forecasters interpreted above-target inflation as transitory.

Cross-sectional IQR of one-year and two-year inflation expectations (ECB SPF and Philadelphia Fed SPF), 2015–2025
Figure 11: Cross-sectional IQR of one-year and two-year inflation expectations (ECB SPF and Philadelphia Fed SPF), 2015–2025

Source: Bruegel. Note: disagreement rose sharply during the surge, particularly at one year for the ECB, so the stable mean expectations reported in Tables 1 and 4 coexisted with a wider dispersion of individual views.

Market-based long-horizon expectations: the euro-area 5y5y inflation swap

For a market-based cross-check we use the euro-area five-year/five-year forward inflation swap, the most closely watched gauge of long-horizon euro-area expectations during the period when Mario Draghi was ECB President (2011–2019). At a horizon matched to the US T5YIFR (both are five-year/five-year forwards, unlike the survey comparison of an ECB five-year against a Philadelphia ten-year), the pass-through of current HICP inflation into the swap during the surge is 0.081 (HAC SE, p<0.01), against 0.029 for the T5YIFR. Taken at face value, this reverses the survey-based ranking, under which the ECB SPF five-year coefficient (0.040) sits well below the Philadelphia ten-year (0.105).

The reversal is, however, largely a risk-premium artefact. Decomposing the swap into the survey-based expectation and a residual wedge (the inflation-risk and liquidity premium, defined as the gap between the swap and the ECB SPF five-year measure), pass-through is additive by construction: because the swap equals the survey measure plus the wedge, regressing each on current inflation gives βswap = βsurvey + βwedge, and during the surge the split is 0.081 = 0.040 + 0.041, so half the market response is for premium rather than expected inflation. The wedge level itself swung from -0.45 pp in 2020–2021 (a deflation premium) to +0.23 pp post-peak, consistent with markets repricing inflation risk on a contract whose euro-area liquidity is notoriously thin.

A survey-anchored state-space model that lets the swap and the survey carry independent measurement error reproduces this decomposition exactly (the model-implied premium correlates with the wedge at 1.00), which confirms its robustness but adds no separate identification: a genuine no-arbitrage split of expected inflation from premium would require the euro-area real (inflation-linked) yield curve, which the ECB does not publish through its SDMX interface and which we leave to future work. The matched-instrument long-short test points the same way: during the surge, the swap shows no sensitivity to one-year revisions (δ̂ = -0.00) while the T5YIFR does (δ̂ = 0.20, p<0.05). We can clean only the euro-area side. The US T5YIFR carries the same inflation-risk and liquidity premium, but with no US five-year-ahead survey to subtract, we cannot decompose it, so the 0.029 figure is itself premium-contaminated and neither raw market measure is premium-free. We therefore read the swap evidence as consistent with the survey-based conclusion once the premium is accounted for, while noting that the cleanest cross-jurisdiction comparison remains survey-to-survey.

7. Discussion

sec:discussion

The complete empirical picture supports a single coherent narrative that can be stated precisely. The ECB did not lose focus on its long-run commitment to the two percent target: professional forecasters never stopped believing euro-area inflation would eventually return to two percent, and the five-year ECB SPF series barely moved throughout the surge. However, there was no expectation that the ECB would act quickly enough to deliver near-two-percent inflation within the next twelve months. We label this the distinction between speed credibility (the perceived willingness to respond promptly) and destination credibility (the perceived commitment to the long-run target). While this distinction was never tested during the pre-surge decade of below-target inflation, the 2021–2022 episode exposed it.

Three pieces of evidence crystallise the communication-lag interpretation. First, the 11:1 ratio of one-year to five-year ECB pass-through during the surge (0.453 versus 0.040) is too large to be explained by a general loss of confidence: a forecaster who doubted the ECB’s fundamental commitment to two percent would have revised five-year expectations upward, not kept them flat. Second, the Bai-Perron break in ECB one-year expectations occurred in March 2021, five months before the Fed’s break in August 2021 and well before the ECB’s first rate increase (July 2022). The global inflationary shock was simultaneous; the policy response was not. Professional forecasters sensed the divergence in central-bank communication in real time. Third, local projections show that the euro-area one-year response to an inflation shock peaked at h=13 months compared to h=7 months for the US, independently replicating the five-month communication lag that Bai-Perron dates from the series level.

The below-target baseline further shapes the interpretation of these numbers. ECB SPF one-year expectations stood at 1.40 percent in the pre-crisis period, 0.60 percentage points below the symmetric two percent target adopted in July 2021. This is consistent with Paloviita et al (2021) and implies that the ECB’s credibility was calibrated to an implicit below-two-percent target. When the symmetric target was announced, it was immediately tested from above. The apparent de-anchoring was in part a correction of a previous misalignment. The post-peak five-year ECB expectation of 2.06 percent (versus the pre-crisis 1.81 percent) suggests the symmetric target revision may have been internalised by professional forecasters, a genuine improvement in long-run anchoring relative to the pre-crisis baseline.

The US comparison adds a dimension that previous analyses lacked. US one-year credibility eroded less, though not negligibly: the Philadelphia SPF pre-crisis versus surge break is statistically significant but roughly three times smaller than the ECB’s (Chow F=7.6 versus 22.7; euro-area pass-through exceeds the US by 0.224 during the surge, p=0.018), but US ten-year credibility moved more than ECB five-year credibility during the surge (Philly SPF 10yr β̂=0.105 versus ECB SPF 5yr β̂=0.040). The Fed better preserved the public’s confidence in its short-term responsiveness, but allowed greater long-run drift in professional expectations. The ECB pattern was the mirror image.

This asymmetry reflects differences in institutional architecture and in the structure of the pre-existing credibility, and the two dimensions work in opposite directions. On the speed dimension, the Fed’s advantage is clear: the Federal Open Market Committee can reach decisions more quickly than the ECB Governing Council, no formal sequencing commitment between asset purchases and rate decisions prevailed, and the Fed communication culture was accustomed to rapid pivots. The slowness to pivot is sometimes read as a structural ECB trait. The evidence is mixed: the ECB was indeed slower than the Fed to begin tightening in 2022, but it was not uniformly slower to turn, having cut rates in June 2024, three months ahead of the Fed. What our results identify is narrower and better supported: a slowness to signal a tightening pivot tied to the asset-purchase sequencing, rather than a general institutional inertia in both directions.

On the destination dimension, the ECB’s advantage is less obvious but equally real. A decade of below-target inflation led professional forecasters to expect that the ECB would struggle to bring inflation persistently to two percent from below, not that it would fail to return it to target from above. This asymmetric prior, accumulated through years of undershooting and countered by the July 2021 symmetric target revision, provided a structural long-run anchor that the Fed, whose pre-crisis expectations were already at or above target, did not possess. When the surge arrived, US professional forecasters faced genuine uncertainty about the long-run landing point: whether the Fed would tolerate a marginally higher inflation norm, whether widening structural fiscal deficits would eventually constrain the policy response or whether upward revisions to the neutral rate had shifted the long-run equilibrium above two percent. None of these concerns raised doubts about the Fed’s fundamental commitment to price stability. Rather, they reflected uncertainty about the destination that the ECB’s prior undershooting had largely pre-resolved for the euro area. The post-peak T5YIFR stabilising near 2.3 percent (versus the pre-crisis 1.95 percent) is consistent with this reading: a modest destination premium, not a credibility collapse.

A caveat about the measure frames these conclusions. We assess credibility through professional forecasters, but they are only one of the audiences a central bank must convince, and not the one to which it is formally accountable. The ECB’s mandate is price stability over the medium term and it is accountable to the European Parliament and to the European constituency more generally, not to SPF respondents. An ECB that loses speed credibility with professional forecasters while retaining it with bond markets and wage-setters may have made a defensible institutional trade-off rather than a mistake. Reis (2023) made a related point from the opposite direction, noting that professional forecasters are the least appropriate population for gauging the expectations channel of the Phillips curve. We use them precisely because our object is institutional credibility (the belief on the part of informed agents that the target will be met), rather than price-setting transmission. The speed-credibility loss we document should therefore be read as a deterioration in one specific, if informative, gauge, not as evidence that the ECB’s broader credibility was impaired.

These findings have direct implications for central-bank practice. The ECB’s commitment not to raise rates before ending net asset purchases prevented early communication of a policy pivot, exactly the forward guidance the Fed provided from November 2021. The contrast with the Bank of England is instructive: because its asset-purchase programme was defined by a total stock envelope rather than an ongoing flow, the Bank of England was able to raise rates as early as December 2021, whereas the ECB perceived itself as constrained to wind down net purchases first.2 The Bai-Perron evidence suggests professional forecasters identified this constraint in March 2021. State-contingent language tied to observed inflation paths rather than predetermined policy sequences would restore the flexibility to signal a change in the reaction function before the asset purchase programme formally ends.

More broadly, the core pass-through finding argues for greater transparency about core inflation forecasts: professional forecasters were more responsive to core than to headline inflation during the surge, meaning they saw through energy factors, but not through underlying price pressures. Central-bank communication that actively distinguishes its assessment of transitory versus persistent inflation components would reduce the information gap that allowed the `transitory’ narrative to persist longer than the core data warranted. The slow post-peak re-anchoring at one year (visible in Figure 4) suggests that this communication lesson remains relevant even after the inflation surge passed.

One issue we do not cover in this paper, but which is germane to its topic, is the evidence that both the ECB and the Fed have systematic problems in forecasting inflation, which affects their interest-rate policies.

8. Conclusion

sec:conclusion

The 2021–2022 episode was a credibility test, and the results depend critically on which horizon is used. At the one-year horizon, short-term expectations drifted in both jurisdictions, but the euro-area break was significantly larger than the US one (a surge-period pass-through gap of 0.224, p=0.018); the break in the sensitivity of expected to actual inflation occurred five months earlier in the euro area than in the US, and re-anchoring was slow. At the two-year horizon, the ECB performed comparably to the Fed, consistent with its medium-term mandate. At longer horizons, the comparison reverses: ECB five-year institutional credibility was essentially unimpaired throughout (pass-through 0.04 during the surge), while US ten-year professional expectations showed greater drift (pass-through 0.11). The ECB lost more speed credibility but maintained destination credibility; the Fed lost less speed credibility but experienced more long-run drift.

The long-short sensitivity test confirms that the ECB five-year result is not mechanical: even at the peak of the surge, ECB five-year expectations showed zero statistical sensitivity to one-year revisions (δ̂=-0.002), and the forecast-error pass-through was slightly negative. Informed professional observers believed the surge was transitory and expected a return to below-target levels at the five-year horizon. Data-driven structural break detection independently dates the ECB’s short-term increase in the impact of actual on expected inflation to March 2021, and local projections independently recover the same six-month communication lag. Core inflation pass-through during the surge (≈ 0.92) exceeded headline pass-through (0.45), ruling out energy-import exposure as the driver.

The broader lesson is that credibility assessment requires a time dimension. Short-term and long-term impact of actual on expected inflation are distinct phenomena, affected by different institutional features and calling for different remedies. The ECB protected its long-term anchor while not forcing an early return of inflation expectations to its objective. The priority before the next shock is not to strengthen the ECB’s commitment to the two percent target (that commitment was not in doubt) but to reform how the ECB signals the speed of its reaction function, so that the next episode of above-target inflation does not require eighteen months before professional forecasters receive a credible pivot signal.

References

Afrouzi, H., M. Halac, K.S. Rogoff and P. Yared (2024) ‘Changing Central Bank Pressures and Inflation’, Brookings Papers on Economic Activity, Spring: 205–64, available at https://www. brookings.edu/articles/changing-central-bank-pressures-and-inflation/

Ahn, H.J. and L.E. Farmer (2024) ‘Disagreement about the Term Structure of Inflation Expectations’, Finance and Economics Discussion Series 2024-084, Federal Reserve Board, available at https://doi.org/10.17016/FEDS.2024.084

Baumann, U., A. Ferrando, D. Georgarakos, Y. Gorodnichenko and T. Reinelt (2025) ‘Do Inflation Expectations Become More Anchored During a Disinflation Episode? Evidence for Euro- area Firms’, AEA Papers and Proceedings 115: 266–70, available at https://doi.org/10.1257/ pandp.20251019

Bernanke, B.S. and O.J. Blanchard (2023) ‘What Caused the US Pandemic-Era Inflation?’ NBER Working Paper 31417, National Bureau of Economic Research, available at https://www.nber. org/papers/w31417

Bernanke, B.S. and O.J. Blanchard (2024) ‘An Analysis of Pandemic-Era Inflation in 11 Economies’, Hutchins Center Working Paper 91, The Brookings Institution, available at https://www. brookings.edu/articles/an-analysis-of-pandemic-era-inflation-in-11-economies/

Bianchi, F., R. Faccini and L. Melosi (2023) ‘A Fiscal Theory of Persistent Inflation’, Quarterly Journal of Economics 138(4): 2127–79, available at https://doi.org/10.1093/qje/qjad027

Blinder, A.S. (1997) ‘What Central Bankers Could Learn from Academics, and Vice Versa’, Journal of Economic Perspectives 11(2): 3–19, available at https://doi.org/10.1257/jep.11.2.3

Burban, V., B. De Backer and A.L. Vladu (2024) ‘Inflation (de-)anchoring in the euro area’, Working Paper 2964, European Central Bank, available at https://dx.doi.org/10.2139/ssrn.4912626

Coibion, O. and Y. Gorodnichenko (2015) ‘Information Rigidity and the Expectations Formation Process: A Simple Framework and New Facts’, American Economic Review 105(8): 2644–78, available at https://doi.org/10.1257/aer.20110306

Coibion, O. and Y. Gorodnichenko (2025) ‘Inflation, Expectations and Monetary Policy: What Have We Learned and to What End?’ NBER Working Paper 33858, National Bureau of Economic Research, available at https://doi.org/10.3386/w33858

Corsello, F., S. Neri and A. Tagliabracci (2021) ‘Anchored or de-Anchored? That Is the Question’, European Journal of Political Economy 69: 102031, available at https://doi.org/10.1016/j. ejpoleco.2021.102031

D’Acunto, F., Y. Gorodnichenko, O. Coibion and M. Weber (2022) ‘The Subjective Inflation Expectations of Households and Firms: Measurement, Determinants, and Implications’, Journal of Economic Perspectives 36(3): 157–84, available at https://doi.org/10.1257/jep.36.3.157

Dovern, J. and G. Kenny (2020) ‘Anchoring Inflation Expectations in Unconventional Times: Micro Evidence for the Euro-area’, International Journal of Central Banking 16(5): 309–47, available at https://www.ijcb.org/journal/v16n5/anchoring-inflation-expectations-unconventional-times- micro-evidence-euro-area

Erceg, C.J. and A.T. Levin (2003) ‘Imperfect Credibility and Inflation Persistence’, Journal of Monetary Economics 50(4): 915–44, available at https://doi.org/10.1016/S0304- 3932(03)00036-9 Working Paper | Issue n˚22/26 | August 2026

Gennaioli, N., M. Leva, R. Schoenle and A. Shleifer (2024) ‘How Inflation Expectations De-Anchor: The Role of Selective Memory Cues’, NBER Working Paper 32633, National Bureau of Economic Research, available at https://doi.org/10.3386/w32633

Gürkaynak, R.S., A.T. Levin and E.T. Swanson (2010) ‘Does Inflation Targeting Anchor Long-Run Inflation Expectations? Evidence from the US, UK, and Sweden’, Journal of the European Economic Association 8(6): 1208–42, available at https://doi.org/10.1111/j.1542-4774.2010. tb00553.x

Gürkaynak, R.S., B. Sack and E.T. Swanson (2005) ‘The Sensitivity of Long-Term Interest Rates to Economic News: Evidence and Implications for Macroeconomic Models’, American Economic Review 95(1): 425–36, available at https://doi.org/10.1257/0002828053828446

Jordà, Ò. (2005) ‘Estimation and Inference of Impulse Responses by Local Projections’, American Economic Review 95(1): 161–82, available at https://doi.org/10.1257/0002828053828518

Levin, A.T., F.M. Natalucci and J.M. Piger (2004) ‘The Macroeconomic Effects of Inflation Targeting’, Federal Reserve Bank of St. Louis Review 86(4): 51–80, available at https://doi. org/10.20955/r.86.51-80

Łyziak, T. and M. Paloviita (2017) ‘Anchoring of Inflation Expectations in the Euro-area: Recent Evidence Based on Survey Data’, European Journal of Political Economy 46: 52–73, available at https://doi.org/10.1016/j.ejpoleco.2016.11.001

Natoli, F. and L. Sigalotti (2018) ‘Tail Co-Movement in Inflation Expectations as an Indicator of Anchoring’, International Journal of Central Banking 14(1): 35–71, available at https://www.ijcb. org/journal/v14n1/tail-co-movement-inflation-expectations-indicator-anchoring

Neri, S., G. Bulligan, S. Cecchetti, F. Corsello, A. Papetti, M. Riggi, C. Rondinelli and A. Tagliabracci (2022) ‘On the Anchoring of Inflation Expectations in the Euro-area’, Occasional Papers 712, Banca d’Italia, available at https://www.bancaditalia.it/pubblicazioni/qef/2022-0712/index.html

Neri, S., F. Busetti, C. Conflitti, F. Corsello, D. Delle Monache and A. Tagliabracci (2023) ‘Energy Price Shocks and Inflation in the Euro Area’, Occasional Papers 792, Banca d’Italia, available at https://www.bancaditalia.it/pubblicazioni/qef/2023-0792/index.html

Orphanides, A. and J.C. Williams (2005) ‘Imperfect Knowledge, Inflation Expectations, and Monetary Policy’, in B.S. Bernanke and M. Woodford (eds) The Inflation-Targeting Debate, University of Chicago Press, available at https://www.nber.org/books-and-chapters/inflation- targeting-debate

Paloviita, M., M. Haavio, P. Jalasjoki and J. Kilponen (2021) ‘What Does “Below, but Close to, 2 Percent” Mean? Assessing the ECB’s Reaction Function with Real-Time Data’, International Journal of Central Banking 17(2): 125–69, available at https://www.ijcb.org/journal/v17n2/what- does-below-close-2-percent-mean-assessing-ecbs-reaction-function-real-time-data

Reis, R. (2023) ‘Four Mistakes in the Use of Measures of Expected Inflation’, AEA Papers and Proceedings 113: 95–99, https://doi.org/10.1257/pandp.20231033

Shapiro, A.H. (2022) ‘Decomposing Supply and Demand Driven Inflation’, Working Paper 2022-18, Federal Reserve Bank of San Francisco, available at https://www.frbsf.org/wp-content/uploads/ wp2022-18.pdf Working Paper | Issue n˚22/26 | August 2026


  1. We use the terms expectations and forecasts interchangeably. ↩︎
  2. In practice the Fed overcame a similar constraint by compressing the end of net purchases and the first rate increase into a couple of weeks. ↩︎