Why Central Banks Systematically Miss Inflation Forecasts

Central-bank inflation forecasts: systematically biased in the same direction across decades — and the empirical pattern reveals more about institutional incentives than about forecasting limits.

Twenty-five years of ECB inflation projections systematically undershot realised inflation in 2021-2023, just as Fed projections systematically overshot during the 2010-2019 deflationary scare. The errors are not random — they cluster in directions the framework would predict.

The IMF’s evaluation of central-bank forecasts (Working Paper 23/53, 2023) documented systematic forecast biases across 16 advanced-economy central banks since 2000. Underprediction during inflation surges and overprediction during disinflation are both signatures of the same underlying mechanism: forecasts are anchored to the policy framework’s implicit assumptions about how the world works.

The empirical record of central-bank forecast errors

Central-bank inflation forecasts published quarterly or semi-annually are among the most scrutinised macro variables in modern finance. They influence market positioning, fiscal planning, wage negotiations, and corporate budgeting. Their accuracy determines policy credibility. The empirical record over 25 years is uncomfortably consistent: the errors are large and persistent, and they cluster in the same direction across institutions during similar regime episodes.

The IMF Working Paper 23/53 (2023) provides the most systematic evaluation. Across 16 advanced-economy central banks since 2000, the average two-year-ahead inflation forecast error has been substantial — typically 0.5 to 1.5 percentage points in absolute terms during stable periods, and 3-7 percentage points during regime-shift episodes (2008-2009, 2021-2023). More importantly, the errors exhibit serial correlation: when a central bank under-predicts inflation in one period, it tends to under-predict in the next period as well. This is the signature of structural rather than random error.

The 2021-2023 systematic underprediction episode

The post-pandemic period produced the most visible recent example. The European Central Bank’s June 2021 staff projections forecast 2022 HICP inflation at 1.5%; the realised value was 8.4%, an error of 6.9 percentage points. The Federal Reserve’s Summary of Economic Projections from June 2021 forecast 2022 PCE inflation at 2.3%; the realised value was 6.5%, an error of 4.2 percentage points. The Bank of England’s August 2021 Inflation Report forecast 2022 CPI at 4%; the realised value was 9.0%, an error of 5 percentage points. These were not minor calibration mistakes; they were systematic mispredictions of the regime in real time.

What makes the episode diagnostic rather than merely embarrassing is that the underprediction was largely uniform across institutions and methodologies. Different central banks using different models and different data sets all underpredicted inflation in the same direction at the same time. This rules out the explanation that any single institution’s framework was responsible. The common factor is the implicit assumption embedded in inflation-targeting frameworks: that long-run inflation will return to the 2% target, and that any deviation will mean-revert. When the world delivers a regime that violates this assumption, the entire profession underpredicts in unison. Our analysis of how inflation expectations form and why they matter develops the related question of private-sector expectational anchoring.

The 2010-2019 mirror image

The opposite error operated during the 2010-2019 period. Central banks across advanced economies systematically overpredicted inflation throughout the post-Global Financial Crisis decade. The ECB’s June 2010 staff projections forecast 2012 HICP at 1.7%; the realised value was 2.5% — modest overshoot, but the pattern was repeated for nearly every year. The Fed’s Summary of Economic Projections systematically overpredicted PCE inflation from 2012 through 2019, with the median FOMC participant forecast exceeding realised inflation in nine of ten years (Federal Reserve SEP archive). The Bank of England’s MPC inflation projections showed the same overshoot pattern.

The mechanism was the inverse of 2021-2023. With the same anchoring assumption — that inflation would mean-revert to 2% — central banks expected each disinflation episode to be temporary and inflation to “normalise” back to target. Instead, structural deflationary forces (demographic, technological, globalisation) kept actual inflation persistently below 2% throughout the period. The empirical record is gathered in the regime-by-regime breakdown of inflation dynamics. Once again, different institutions made the same error in the same direction — and once again, the common factor was the implicit framework assumption. Our coverage of cost-push vs demand-pull inflation develops the diagnostic challenge of distinguishing temporary from structural inflation drivers.

The institutional anchor explanation

The pattern across both decades suggests a common mechanism: central-bank forecasts are partially generated by their target rather than purely by their data. When the target is 2% and the framework assumes mean-reversion, the forecast naturally drifts toward 2% over the medium-term horizon, regardless of what the data is currently saying. This is not a flaw in any individual model but an emergent property of inflation-targeting frameworks themselves. The forecast is a partial restatement of the policy commitment — and when reality persistently violates the commitment, the forecast persistently misses in the same direction.

Romer and Romer (2004) documented a similar pattern in their analysis of historical Fed forecasts: the Fed’s internal forecasts during the 1970s consistently underpredicted inflation when the Fed was being too accommodative, and consistently overpredicted inflation when the Fed was being too restrictive. The pattern is consistent with motivated reasoning rather than simple model error: the forecast tends to support the policy stance the institution has already adopted. This is not evidence of bad faith; it is evidence that institutional context shapes empirical perception in ways that academic frameworks cannot fully neutralise. Our examination of the Phillips curve and its limits connects to this: forecast errors and Phillips-curve flattening together suggest that central-bank models embed assumptions that may not match the structural reality.

Why the anchoring is hard to fix

The institutional response to the 2021-2023 underprediction has been incremental rather than fundamental. Central banks have updated their inflation models, expanded their judgement frameworks, increased their attention to supply-side variables. But the basic architecture — anchored expectations as the medium-term anchor, mean-reversion to target as the operational assumption — remains. The reason is partly that no superior alternative has emerged, and partly that abandoning the anchor would risk the credibility that the framework provides.

The Fed’s Flexible Average Inflation Targeting (FAIT) adopted in August 2020 was an attempt to address the 2010-2019 systematic overprediction by allowing inflation to overshoot the target temporarily after periods of undershooting. The framework was almost immediately overwhelmed by the 2021-2023 episode, exposing the asymmetric design: FAIT made the central bank systematically slower to tighten than the realised inflation justified. Our examination of why the Fed targets 2% inflation specifically develops the rationale for the chosen anchor level. The pillar piece on the complete framework of inflation mechanisms, measurement, history and effects integrates these institutional considerations.

Implications for forecast users

For market participants, fiscal planners and corporate budgeters who use central-bank forecasts as inputs, the empirical record carries operational implications. First, the forecasts are most reliable during periods of regime stability and least reliable during regime transitions — exactly the periods when accurate forecasts would be most valuable. Second, the directional bias is correlated with the policy stance of the institution: an aggressive central bank tends to overpredict inflation (justifying its tightening); an accommodative central bank tends to underpredict (justifying its easing). Third, alternative forecast sources (Survey of Professional Forecasters, market-implied breakevens) provide useful triangulation but are themselves anchored to similar frameworks.

The diagnostic discipline is to weight central-bank forecasts as one input among several rather than as authoritative predictions. Triangulating across the SPF, market measures (TIPS breakevens, inflation swaps), and structural analyses produces a more robust forecast distribution than any single source. This mechanism fits within our taxonomy of inflation regimes. The dataset on market-implied breakeven inflation rates provides one such alternative measure. Our work on how central banks calibrate rate hikes reflects the operational importance of forecast accuracy for policy timing.

What credible forecasting would require

The intellectual honest version of this analysis acknowledges that no forecasting institution — central bank, IMF, OECD, private bank — has a clean track record on inflation regime transitions. The errors are not unique to central banks; they are characteristic of macroeconomic forecasting in regimes that violate the framework’s implicit assumptions. The solution is not better models but better epistemic discipline about model limits. Atlanta Fed’s GDPNow nowcasting framework is one institutional attempt to address this: rather than producing a single point forecast, it produces a real-time updated estimate based on incoming data, with explicit acknowledgement that the estimate evolves with information.

For the analyst seeking a more durable framework, the relevant disciplines are: monitor the gap between target and realised inflation as a leading indicator of regime stress; weight forecasts inversely to recent forecast errors of the same institution; and integrate alternative sources (market-implied measures, structural macro analysis) as triangulation. Our analysis of core vs headline inflation develops the measurement layer that interacts with forecast accuracy. The dataset on U.S. inflation history since 1913 provides the long-run perspective that puts any current forecast into structural context.

The diagnostic posture

The honest analytical takeaway is that central-bank inflation forecasts are useful but biased instruments. They reflect the institutional commitment to the inflation target as much as they reflect the underlying data. Treating them as authoritative is a category error; treating them as worthless is also wrong. They are one input in a triangulation framework that should also include market-implied measures, structural analyses, and explicit acknowledgement of regime-shift risk.

For the next inflation episode, the diagnostic challenge is to identify the regime-shift signal early enough to discount the central-bank forecast appropriately. The 2021-2023 episode taught the lesson the hard way; the operational discipline is to apply the lesson before the next regime test rather than during it.

Last updated — 7 May 2026