Empirical benchmarks for ESCB models: A quantitative assessment of the macroeconomic impact of the 2022-23 monetary policy tightening

Abstract

In this policy note, drawing on a recent ECB Occasional Paper (November 2025, No. 377) and an independent meta-analysis on the effectiveness of monetary policy, we cross-check the ESCB’s workhorse models against selected empirical evidence. We address two central questions: first, what is the effect of a standard monetary policy shock on macroeconomic variables according to ESCB models, and how does this compare to findings from the empirical (mostly VAR-based) literature? Second, what has been the effect of the 2021–23 monetary policy tightening through the lens of these results? To answer the first question, we compare the findings of a recent meta-analysis of more than 400 empirical studies with a common simulation exercise conducted across several ESCB models. The results show that the size, dynamics and persistence of model-based and empirical responses are broadly similar. However, DSGE models tend to imply stronger effects—particularly on inflation—while semi-structural models are closer to the empirical benchmarks, reflecting their richer treatment of consumption dynamics and weaker reliance on intertemporal substitution.

Applying these insights to the 2022–23 tightening episode, we simulate all models under a counterfactual “no-tightening” scenario. The difference between the actual and counterfactual policy paths delivers an estimate of the impact of monetary policy on macroeconomic variables. The results suggest that monetary policy has been effective in curbing inflation, though at the cost of lower GDP growth, with notable differences across modelling approaches. Moreover, the results suggest that most of the impact of monetary policy on growth and inflation comes from the systematic response to changes in the macroeconomic environment.

Introduction

In this policy brief, drawing on a recent ECB OP paper (November 2025, No. 377) and an independent meta-analysis on the effectiveness of monetary policy, we cross-check the ECB models against selected empirical evidence. We address two central questions: First, what is the effect of a standard monetary policy shock on macroeconomic variables according to workhorse ESCB models, and how this compares to findings from the empirical (mostly VAR-based) literature? Second, what has been the effect of the 2021-23 monetary policy tightening through the lens of those results?

We answer the first question by comparing the findings of a recent meta-analysis of the empirical literature (Enzinger et al. 2025 full paper preprint, SUERF Policy Brief 1287 summary) with a common simulation exercise conducted across several ESCB models with unanticipated shocks and different policy rules across models. These are the most comparable sets of results based on “pure” shocks with possibly different empirical rules, which provides the closest possible counterpart to real-world conditions, where identification schemes and economic contexts naturally differ across jurisdictions and modelling frameworks. For the second question, we simulate all models under a counterfactual policy path in a “no-tightening scenario”, in deviation from the realised interest rate path, which is implemented via unexpected monetary policy shocks.1 The difference between the actual policy and the counterfactual policy delivers an estimate of the impact of monetary policy on the macroeconomic variables.2

ESCB models

The models routinely used for policy analysis within the ESCB generally fall into two broad categories: structural and semi-structural models. Structural models, such as Dynamic Stochastic General Equilibrium (DSGE) models, are micro-founded, firmly rooted in New Keynesian economic theory, and allow policymakers to interpret economic developments through the lens of structural shocks. Traditionally, in such models, agents are assumed to have rational expectations and engage in strong intertemporal substitution of consumption and labour after monetary policy shocks. Semi-structural models on the other hand, prioritize flexibility and alignment with empirical data to ensure timely responses to emerging economic challenges. They are usually larger than DSGE models and, in most central banks, they are used as core models for projection exercises. Equations are typically chosen to ensure a good data fit and, although consistent with economic theory, they are typically more eclectic and show less internal coherence compared to DSGE models. In particular, semi-structural models put less emphasize on intertemporal substitution effects.

Both types of models are essential for the ESCB’s policy work and are used complementarily. Together, they help guide medium-run projections and inform decision-making.

Empirical literature

There are hundreds of macroeconometric estimates of the impact of conventional monetary policy on output and prices. The literature covers a wide range of countries, time periods, econometric specifications and identification strategies for exogenous monetary policy shocks. In fact, this heterogeneity of approaches, in combination with notorious noise in the data implies a considerable variation in the reported output and inflation effects. Although most empirical results are consistent with the qualitative theoretical prediction that contractionary (expansionary) monetary policy leads to a hump-shaped decline (increase) in economic activity and inflation, there is disagreement with regard to the strength, the transmission time, and even the initial sign of the effects (see e.g. Coibion 2012, Ramey 2016). This diversity of findings underscores the need for a quantitative synthesis based on rigorous statistical criteria, i.e. a meta-analysis.

A recent study (Enzinger et al. 2025) provides such an analysis, covering more than 400 empirical papers that contain almost 5000 estimates of the responses of output and prices to conventional monetary policy shocks. This selection was based on a comprehensive, preregistered and replicable search process that filtered out the universe of comparable empirical estimates. The meta-analysis scraps and aggregates the point estimates, confidence intervals from the impulse-response functions as well as study characteristics of all these primary studies to derive an average estimate, detect if there are peculiarities in the reporting of findings and consider the influence of study and sample characteristics on the estimates. The main findings of the meta-analysis can be summarized as follows:

• Taking the simple mean across these studies, a 25 basis point (bp) interest rate hike reduces output by about 25 percent at the peak after around two years, while prices fall by about 0.2 percent after about four to five years. Uncertainty is substantial and one cannot exclude zero responses according to the 95 percent confidence levels (while the 68 percent confidence bounds do not contain zero effects after some time). The interest rate itself fades back to zero two years after the shock.
• The median responses of output and prices point to weaker dampening effects of contractionary monetary policy shocks at all response horizons. The median responses of the output and the price level are zero on impact and peak at similar horizons as the mean responses. However, the median effect sizes are only about 60 percent of the mean. The peak median response is around -0.15 percent for output and -0.1 percent for prices (median confidence bounds are somewhat narrower than for the mean but do also contain zero effects at the 95 percent level). This indicates that the distribution of output and price-level effect sizes is substantially left-skewed, while the interest response has a symmetric distribution.

These findings point to publication selection bias (Gechert et al. 2025), a preference of researchers for statistically significant and theory-confirming findings, as laid out in more detail in SUERF Policy Brief 1287. For the remainder of this paper, we refer to the median responses as a simple correction method of this selection process.

Monetary policy impact through the lens of ESCB models and empirical benchmarks

When we compare model-based effects of a (standard) monetary policy shock with empirical benchmarks, an important caveat is needed: the comparison can only be carried out along selected dimensions, and full harmonisation across models—as well as between models and the empirical literature—is not possible given their inherent heterogeneity. We take the responses of GDP and inflation in deviation from the baseline to a standard 25 bp unexpected shock to the interest rate. It is worth noting that, on average, the size, dynamics and the persistence of the shock, are very similar across the two sets of results.

Results are reported in Chart 1, where we compare the results of all models together, and in Chart 2, where we discriminate between structural and semi-structural models.

The median response of output is similar across theoretical and empirical models with a peak around -0.15 percent after about 2 years in both cases, although theoretical models suggest a somewhat faster pass-through in the short-term. The interpercentile range of the ESCB models (blue shaded area in Chart 1) suggests potentially stronger dampening effects on output during the first 2 years after the shock relative to the average 68% confidence interval in the empirical literature (yellow shaded area).

Turning to prices, the median price level response is predicted to be stronger in the ESCB models relative to the empirical literature, with a peak of -0.15 versus -0.11 percent after four years. In particular, during the first 2 years after the shock the median price level response in the models falls outside the average 68% confidence interval from the literature, getting back inside the interval from quarter 9 onwards.

Overall, the two sets of results are broadly consistent, but they also provide a clear message that the empirical evidence supports a less effective monetary policy on inflation than the one found across the average of models underlying the results in Chart 1.

Chart 1. Reconciling theoretical models and empirical evidence

Considering modelling approaches in more detail, Chart 2 shows some interesting differences. It is in particular DSGE models, which predict stronger and more immediate effects on output and inflation in comparison to the meta-analysis, partly driven by the strong intertemporal substitution effects of rising interest rates in these models. By contrast, estimated semi-structural models tend instead to be closer to the empirical evidence, especially over short and medium run. By abstracting from some of the cross-equation restrictions required for full structural identification, the semi-structural models allow for a richer representation of consumption – incorporating individual income risk and varying propensities to consume from different income sources. As a result, their consumption dynamics better reflect the observed hump-shape and persistence in consumption, thereby aligning more closely to the empirical evidence.

Discrepancies between euro-area semi-structural models – typically used for projections – and structural models – which are instead more used in policy simulations – appear significant mainly as regards the more front-loaded responses and would deserve further attention in the estimation of models.

Chart 2. Reconciling theoretical models and empirical evidence: Comparison across sets of models

Another useful point of comparison comes from the Basic Model Elasticities (BMEs), though a direct comparison with this tool is not feasible, as BMEs operate within a partial equilibrium framework.3 Instead, we use an EA-BMEs model, which builds on the Eurosystem’s BMEs while incorporating a forward-looking financial block à la Dornbusch.4 This framework maintains the real-side propagation mechanisms of the Eurosystem projections, ensuring internal consistency. The responses to monetary policy shocks in EA-BMEs are broadly in line with those observed in the semi-structural models and in the empirical meta-analysis, particularly with the impulse responses obtained in studies that identified the monetary policy shocks based on a more agnostic or Cholesky-type decomposition.5 This similarity reflects that most central banks in the Eurosystem rely on large-scale semi-structural models for projections that typically feature a relatively low GDP elasticity to interest rates and a Phillips curve that does not generate excessive amplification.

Finally, notice that the relative differences across models are maintained even if we compute the cumulative total impacts in Chart 3, which reflects the areas under each average IRF of Chart 2, notwithstanding the different persistence across the various sets of models.

Chart 3. Reconciling theoretical models and empirical evidence: Cumulative effects

A quantitative assessment of the macroeconomic impact of the 2022-23 tightening

How effective has monetary policy been over the recent tightening cycle? Addressing this question is relevant for the ECB to build a robust framework for incorporating monetary policy impacts into the staff projections and derive accurate forecasts and optimal policies conditional on those forecasts. We answer this questions with a conditional forecasting exercise. The exercise can give an idea of how much monetary policy has been effective in curbing inflation and at what cost in terms of output. To compute the impact of monetary policy, one can simulate the models using a counterfactual policy path in a “no-tightening scenario”, in deviation from the realised interest rate path, which is implemented via unexpected monetary policy shocks.6 The difference between the actual policy and the counterfactual policy delivers an estimate of the impact of monetary policy on the macroeconomic variables.

The results from different model categories (empirical models from the meta-analysis, semi-structural models and DSGE models) are reported in Chart4.7 To enhance the robustness of the findings, Chart 4 also reports a ‘confidence’ range around the results of the various model categories. Notice, however, that the uncertainty for each class of models is not comparable across models, since for VAR it is the variance of published results in the empirical literature, while for ESCB models it is the dispersion across models. With this caveat in mind, the exercise shows that the estimated effects of policy tightening on GDP growth and inflation are larger according to the average impulse responses from the DSGE models. By contrast, semi-structural models deliver the smallest impacts for GDP growth while VAR/LP models deliver the smallest estimated effects for inflation. This exercise can be very useful in a model averaging framework, which provides a more comprehensive assessment of monetary policy effectiveness in given circumstances.

Chart 4. Macroeconomic effects of the 2022-23 tightening according to empirical benchmarks

Conclusions and caveats

Comparing model-based responses to monetary policy shocks with empirical benchmarks shows broad consistency, but also some systematic differences. DSGE models tend to imply stronger effects of monetary policy—particularly on inflation—than those found in the empirical literature, while semi-structural models deliver responses closer to observed outcomes. This reflects their richer treatment of consumption dynamics and their weaker reliance on the intertemporal substitution channel. Applying these insights to the recent tightening cycle suggests that monetary policy has been effective in containing inflation, though at the cost of lower GDP growth. The magnitude of these effects, however, varies across modelling approaches: DSGE models generally imply stronger impacts, while semi-structural and empirical (mostly VAR/LP) models provide more moderate estimates. This underlines the importance of model averaging and of recognising model uncertainty when assessing policy effectiveness.

The previous counterfactual “no-tightening scenario” is obtained by adding to the historical data the monetary easing impulses consistent with keeping the policy rate unchanged relative to the financial markets’ expectations in December 2021. This approach assumes that the entire monetary policy shock was unexpected, meaning that all deviations from the projected path were unanticipated by economic agents. This seems to suggest that a more and accurate refined manner to gauge the impact of monetary policy should be to disentangle between the systematic response of the short-term nominal interest rate to macroeconomic developments from the ‘purely unexpected’ monetary policy shocks that shift the policy rate for a given set of macroeconomic data.

The typical models used in the Eurosystem projections can be used for such a decomposition if model simulations are harmonised across the board and conducted with the same policy rule, which would facilitate the aggregation over country models. For models of the euro area, the exercise would be relatively straightforward.

Chart 5. Decomposition on the impact of monetary policy according to ECB-BASE model

Chart 5 shows a historical decomposition over the sample 2020Q1-2024Q4 using the ECB-BASE model, that identifies the impact of monetary policy shocks (non-systematic component) and the impact of the systematic component of monetary policy since January 2020. The latter refers to the historical response in the short-term nominal interest rate to changes in inflation and output according to the policy rule, while the non-systematic component accounts for deviations in the policy rate from the rule.8  The results suggest that most of the impact of monetary policy on growth and inflation comes from the systematic response to changes in the macroeconomic environment. By contrast in 2022 results show a mitigating effect from accommodative monetary policy shocks as the exercise does not account for the anticipation of the interest rate increases due to the backward-looking nature of the ECB-BASE model. Quantitatively, the impact on annual growth peaks in the first quarter of 2024, offsetting about 2pp of annual growth, while policy has only limited disinflationary effects, materializing since the first quarter of 2024.