The conflict in the Middle East has once again exposed the euro area to sizeable energy shocks. One key lesson from the 2021-22 inflation surge was that when large shocks hit, inflation can react disproportionally, meaning in a non-linear way. Back then euro area inflation reached double digits and traditional linear models were unable to reconcile the facts. This experience has reshaped how policymakers think about inflation dynamics. In its 2025 monetary policy strategy assessment, the ECB Governing Council affirmed that monetary policy should be forceful or persistent in responding to large, persistent deviations of inflation above target as well as below target owing to non-linearities in price and wage setting that can kick in when inflationary shocks hit (ECB 2025). 

The current energy shock shares some similarities with the one following Russia’s invasion of Ukraine, but there are some key differences for the euro area economy. When the 2022 energy shock occurred, energy prices had already been rising steeply for months beforehand. In contrast, euro area energy prices had been decreasing for 12 months when the war in Iran started. The current energy shock is so far less broad-based for the euro area; while oil price dynamics have resembled those observed after the Ukraine invasion, wholesale gas and electricity prices remain well below their 2022 peaks. Moreover, the macroeconomic backdrop is more benign today, without the strong reopening effects that followed the pandemic (e.g. Lagarde 2026). On the other hand, while both shocks have a geopolitical component which can hit harder than other energy shocks (see Verduzco-Bustos and Zanetti 2026), the current shock is more global in nature compared to the one following the Ukraine invasion, which can amplify its effects.

While economists tend to agree that inflation can behave in a non-linear way following large shocks, pinning this down empirically and in real time is hard. In general, traditional models typically impose specific assumptions about these non-linearities, often assuming either smooth changes over time or a limited number of discrete economic ‘regimes’. However, these assumptions are rarely directly observable in the data. In Bobeica et al. (2025) we develop a more flexible approach that does not pre-impose a specific form of non-linearity. Importantly, the model also allows structural shocks to affect economic variables in a non-linear way, also contemporaneously.

The framework builds on machine learning techniques in the spirit of Clark et al. (2023), combining traditional vector autoregressions (VARs) with Bayesian Additive Regression Trees (BART, Chipman et al. 2010). This allows the data to determine whether relationships are linear or non-linear. While such models have been widely used for forecasting, we show that they can also be used to identify inflationary shocks and study how they propagate in a non-linear way.

To identify structural shocks, we assume that the reduced-form errors are driven by a small number of latent factors. These factors enter the model as the sum of a linear component and an unknown non-linear component, which allows for non-linear impact effects. The factors are interpreted as fundamental economic shocks, and their identification follows Korobilis (2022) using a combination of sign, zero, and magnitude restrictions on the linear component. To capture energy price dynamics, we use a synthetic energy commodity price indicator that combines crude oil and wholesale gas prices, which is more appropriate to characterise inflationary pressures.

We find that large energy shocks have disproportionately strong effects on inflation. Figure 1 shows the responses of euro area headline inflation to small, medium, and large shocks, all scaled for comparability. If inflation responded proportionally to all shock sizes (so that, for example, a two-standard-deviation shock simply doubles the effect of a one-standard-deviation shock) the three lines would overlap. Instead, they clearly diverge. Small shocks have little effect on inflation, while medium and large shocks generate increasingly strong responses as firms pass more of the higher costs through to prices. In terms of magnitude, a medium-sized shock produces an inflation response around 2.5 times larger than a small shock. One explanation for such non-linearity is that it is costly to change prices and so when the cost shock is bigger firms become increasingly willing to adjust prices (see Ball and Mankiw 1995). Empirically, micro-based evidence pointed to more frequent price adjustments in the post-pandemic period (see Gautier et al. 2026 for the euro area and Montag and Villar 2023 for the US). This is consistent with state-dependent pricing models, whereby firms adjust prices more readily after large shocks. Cavallo et al. (2023) show that price changes become more frequent and shocks transmit faster when they are large. A second mechanism giving rise to non-linearities resides in the amplification impact via second-round effects. Wages and inflation expectations adjust with a lag, reinforcing the initial shock through subsequent price increases. Ampudia et al. (2024) find a stronger link between wages and inflation in the recent high-inflation period, while Acharya et al. (2023) show that higher inflation expectations, combined with greater pricing power, can make supply shocks more persistent. In normal times, firms are cautious about raising prices to avoid losing customers; when inflation expectations rise, this constraint weakens and pass-through increases.

Figure 1 Reaction of inflation to small, medium, and large energy shocks (percentage points)

Notes: Thick lines are median estimates and the shaded areas are the 68% credible interval. Impulse response functions (IRFs) are normalised to a one standard deviation shock for comparability.                

We label energy shocks as being small, medium, and large based on the impact they have on the synthetic energy commodity price indicator. Specifically, a one standard deviation shock triggers a small increase in the synthetic indicator, occurring in less than 10% of all increases in our sample. A two standard deviation shock elicits an average-sized increase in the synthetic indicator, with half of the increases in our sample being smaller and half larger. A three standard deviation shock causes a large increase, occurring in around 10% of all recorded increases.

The recent energy shock can so far be characterised as medium-sized and already points to an increased likelihood of some non-linear effects that can intensify with the size of the shock. The taxonomy of shocks according to their size is model-dependent. Through the lens of this model, in April 2026 the dynamics in the synthetic energy commodity price indicator were akin to a medium-sized shock with an annual increase of approximately 45% (Figure 2). Specifically, a one-off medium-sized shock of two standard deviations raises headline inflation by about 0.4 percentage points at its peak after nine months, while the impact on core inflation is more delayed and muted, reaching around 0.1 percentage points after 16 months. While classifying shocks as small, medium, or large helps to communicate results more clearly, the underlying economic relationship is better thought of as a continuum. Inflation responses become progressively stronger as shocks grow, rather than shifting abruptly between regimes. Figure 3 focuses on a fixed horizon after the energy shock and shows the total inflation response to shocks of different sizes (on the horizontal axis). All responses are standardised for comparability; if effects were proportional, the line would be flat. Instead, the figure shows increasingly strong non-linear effects as the shock size rises, suggesting that models with only a few discrete regimes oversimplify inflation dynamics.

Figure 2 Synthetic energy commodity indicator (annual log difference)

Notes: Annual log differences in the synthetic indicator of energy commodity prices, combining oil and gas prices using euro area’s energy import shares. Red, yellow, and green dotted lines mark thresholds for small, medium, and large energy shocks. Latest observation: April 2026.

Figure 3 Reaction of inflation to energy shocks of different size

Notes: Black lines are median estimates and the blue lines denote the 16/84 percentiles of the posterior distribution. The horizontal axis denotes the sizes grid considered for the shocks in terms of standard deviations. Responses refer to the one-year horizon and are normalised to a one standard deviation shock for comparability. 

While the size of the shock is important, also the persistence – how long energy prices remain elevated – and breadth – how widely cost increases spread across sectors, are crucial for inflation. From a firm’s perspective, it makes little sense to react disproportionately to a one-off shock, especially if it is idiosyncratic and does not affect competitors or the broader economy. The 2021–22 inflation episode was unusual because it combined all three features: it was large, persistent and affected many sectors through widespread global supply chain bottlenecks. In practice, it is difficult to disentangle the relative importance of size, persistence, and breadth, as the post-pandemic period, which is key for identifying non-linearities, was exceptional along all these dimensions. Figure 4 simulates the impact of large shocks before and after the pandemic and shows that transmission mechanisms have changed. Two main results stand out. First, energy shocks have a much stronger effect on inflation when the model is trained on the more recent period. Second, the shape of the impulse responses differs noticeably across samples, with significantly greater persistence in the post-2019 period, consistent with the higher persistence of the shocks themselves.

Figure 4 Simulated reaction of inflation to large energy shocks before and after the pandemic

Notes: Red areas denote the 68% credible intervals for the period from 2020:01 to 2024:11 while blue areas are the 68% credible intervals for the period from 1996:01 to 2019:12.

The likelihood of non-linear inflation dynamics indicates that risks to the inflation outlook are skewed to the upside. When inflation responds non-linearly to positive shocks, the risks around baseline projections can become tilted to the upside. In the current context, where energy shocks have reached the medium size threshold, this implies that inflationary pressures could be stronger and more durable than suggested by linear models. Our findings suggest that when large energy shocks bite, central banks should pay very close attention to indicators of how the shock is propagating through the economy. 

References

Acharya, V V, M Crosignani, T Eisert and C Eufinger (2023), “How Do Supply Shocks to Inflation Generalize? Evidence from the Pandemic Era in Europe”, NBER Working Papers 31790. 

Ampudia, M, M J Lombardi and T Renault (2024), “The wage-price pass-through across sectors: evidence from the euro area”, Working Paper Series 2948, European Central Bank.

Ball, L and N G Mankiw (1995), “Relative-Price Changes as Aggregate Supply Shocks”, The Quarterly Journal of Economics 110: 161–193.

Bobeica, E, S Holton, F Huber and C Martínez Hernández (2025), “Beware of large shocks! A non-parametric structural inflation model”, Working Paper Series 3052, European Central Bank.

Cavallo, A, F Lippi and K Miyahara (2023), “Inflation and misallocation in New Keynesian models”, Sintra paper, ECB Forum on Central Banking June 2023.

Chipman, H A, E I George and R E McCulloch (2010), “BART: Bayesian additive regression trees”, The Annals of Applied Statistics 4: 266– 298. 

Clark, T E, F Huber, G Koop, M Marcellino and M Pfarrhofer (2023), “Tail forecasting with multivariate Bayesian additive regression trees”, International Economic Review 64: 979–1022.

Dedola, L, L Henkel, C Hoeynck, C Osbat and S Santoro (2024), “What does new micro price evidence tell us about inflation dynamics and monetary policy transmission?”, Economic Bulletin Articles, Issue 3.

European Central Bank (ECB) (2025), “The ECB’s monetary policy strategy statement (2025)”, June.

Gautier, E, C Conflitti, L Fadejeva, A Grimaud, V Jouvanceau, J-O Menz, A Paulus, P Petroulas, P Roldan-Blanco, E Wieland, D Enderle and E Gutiérrez (2026), “Consumer Price Stickiness in the Euro Area During an Inflation Surge”, Working Paper Series 3181, European Central Bank. 

Korobilis, D (2022), “A new algorithm for structural restrictions in Bayesian vector autoregressions”, European Economic Review 148, 104241.

Lagarde, C (2026), “Navigating energy shocks: risks and policy responses”, Keynote speech at “The ECB and Its Watchers” conference organised by the Institute for Monetary and Financial Stability at Goethe University Frankfurt, Frankfurt am Main, 25 March.

Montag, H and D Villar (2023), “Price-Setting During the Covid Era”, FEDS Notes 2023-08-29, Board of Governors of the Federal Reserve System (US).

Verduzco-Bustos, G and F Zanetti (2026), “Geopolitical oil price shocks: Why these shocks hit harder”, VoxEU.org, 28 April.