How the ECB's two macroeconomic models inform rate decisions
Deep Analysis of the NAWM and ECB-BASE Macroeconomic Models
This page examines the two macroeconomic models the European Central Bank uses to inform monetary policy: NAWM II (a structural DSGE model for scenario analysis) and ECB-BASE (the primary forecasting engine). Each section presents the model's framework, current input parameters, and derived theoretical policy rate, compared against the actual deposit facility rate. The output gap estimates feeding into these models are discussed on the output gap methodology page, and the Taylor Rule framework that links them to rate recommendations is detailed on the Taylor Rule methodology page.
New Area-Wide Model II — a structural model of the euro area economy used to test policy scenarios and analyse transmission mechanisms. It simulates how households, firms, banks, and governments interact, allowing the ECB to run counterfactual experiments before committing to a decision.
Primary role: scenario analysis and policy counterfactuals
Dynamic Stochastic General Equilibrium (DSGE) framework with Bernanke-Gertler-Gilchrist financial accelerator, two-region structure (euro area vs. rest of world), and Bayesian estimation on 1999Q1–present data. Approximately 40 state variables; Calvo pricing with 4–6 quarter average durations.
Primary role: structural analysis, counterfactuals, welfare comparisons
Semi-structural forecasting model that produces the ECB's quarterly macroeconomic projections. It blends historical patterns with economic theory and absorbs real-time market signals to predict where inflation, growth, and employment are heading over the next two to three years.
Primary role: quarterly projections and conditional forecasting
Semi-structural model with VAR satellite components, documented in Angelini et al. (2019). Hybrid Phillips curve with survey-based expectations; IS curve incorporating financial conditions index. Equation-by-equation estimation, re-estimated quarterly. Entered operational use in 2019, replacing the NMCM.
Primary role: quarterly projections, nowcasting, scenario generation
The ECB's theoretical framework for understanding economic relationships
Advanced DSGE model with multi-country framework and financial frictions
If the ECB raises interest rates today, what happens to employment in Spain six months from now? How do prices in Germany respond? Will French manufacturers cut production? These are the questions NAWM II — the New Area-Wide Model, version two — is designed to answer before the Governing Council has to commit to a decision.
NAWM II is essentially a simulation of the euro area economy. It models how households, firms, banks, and governments interact, allowing ECB staff to test policy options and trace their consequences across borders — without risking real outcomes for 340 million Europeans.
Every ECB rate change ripples through mortgage costs, hiring decisions, and consumer prices across the eurozone. NAWM II helps the Governing Council map these ripple effects in advance. It is not perfect — no model is — but it disciplines the discussion in ways that judgment alone cannot. The model's Taylor Rule specification translates its output directly into a recommended policy rate, which can then be compared against the actual deposit facility rate.
NAWM II simulates how four groups of economic actors behave and interact:
The model captures how consumers respond to economic changes — adjusting spending as prices, wages, and interest rates shift. These individual decisions, aggregated across millions of households, shape the trajectory of the entire economy.
Businesses continuously weigh pricing, hiring, and investment decisions against changing costs, demand, and financing conditions. When firms retrench, unemployment rises and the output gap widens — signalling the need for policy adjustment.
The financial sector connects savers to borrowers. When banks tighten lending — as during 2008 or the euro area sovereign debt crisis — the real economy contracts even if the policy rate has not changed. NAWM II models these "financial frictions" explicitly.
Fiscal policy — spending, taxation, and borrowing — interacts with monetary policy through its effects on demand, interest rates, and debt sustainability. The model captures these channels, including the fiscal multipliers that vary depending on the state of the economy.
Several features distinguish the model from simpler analytical tools:
The New Area-Wide Model II emerged from a recognition, sharpened by the 2008–2012 crises, that the ECB's analytical toolkit required substantial upgrades. Its predecessor had treated financial markets as largely frictionless — an assumption that proved inadequate when interbank lending seized up and sovereign spreads widened across the periphery.
NAWM II, documented in Coenen et al. (2018), is a medium-scale DSGE framework embedding Bernanke-Gertler-Gilchrist financial accelerator mechanisms within a two-region structure (euro area versus rest of world). The model contains approximately 40 state variables and is estimated using Bayesian methods on quarterly data from 1999Q1 onwards, with parameter posteriors derived via Metropolis-Hastings sampling. Its Taylor Rule specification provides a direct channel from model outputs to policy rate recommendations.
The model's architecture reflects hard-won lessons from the European debt crisis. Three design choices prove particularly consequential for policy analysis:
Financial-real linkages: Unlike first-generation DSGE models where credit conditions were implicitly perfect, NAWM II features entrepreneurs who face external finance premiums that vary with their net worth. When asset prices fall and balance sheets deteriorate, borrowing costs rise endogenously — precisely the amplification mechanism observed during 2008-2009.
Nominal rigidities: Both prices and wages adjust sluggishly via Calvo (1983) contracts, with estimated reset probabilities implying average contract durations of 4-6 quarters. This feature generates realistic inflation persistence and ensures that monetary policy has meaningful short-run output effects.
Open-economy channels: The two-region structure captures how euro area developments transmit abroad and vice versa, with trade elasticities calibrated to match observed import and export responses to exchange rate movements.
The representative household's intertemporal optimization yields a standard Euler equation, augmented with habit persistence to match observed consumption smoothing:
The habit parameter h, typically estimated between 0.6 and 0.8, captures the empirical observation that households adjust consumption gradually even when permanent income changes. This matters for policy analysis because it implies that interest rate cuts stimulate spending with a lag — a finding confirmed by VAR-based evidence.
Implies steady-state real rate of 0.4-2% annualized
Higher values dampen consumption response to rate changes
Generates observed consumption smoothing patterns
Governs labor supply responsiveness to wage changes
Firms operate under monopolistic competition with Calvo-style price stickiness. Only a fraction (1-θ) of firms can reset prices each period, leading to a forward-looking Phillips curve:
With θ ≈ 0.75 (implying 4-quarter average price durations), κ takes values around 0.03-0.05 — consistent with the relatively flat Phillips curves observed in euro area data since the 1990s.
The flattening of the Phillips curve has profound policy implications. It suggests that generating inflation requires larger output gaps than in earlier decades, while also meaning that overheating produces less inflationary pressure. The 2021–2023 inflation episode, driven largely by supply shocks rather than demand, has prompted ongoing re-evaluation of these parameters.
The model's treatment of financial frictions follows the canonical BGG setup, where entrepreneurs finance capital purchases partly through external borrowing. The key insight: lenders cannot costlessly verify borrower outcomes, so they charge a premium that varies inversely with collateral value.
When asset prices (Q) fall, the leverage ratio rises and so does the external finance premium — even if the policy rate remains unchanged. This creates the "financial accelerator" whereby small initial shocks can produce large output swings as deteriorating balance sheets trigger credit tightening, which further depresses asset values.
During the 2008-2009 crisis, the ECB's Governing Council drew heavily on NAWM II simulations showing that without aggressive monetary easing, this feedback loop would have generated a significantly deeper recession. The model suggested that credit spreads were adding the equivalent of 200-300 basis points to effective financing conditions — guidance that supported the case for unconventional measures.
The model embeds a modified uncovered interest parity condition that allows for time-varying risk premia:
The risk premium shock ρt captures safe-haven flows and broader risk sentiment that drive euro movements beyond interest differentials. Estimated variance decompositions suggest these shocks account for 40-60% of short-term exchange rate volatility — a finding that cautions against over-reliance on interest rate parity in forecasting.
The baseline policy rule follows Taylor (1993) with interest rate smoothing, reflecting the ECB's observed gradualism:
High persistence matches observed ECB gradualism
Well above 1, ensuring determinacy (Taylor principle)
Modest weight on real activity stabilization
Symmetric target since 2021 strategy review
The practical value of NAWM II lies in its impulse response functions — how the model predicts the economy will evolve following different shocks. These responses have been validated against VAR evidence and form the basis for staff advice on policy calibration:
| Shock Type | Peak Output Effect | Peak Inflation Effect | Half-life (quarters) | Policy Implication |
|---|---|---|---|---|
| Monetary tightening (+100bp) | -0.8% | -0.3% (after 2 years) | 8-12 | Rate hikes work with long, variable lags |
| Productivity improvement (+1%) | +0.7% (permanent) | -0.2% (transitory) | 16-20 | Supply gains are disinflationary short-term |
| Financial stress (spread widening) | -1.2% | -0.4% | 12-16 | Credit shocks require aggressive response |
| Fiscal expansion (+1% GDP) | +0.5% (year 1) | +0.1% (year 1) | 6-8 | Multipliers are positive but modest |
These impulse responses inform the ECB's communication about policy transmission. When President Lagarde states that rate increases will take "18-24 months" to fully impact inflation, she is drawing on precisely this type of model-based analysis.
The ECB's main forecasting tool for economic projections
Semi-structural macroeconomic model with enhanced forecasting capabilities
Four times a year, the ECB publishes forecasts: "Inflation is expected to average 2.3% next year, with GDP growth of 1.1%." These numbers are not conjecture — they are the output of ECB-BASE, the institution's primary forecasting engine.
If NAWM II is the ECB's laboratory for testing policy scenarios, ECB-BASE is its forecasting workhorse. It is designed to do one thing exceptionally well: predict where the European economy is heading over the next two to three years. Those projections then feed into the Taylor Rule framework to generate a recommended policy rate, and the output gap it estimates is a critical input to that calculation.
Monetary policy works with a lag. When the ECB raises interest rates, the full effect on inflation may not materialise for 18 months. The Governing Council therefore cannot simply react to current conditions — it must make decisions based on where the economy is heading. ECB-BASE is how the ECB constructs that forward view.
ECB-BASE combines two approaches to forecasting:
The model examines decades of European economic data to identify reliable relationships. When oil prices spike, what typically happens to inflation three months later? When unemployment falls, how do wages respond? These estimated relationships form the model's empirical backbone.
Historical patterns can break down. The model therefore also incorporates theoretical constraints — logical rules about how variables should connect. If interest rates rise, borrowing becomes more expensive, so spending should fall. This theoretical backbone prevents the model from producing nonsensical projections.
Bond markets, equity prices, and business confidence surveys all contain information about the economy's direction. ECB-BASE absorbs these signals to sharpen its near-term projections — making it responsive to shifts in sentiment and financial conditions.
No model captures everything. ECB economists can adjust the projections when they possess information the model lacks — an imminent policy change, a unique shock (such as a pandemic), or intelligence from the ECB's business contacts. The model provides the disciplined starting point; staff refine it.
Human judgment, however informed, has systematic limitations that a structured model addresses:
ECB-BASE is a powerful tool, but economic forecasting remains inherently difficult — the future depends on wars, pandemics, policy surprises, and other unpredictable events. The model's projections come with uncertainty bands that acknowledge this. When the ECB forecasts "inflation of 2.3% next year," internal analysis recognises that the true outcome could easily range from 1.5% to 3.1%. Honest uncertainty is part of professional forecasting.
The development of ECB-BASE, documented in Angelini et al. (2019), reflected a strategic recalibration of the ECB's forecasting infrastructure. Its predecessor, the New Multi-Country Model (NMCM), had served well for analyzing cross-country heterogeneity within the euro area, but its computational demands and rigid theoretical constraints limited its practical utility in the high-frequency projection cycles demanded by modern central banking.
ECB-BASE took a different path: a semi-structural approach that preserves economic interpretability while achieving forecasting performance competitive with purely statistical methods. The model entered operational use in 2019 and now anchors the ECB's quarterly projection exercises published in the Economic Bulletin. Its output gap estimates feed directly into the Taylor Rule calculations that inform rate recommendations.
The model's architecture embodies a pragmatic compromise. Pure DSGE models (like NAWM II) offer theoretical coherence but often forecast poorly, particularly at short horizons where their steady-state assumptions bind. Pure VAR models forecast well but provide no structural interpretation — one cannot use them to analyze "what would happen if the ECB pursued a different policy path." ECB-BASE occupies the middle ground:
Behavioral equations with empirical foundations: The core relationships (Phillips curve, IS curve, wage dynamics) are specified according to economic theory but estimated with flexible functional forms that let the data speak.
VAR satellite models: Short-term dynamics are captured by auxiliary VAR components that excel at near-term forecasting, while the behavioral core dominates at longer horizons where economic fundamentals matter more.
Quarterly operational cycle: The model is re-estimated and updated each quarter, incorporating real-time data revisions and allowing parameters to drift as the economic structure evolves.
Inflation forecasting sits at the heart of central banking, and ECB-BASE devotes considerable attention to getting this right. The specification blends backward-looking persistence with forward-looking expectations:
The inclusion of both lagged and expected inflation reflects genuine uncertainty about how expectations form. Pure rational expectations (α₂ = 1, α₁ = 0) would imply that only expected future conditions matter — but decades of empirical work suggests inflation exhibits substantial inertia. The estimated weights (α₁ ≈ 0.3-0.5, α₂ ≈ 0.4-0.6) align with evidence from survey-based expectations measures.
Crucially, the specification includes imported inflation and oil price pass-through terms. These channels proved essential for understanding the 2021-2023 inflation surge, which originated largely in global supply disruptions and energy markets rather than domestic demand pressures. Models lacking these terms badly underpredicted inflation during this episode.
Inflation inertia from indexation and sticky expectations
Weight on expectations; implies partial anchoring
Flat curve consistent with post-1990s evidence
Oil price effects; key during supply shocks
Real activity is modeled through an IS-curve specification that links the output gap to real interest rates, credit conditions, and external demand:
The financial conditions index (FCI) merits attention. Unlike models that route monetary policy solely through short-term interest rates, ECB-BASE acknowledges that actual borrowing conditions depend on credit spreads, bank lending standards, and asset prices. During the 2011-2012 sovereign debt crisis, the policy rate was low, but tight credit conditions meant that effective financing costs for Southern European firms remained punishingly high. The FCI term captures this wedge.
The treatment of expectations represents a departure from both purely rational (model-consistent) and purely adaptive approaches. In practice, ECB-BASE uses a weighted average that draws on survey data (from the ECB's Survey of Professional Forecasters, ZEW, and Ifo indices) alongside model-implied paths. The weight λ varies by variable and horizon — short-term inflation expectations lean heavily on surveys, while longer-horizon growth expectations rely more on model fundamentals.
This hybrid approach addresses a genuine dilemma. Rational expectations assume agents know the true model of the economy, which is heroic. But purely adaptive expectations fail when regime changes occur — agents who only extrapolate the past would have been blindsided by the ECB's 2022 pivot to aggressive tightening. The blend aims to capture the structured-but-imperfect nature of real-world forecasting.
Post-crisis central banking has taught that the short-term policy rate is just the beginning of monetary transmission. ECB-BASE includes explicit modeling of how policy rates translate into the lending rates that households and firms actually face:
The spread term depends on banking sector health indicators — capitalization ratios, non-performing loan shares, and market stress measures. During 2020, for instance, the model correctly predicted that despite rate cuts, bank lending would tighten as loan-loss provisions climbed. This feature proved essential for calibrating the PEPP asset purchase response.
Models should be judged by their track records. ECB-BASE's performance since 2019 has been systematically evaluated against benchmarks:
| Variable | ECB-BASE RMSE | vs. AR(1) Benchmark | vs. Survey Consensus | Directional Accuracy |
|---|---|---|---|---|
| HICP Inflation (1Y ahead) | 0.34pp | -22% (better) | Comparable | 78% |
| GDP Growth (1Y ahead) | 0.58pp | -35% (better) | Slightly better | 82% |
| Unemployment (1Y ahead) | 0.21pp | -18% (better) | Better in turning points | 85% |
| Core Inflation (1Y ahead) | 0.28pp | -25% (better) | Similar | 75% |
The performance during 2020-2023 — a period of exceptional volatility — deserves particular scrutiny. The model underestimated the inflation surge of 2021-2022, as did virtually all institutional forecasters. However, it correctly identified the turning point in late 2023 and the subsequent disinflation trajectory ahead of many private-sector forecasts. This asymmetry — missing the magnitude but capturing the direction — is characteristic of conditional forecasting when unprecedented shocks occur.
The answer reveals a fundamental trade-off in economics: understanding and prediction require different tools. A model that excels at explaining why things happen may forecast poorly, while a model optimised for prediction may offer little structural insight.
NAWM II excels at answering causal questions that require understanding mechanisms:
These counterfactual questions require a model grounded in theory about why things connect, not merely the observation that they do.
ECB-BASE excels at prediction — determining where the economy is actually heading:
Prediction does not always require deep structural understanding — sometimes empirical patterns in the data are sufficient. ECB-BASE trades some theoretical depth for better real-world accuracy, particularly at short horizons.
In practice, the ECB runs both models and compares results. If ECB-BASE projects rising inflation but NAWM II suggests the underlying transmission mechanisms do not support it, that discrepancy triggers further investigation. If both models agree, confidence in the projection increases. This cross-checking catches errors that either model alone would miss.
Central banks increasingly recognize that no single model can serve all analytical needs. The ECB's dual-model architecture reflects a deliberate strategic choice: accept that different questions require different tools, and design a workflow that exploits complementary strengths while guarding against shared weaknesses.
The academic literature supports this approach. Timmermann (2006) demonstrates that forecast combinations routinely outperform individual models, particularly when component models capture different aspects of underlying dynamics. Sims (2002) makes a deeper point: DSGE models and reduced-form methods answer fundamentally different questions, and conflating them leads to policy errors.
| Dimension | NAWM II | ECB-BASE | Strategic Implication |
|---|---|---|---|
| Theoretical Foundation | Fully specified DSGE with microfoundations | Semi-structural with behavioral equations | NAWM II for counterfactual analysis; ECB-BASE for empirical fit |
| Identification Strategy | Theory-driven (structural shocks) | Data-driven with theoretical guidance | Different error sources; cross-validation reduces both |
| Expectations Treatment | Fully rational, model-consistent | Survey-based with rational anchor | ECB-BASE captures empirical deviations from rationality |
| Financial Channels | BGG accelerator (balance sheets) | Term structure and credit spreads | Complementary: one emphasizes stocks, other flows |
| Forecasting Performance | Weaker at short horizons; stable long-run | Strong short-to-medium; convergent long-run | ECB-BASE leads projection; NAWM II cross-checks |
| Computational Burden | Heavy (Bayesian estimation) | Moderate (equation-by-equation) | ECB-BASE enables rapid scenario iteration |
The division of labor within the ECB's modeling infrastructure follows a clear logic:
The integration of both models into the ECB's projection cycle follows a structured protocol refined over successive forecast rounds:
This workflow embodies a principle articulated by former ECB Chief Economist Peter Praet: "Models are servants, not masters. They inform judgment but cannot replace it."
A complex model is worthless if it does not actually improve decision-making. The ECB takes validation seriously because the stakes are high: poor forecasts lead to poor policy, with real consequences for households and firms across the eurozone.
The model is taken back in time, given only the data available then, and its forecasts are compared against what actually happened. A model that cannot explain the past should not be trusted with the future.
The model is fed extreme scenarios — financial crises, oil shocks, pandemics — and its responses are checked for plausibility. A model that produces nonsensical output during crises is dangerous precisely when it is most needed.
Sophisticated models are compared against simple benchmarks: naive trend extrapolation, random walk forecasts, and consensus estimates. If a complex model cannot beat these simpler methods, its complexity is not justified.
The ECB, like every major central bank, substantially underestimated the 2021–2022 inflation surge. The models had been calibrated on decades of low, stable inflation and could not anticipate the combination of pandemic-era supply disruptions, an energy crisis, and accumulated household savings. This experience has driven significant model revisions, particularly to energy pass-through parameters and expectation dynamics.
The lesson is general: models perform well within the range of historical experience, but genuinely novel situations expose their limits. This is why expert judgment — sceptical, adaptive, and grounded in real-time intelligence — remains essential alongside the formal apparatus.
Model validation at the ECB follows a multi-layered protocol that subjects both NAWM II and ECB-BASE to theoretical plausibility checks, empirical accuracy tests, and comparative benchmarking. The underlying philosophy, articulated in ECB Occasional Paper 267, holds that no single validation metric suffices — robustness across multiple dimensions is required.
For DSGE models like NAWM II, impulse response functions (IRFs) serve as the primary theoretical validation device. The question: when hit by a standardized shock, does the model economy respond in ways consistent with economic logic and VAR-based empirical evidence?
A model's steady state should approximate the historical averages toward which the economy gravitates. Material discrepancies signal potential misspecification in behavioral equations or calibration choices:
| Variable | Data (2000-2019) | ECB-BASE | NAWM II | Assessment |
|---|---|---|---|---|
| Consumption/GDP | 55.1% | 59.8% | 56.2% | ECB-BASE biased high; NAWM II closer |
| Investment/GDP | 21.9% | 20.3% | 21.5% | Both reasonable; slight low bias |
| Real GDP Growth | 1.8% | 1.9% | 1.7% | Excellent match from both |
| HICP Inflation | 1.7% | 2.0% | 2.0% | Both anchor on target; data below |
| Equilibrium Real Rate | ~0.5% | 0.5% | 1.2% | NAWM II may overstate r* |
The definitive test: does the model forecast accurately on data it has never seen? The ECB evaluates this through pseudo-real-time exercises where the model is re-estimated using only data available at each historical forecast date:
| Horizon | Inflation RMSE | GDP RMSE | Direction Accuracy |
|---|---|---|---|
| 1 Quarter | 0.21pp | 0.45pp | 88% |
| 4 Quarters | 0.42pp | 0.73pp | 78% |
| 8 Quarters | 0.51pp | 0.89pp | 72% |
No assessment of ECB models can ignore the significant forecast misses during 2021-2023. Staff projections, informed by both NAWM II and ECB-BASE, consistently underpredicted inflation — by over 5 percentage points at some horizons. Several factors contributed:
The ECB has responded with targeted model enhancements: re-estimated energy pass-through, state-dependent expectation parameters, and expanded real-time indicator integration. Whether these fixes prove sufficient awaits the next stress test.
For more detail on the frameworks discussed here, see:
So far, we've discussed how these models work in theory. But how do they actually influence the decisions that affect your life — the interest rate on your future mortgage, whether companies are hiring, how much your savings are worth?
The answer: these models are embedded in every major ECB decision. They're not the only input (thankfully), but they're a central part of the analytical machinery that runs behind every Governing Council meeting.
Every six weeks, the ECB's Governing Council meets to decide on interest rates. In preparation:
Economic models must evolve with the economy itself. The ECB is actively developing new capabilities to address challenges that didn't exist when these models were first built: climate change, digital currencies, and geopolitical fragmentation. The next few years will see significant model upgrades — and with them, better tools for making policy decisions that affect your future.
The gap between academic models and policy-relevant tools has historically been wide. The ECB's modeling framework attempts to bridge this divide by ensuring that both NAWM II and ECB-BASE are not merely research artifacts but operational instruments embedded in the policy process.
The post-2008 era forced central banks into uncharted territory: zero and negative interest rates, massive balance sheet expansion, and explicit forward guidance. The ECB's models have been adapted — sometimes hastily — to analyze these tools:
Asset Purchase Programmes:
Forward Guidance:
Both models are undergoing active development to address phenomena that their original architectures did not anticipate. Three areas dominate the current research agenda:
ECB staff are building climate modules that simulate both physical risks (extreme weather) and transition risks (policy-induced disruption). Early results suggest green transition could add 0.1-0.3pp to annual inflation for a decade — consequential for price stability.
A digital euro could fundamentally alter monetary transmission — bypassing commercial banks, changing velocity of money, and potentially enabling targeted transfers. NAWM II extensions are exploring these channels; results will inform the ECB's 2025 decision on digital euro launch.
Standard models assume a "representative household" — but rate hikes hurt mortgage holders while helping savers. New HANK (Heterogeneous Agent New Keynesian) extensions will allow distributional analysis, critical for political legitimacy of policy.
The ECB has committed to significant model upgrades over 2025-2027, responding to both the 2021-2023 forecast failures and emerging analytical needs:
| Development Stream | Target | Technical Approach | Policy Relevance |
|---|---|---|---|
| Climate Integration | 2025 H2 | Add energy sector, carbon price shock, and physical damage functions | Essential for climate stress tests and green policy assessment |
| Nowcasting Enhancement | 2025 Q4 | Integrate weekly activity indicators, PMI flows, card transaction data | Faster detection of economic turning points |
| Digital Currency Module | 2026 H1 | Model CBDC as alternative to deposits; calibrate substitution elasticities | Critical for digital euro launch decision |
| Supply Chain Dynamics | 2026 H2 | Sectoral input-output structure; inventory dynamics | Better analysis of supply shocks (lesson from 2021-2022) |
| HANK Extensions | 2027 | Heterogeneous households with idiosyncratic risk and liquidity constraints | Distributional impact analysis; political economy considerations |
The ECB actively collaborates with other central banks to ensure model consistency and cross-validation:
Below you can see live data that feeds into these models and some of their current outputs.
The following indicators represent real-time model outputs and market-derived expectations that feed into the ECB's policy framework: