## 15. ECB’s backward-looking communication index: Robustness to dictionary

## Source details

**Canonical URL:** [15. ECB’s backward-looking communication index: Robustness to dictionary](https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026133-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2026/english/wpiea2026133-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2026/english/wpiea2026133-source-pdf.pdf.json)

---

### I. Introduction — motivation and questions
- Successive large shocks raised uncertainty to historically high levels, complicating forecasting and weakening traditional policy signals.
- Central bank communication is increasingly important to anchor and guide expectations.
- Four key questions:
  - Do central banks respond to heightened uncertainty with more cautious/data-dependent (backward-looking) or more assertive/forward-looking messaging?
  - Do reactions differ by source of uncertainty (monetary-policy-influenced like inflation uncertainty vs. largely exogenous like financial volatility)?
  - Do emerging market (EM) central banks react differently from advanced economy (AE) counterparts?
  - How do central banks’ own assessments of uncertainty compare with market participants’ perceptions?

### II. Survey of central bank communication practices in Europe
- Sample: twenty European central banks; formal toolkits similar but intensity and transparency vary.
- Typical toolkit: monetary policy statements (MPS), monetary policy reports (MPRs), meeting minutes/transcripts, press conferences, online presence, social media.
- AE vs EM differences:
  - AE: more likely to publish meeting minutes, hold press conferences, publish scenarios, fan charts, projected interest rate paths.
  - EM: most publish MPRs and MPSs and maintain online presence; around one-third do not regularly hold press conferences.
- Content differences:
  - All AE and more than 80 percent of EM central banks publish quantitative forecasts or projections for inflation or interest rates.
  - Over 60 percent in either group provide explicit forward communication regarding the likely future path of monetary policy.
  - Fewer than 10 percent of EM central banks publish forward-looking scenarios with explicit policy assumptions, narratives, and simulated economic paths.

### III. Construction of communication indices (methodology)
- Focus documents: monetary policy statements (MPS) for cross-country comparability.
- Indices for 11 central banks: ECB and national central banks of the United Kingdom, Iceland, Israel, Sweden, the Czech Republic, Hungary, Poland, Serbia, Romania, and Türkiye.
- Text-mining frequency approach using AI-assisted dictionaries with expert checking.
- Dictionaries (examples as used):
  - Backward-looking: “Assessment”, “Data-dependent”, “Historical”, “Indicator”, “Data dependent”, “Previous”, “Past”, “Current framework”, “experience”, “Cyclical drivers”, “Current data”, “Current Inflation”
  - Forward-looking: “Outlook”, “Projected”, “Expectation”, “Guidance”, “Future path”, “Expect”, “Future”, “Projection”, “Guiding us forward”, “New frontier”, “Proactive”, “Emerging threat”, “Future”, “Anticipate”, “Prepared for change”, “Flexibility”, “Evolving framework”, “Stabilizing inflation”, “Timely return”, “Commitment to target”, “Expectation”, “Path”, “Commitment”, “Inflation expectation”, “Forecast”
- Index definitions:
  - Forward-/backward-looking index = 100 × (N. of sentences containing dictionary words) / (Total N. of sentences in the monetary policy statement)
  - Net forward-looking index = Forward-looking index − Backward-looking index
- Interpretation notes:
  - Forward-looking index captures information-based forward-looking communication, not necessarily commitment-based forward guidance.
  - Net index captures relative emphasis; forward- and backward-looking language can co-occur.

### IV. Stylized facts on central bank communication
- AE central banks tend to make greater use of forward-looking language relative to EMs.
  - Examples: ECB net forward-looking index rises during COVID‑19 and declines during ECB’s hiking cycle; Czech National Bank shows downward trend post-COVID‑19; Bank of England increased forward-looking focus since the 2024 Bernanke review.
- Forward- and backward-looking communication often used simultaneously, especially around policy turning points; EM central banks often show clearer separation.
- Communication indices are persistent/inertial (e.g., about 0.5 one-year serial correlation for Bank of England and ECB forward-looking index), motivating lagged dependent variables in empirical models.

### V. Empirical analysis — how central banks react to uncertainty
- Dynamic panel specification (monthly):
  - NetIndex_i,t = α_i + β1 NetIndex_i,t−3 + β2 VSTOXXX_t−1 + β3 Π_U,t−1 + β4 Policyrate_t−1 + ε_i,t
  - Uncertainty measures include VSTOXX (financial volatility), Π_U (inflation expectations uncertainty proxied by forecaster disagreement), and policy rate.
- Baseline (Table 1, dynamic model) key coefficients and statistics:
  - Lag net forward-looking index: 0.565 (full sample, p<0.01); 0.547 (AE, p<0.01); 0.526 (EM, p<0.01).
  - Inflation uncertainty:
    - Full sample: -0.183 (p<0.10).
    - AE: 2.963 (p<0.01).
    - EM: -0.350 (p<0.01).
  - Policy rate:
    - Full sample: -0.059 (p<0.10).
    - AE: -0.801 (p<0.01).
    - EM: -0.009 (not significant).
  - Lag financial volatility:
    - Full sample: 0.005 (not significant).
    - AE: -0.104 (p<0.01).
    - EM: 0.072 (not significant).
  - Economic importance (1SD effects reported for AE):
    - 1 standard deviation increase in inflation uncertainty increases net forward-looking language by 3.41.
    - Policy rate: -4.5 effect per 1 standard deviation change.
    - Financial volatility: -0.86 effect per 1 standard deviation change.
  - Observations and fit:
    - Observations: 1512 (full), 756 (AE), 756 (EM).
    - Adj. R-squared: 0.757 (full), 0.860 (AE), 0.450 (EM).
- Rolling-window results:
  - AE central banks’ response to inflation uncertainty changed over time: 2015–2020 less sensitive; post-pandemic and 2020s more forward-looking and conditional.
  - Financial volatility term negative mainly in 2020–2024.
- Expanded model (Table 2) adding GDP uncertainty and EPU:
  - Inflation uncertainty remains primary driver.
  - GDP uncertainty insignificant.
  - Lag EPU: AE coefficient 0.007 (p<0.05) in one specification.
- Decomposition (Table 3) — forward- vs backward-looking indices:
  - Forward-looking index:
    - Lag forward index: ~0.593 (full); 0.601 (AE); 0.550 (EM).
    - Lag financial volatility: negative and significant (e.g., -0.102 full; -0.177 AE; -0.072 EM).
    - Lag inflation uncertainty: AE 2.731 (p<0.01); EM -0.338 (p<0.01).
  - Backward-looking index:
    - Lag backward index: ~0.531 (full); 0.475 (AE); 0.536 (EM).
    - Financial volatility associated with reduced backward-looking communication in full sample and EM (e.g., -0.142 EM, p<0.01).
  - Interpretation:
    - AE: higher inflation uncertainty → more forward-looking communication.
    - EM: higher inflation uncertainty → less forward-looking (more backward-looking) communication.
    - Financial volatility generally reduces communication intensity.
- Anchoring hypothesis — ECB case (Table 4):
  - Market-based monetary policy uncertainty measured as monthly standard deviation of daily market pricing for the policy rate two months ahead.
  - OLS finds net forward-looking index positively associated with monetary policy uncertainty (possible reverse causality).
  - MM-estimator (robust to outliers) yields net forward-looking index coefficient -0.450 (p<0.01), suggesting increased forward-looking communication associated with lower market interest-rate expectation volatility after accounting for outliers.
  - Controls include policy rate, inflation, financial volatility, federal funds rate; observations: 78.

### VI. How central banks communicate uncertainty (central bank uncertainty index — CBUI)
- CBUI construction:
  - Dictionary: “uncertainty”, “uncertain”, “unclear”, “lack of clarity”, “ambiguity”, “unknown”, “unpredictability”, “volatile”, “volatility”.
  - CBUI = frequency of phrases using any dictionary word / total number of phrases in the MPS.
  - CBUI available for 11 countries (five AEs, six EMs).
- Empirical patterns:
  - CBUI spikes coincide with large shocks: Global Financial Crisis, Brexit, COVID-19, tariff announcements.
  - Increasing trend over the last decade.
  - Cross-country correlations limited on average; strong temporary co-movements around global shocks.
  - Relative importance exercise (Figure 13): EPU is the largest driver of CBUI, followed by level and standard deviation of inflation.
- Comparison with market-based monetary policy uncertainty (Riksbank and ECB):
  - Low correlation over full sample, but correlation rose markedly post-pandemic (notably during 2022 hiking cycle), especially for Sweden.
  - Recent divergence after November 2024: markets perceive little near-term uncertainty about central bank actions while MPSs still convey elevated economic uncertainty.
  - Suggestion: scenario analysis and other tools may help anchor expectations amid elevated economic uncertainty.

### VII. Robustness exercises and appendix findings
- Static specification (A.2, Table 6):
  - All sample: no variable statistically significant; examples of coefficients:
    - Lag financial volatility: 0.051 (0.076)
    - Lag GDP uncertainty: -0.637 (1.844)
    - Lag EPU: 0.005 (0.007)
    - Lag inflation uncertainty: -0.237 (0.452)
    - Lag policy rate: -0.110 (0.115)
    - Constant: 9.461*** (1.616)
    - Observations: 1528; Adj. R-squared: 0.645
  - AE (column 2): lag inflation uncertainty 7.251*** (0.543); lag policy rate -0.874* (0.275); Observations: 57; Adj. R-squared: 0.414
  - EM (column 3): no significant variables; Observations: 395; Adj. R-squared: 0.690
- 12-month-difference specification (A.3, Tables 7–9):
  - All sample: lag uncertainty index coefficients around 0.491–0.541 (significant in some specs); lag financial volatility small and insignificant.
  - Advanced economies: lag uncertainty index coefficients up to 0.982** (0.212); lag inflation uncertainty large positive (e.g., 4.987* (1.687)); lag policy rate -1.797** (0.351) in some specs.
  - Emerging/developing: lag uncertainty index around 0.421–0.432 (some significance); lag inflation uncertainty smaller and less robust.
- Controlling for the zero lower bound (A.4, Table 10):
  - Lag net forward-looking index: 0.558*** (0.033) all; 0.509*** (0.058) AE; 0.529*** (0.041) EM.
  - Lag financial volatility: 0.018 (0.033) all; -0.122*** (0.036) AE; 0.056 (0.048) EM.
  - Lag inflation uncertainty: -0.281** (0.137) all; 3.303*** (0.812) AE; -0.436*** (0.136) EM.
  - Lag policy rate: -0.043 (0.034) all; -0.992*** (0.185) AE; -0.005 (0.038) EM.
  - Zero lower bound dummy: 3.267* (1.867) all; Observations: 1512; Adj. R-squared: 0.758.
  - Finding: main results robust to controlling for ZLB.
- Robustness to different dictionary for ECB index (A.5):
  - Alternative backward-looking index excluding data-dependence terms is lower by construction but co-moves strongly with original index.
  - Correlation between original and alternative index: 0.85.
  - ECB increasingly used data-dependence language in recent years; upward trend in backward-looking communication remains evident even after excluding data-dependence terms.
  - Policy context: in July 2022 the ECB discontinued commitment-based forward guidance and adopted a meeting-by-meeting, data-dependent approach.

### VIII. Key substantive findings (bulleted summary)
- Communication toolkits similar across European central banks, but usage intensity and transparency vary between AEs and EMs.
- AE central banks more likely to use forward-looking tools (scenarios, fan charts, projected interest rate paths); EM central banks more selective.
- Forward- and backward-looking language commonly co-occur; net forward-looking index isolates relative emphasis.
- Communication indices are persistent; lagged values strongly predict current indices.
- Inflation expectations uncertainty is a primary determinant:
  - AE: higher inflation uncertainty → more forward-looking communication (information-based).
  - EM: higher inflation uncertainty → more backward-looking communication.
- Financial market volatility tends to reduce forward-looking communication (negative coefficients), especially in AE sample during 2020–2024.
- Economic policy uncertainty (EPU) increases forward communication in AE central banks.
- Anchoring evidence (ECB): MM-estimator result shows net forward-looking index coefficient -0.450 (p<0.01), indicating increased forward-looking communication associated with lower market interest-rate expectation volatility after accounting for outliers.
- Central banks’ own uncertainty communications (CBUI) align more with global economic policy uncertainty than exclusively with inflation or financial uncertainty.

### IX. Policy implications and recommendations
- Use conditional, information-based forward-looking communication to anchor expectations when inflation uncertainty rises, particularly for AE central banks with credibility and forward-looking tools (scenarios, projections).
- For EM central banks with more limited credibility or tools, clearer explanation of past data and data-dependent communication may be optimal; strengthening institutional credibility could expand scope for forward-looking communication.
- Scenario analysis is a valuable tool to convey conditional policy responses and clarify reaction functions; AE experience suggests growing use but empirical assessment remains preliminary due to short history.
- Avoid overuse of forward-looking commitments in periods of high financial market volatility when forecasts are noisy; emphasize conditionality and clarity about how incoming data will affect policy.
- Enhance transparency and frequency of comparable documents (MPS) to facilitate cross-country comparability and public communication.

*Source: Excerpt from wpiea2026133-source-pdf (15. ECB’s backward-looking communication index: Robustness to dictionary).*

### References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  28

### References

### Appendices
- A.   Appendix. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  31
- A.1. Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  31
- A.2. Robustness with static equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  31
- A.3. Robustness with equation in 12-months difference. . . . . . . . . . . . . . . . . . . . . .  32
- A.4. Robustness controlling for the zero lower bound. . . . . . . . . . . . . . . . . . . . . . . .  35
- A.5. Robustness to different dictionary: the ECB index. . . . . . . . . . . . . . . . . . . . . .  36

### List of Tables
- 1.   Dynamic model - lag net forward-looking index. . . . . . . . . . . . . . . . . . . . . . . . .  15
- 2.   Dynamic model - lag net forward-looking index with more uncertainty measures. . . . . . . . .  18
- 3.   Correlates of forward-/backward-looking communication - decomposing the net forward-looking index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  19
- 4.   Anchoring of market interest rate expectations. . . . . . . . . . . . . . . . . . . . . . . .  21
- 5.   Description of variables used among different specifications. . . . . . . . . . . . . . . . . . .  31
- 6.   Static model - Net forward-looking index with more uncertainty measures. . . . . . . . . . . . . .  32
- 7.   Explaining the change in net forward-looking index all sample. . . . . . . . . . . . . . . . . .  33
- 8.   Explaining the change in net forward-looking index advanced. . . . . . . . . . . . . . . . . .  34
- 9.   Explaining the change in net forward-looking index developing. . . . . . . . . . . . . . . . .  34
- 10.  Dynamic model - lag net forward-looking index with more uncertainty measures. . . . . . .  36

### List of Figures
- 1.   Financial market and economic policy uncertainty. . . . . . . . . . . . . . . . . . . . . . . .   4
- 2.   Central banks’ use of key communication tools. . . . . . . . . . . . . . . . . . . . . . . . . .   8
- 3.   Content of central banks’ monetary policy communication. . . . . . . . . . . . . . . . . . . .   8
- 4.   Distribution of the net forward-looking index across country groups. . . . . . . . . . . .  11
- 5.   Net forward-looking index evolution over time for selected countries. . . . . . . . . . . . .  11
- 6.   Correlations between backward- and forward-looking communication indices. . . . . . . .  12
- 7.   Inertia in the forward-looking communication index. . . . . . . . . . . . . . . . . . . . . . .  13
- 8.   AE central banks’ communication responses to inflation uncertainty over time. . . . . . . .  16
- 9.   Monetary policy uncertainty and net forward-looking communication. . . . . . . . . . . . . .  20
- 10.  Central bank uncertainty aligns with large shock events. . . . . . . . . . . . . . . . . . . . .  22
- 11.  Cross-country correlation of the CBUI and policy rates. . . . . . . . . . . . . . . . . .  22
- 12.  Rolling correlation of CBUI. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  23
- 13.  What drives central banks to talk about uncertainty. . . . . . . . . . . . . . . . . . . . . . .  24
- 14.  CBUI Highly Correlated with Policy Rate Pricing Uncertainty During Certain Periods. . . .  25

*Source: wpiea2026133-source-pdf - References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  28*

### 15.  ECB’s backward-looking communication index: Robustness to dictionary. . . . . . . . . . . . . .  37

### 15.  ECB’s backward-looking communication index: Robustness to dictionary

### I. Introduction — motivation and questions
- Successive large shocks (pandemics, armed conflicts, geopolitical fragmentation, trade tensions) pushed uncertainty to historically high levels.
- Central banks face the challenge that elevated uncertainty complicates forecasting, weakens signals of traditional policy tools, and may reduce conventional monetary policy effectiveness.
- Central bank communication is therefore increasingly important to anchor and guide expectations.
- Four key questions the paper addresses:
  - Do central banks respond to heightened uncertainty with more cautious/data-dependent (backward-looking) or more assertive/forward-looking messaging?
  - Do reactions differ by source of uncertainty (monetary-policy-influenced like inflation uncertainty vs. largely exogenous like financial volatility)?
  - Do emerging market (EM) central banks react differently from advanced economy (AE) counterparts?
  - How do central banks’ own assessments of uncertainty compare with market participants’ perceptions?

### II. Survey of central bank communication practices in Europe
- Sample: twenty European central banks; formal communication toolkits similar but intensity and transparency vary.
- Typical toolkit items: monetary policy statements (MPS), monetary policy reports (MPRs), meeting minutes/transcripts, press conferences, online presence, social media.
- AE vs EM differences (summary):
  - AE central banks: more likely to publish meeting minutes, hold press conferences, publish scenarios, fan charts, projected interest rate paths.
  - EM central banks: more selective; most publish MPRs and MPSs and maintain online presence, but meeting minutes/transcripts and regular press conferences less common; around one-third do not regularly hold press conferences.
- Content differences:
  - All AE central banks and more than 80 percent of EM central banks publish quantitative forecasts or projections for inflation or interest rates.
  - Over 60 percent in either group provide explicit forward communication regarding the likely future path of monetary policy.
  - Fewer than 10 percent of EM central banks publish forward-looking scenarios with explicit policy assumptions, narratives, and simulated economic paths.

### III. Construction of communication indices (methodology)
- Focus documents: monetary policy statements (MPS) for cross-country comparability.
- Indices constructed for 11 central banks with sufficient MPSs: the ECB and national central banks of the United Kingdom, Iceland, Israel, Sweden, the Czech Republic, Hungary, Poland, Serbia, Romania, and Türkiye.
- Approach:
  - Text-mining frequency approach using AI-assisted dictionaries with expert checking.
  - Dictionary selection: sample MPSs queried with GPT prompt “what are the key words in this statements that indicate forward (backward)-looking statements”; experts filtered the resulting keywords.
- Backward-looking dictionary (as used):
  - “Assessment”, “Data-dependent”, “Historical”, “Indicator”, “Data dependent”, “Previous”, “Past”, “Current framework”, “experience”, “Cyclical drivers”, “Current data”, “Current Inflation”
- Forward-looking dictionary (as used):
  - “Outlook”, “Projected”, “Expectation”, “Guidance”, “Future path”, “Expect”, “Future”, “Projection”, “Guiding us forward”, “New frontier”, “Proactive”, “Emerging threat”, “Future”, “Anticipate”, “Prepared for change”, “Flexibility”, “Evolving framework”, “Stabilizing inflation”, “Timely return”, “Commitment to target”, “Expectation”, “Path”, “Commitment”, “Inflation expectation”, “Forecast”
- Index definitions:
  - Forward-/backward-looking index = 100 × (N. of sentences containing dictionary words) / (Total N. of sentences in the monetary policy statement)
  - Net forward-looking index = Forward-looking index - Backward-looking index
- Interpretation notes:
  - Forward-looking index captures information-based forward-looking communication (outlook, projections, expectations, possible policy paths), not necessarily commitment-based forward guidance.
  - Net index captures the relative emphasis between forward- and backward-looking communication; forward- and backward-looking language can co-occur.

### IV. Stylized facts on central bank communication
- Stylized fact I: AE central banks make greater use of forward-looking language
  - Distribution of net forward-looking index shows AE central banks trend toward more forward-looking language relative to EMs.
  - Country examples:
    - ECB: net forward-looking index rises during COVID‑19, declines during ECB’s hiking cycle with increased data-dependence.
    - Czech National Bank: downward trend in net forward-looking index post-COVID‑19 (more backward-looking).
    - Bank of England: similar trends but notable increase in forward-looking focus since the 2024 Bernanke review with greater use of scenario analysis.
- Stylized fact II: Forward- and backward-looking communication often used simultaneously
  - Many central banks (particularly AEs) use both styles together, especially around monetary policy decision turning points.
  - EM central banks often show clearer separation (indices uncorrelated or negatively correlated in some cases).
  - Implication: net index abstracts from common movements; separate indices examined as complement.
- Stylized fact III: Persistence/inertia in communication
  - Communication indices are persistent; changes occur gradually.
  - Example: Bank of England and ECB show about 0.5 one-year serial correlation in the forward-looking index.
  - Implication: empirical specifications incorporate lagged dependent variables to capture inertia.

### V. Empirical analysis — how central banks react to uncertainty
- Dynamic panel specification (monthly frequency) estimated:
  - NetIndex_i,t = α_i + β1 NetIndex_i,t−3 + β2 VSTOXXX_t−1 + β3 Π_U,t−1 + β4 Policyrate_t−1 + ε_i,t
  - Explanatory variables lagged one month; dependent variable lagged three months.
  - Uncertainty measures considered include: financial market volatility (VSTOXX), inflation expectations uncertainty (Π_U proxied by forecaster disagreement), policy rate.
- Baseline results (Table 1, dynamic model):
  - Lag net forward-looking index: positive and highly significant (p<0.01) across samples (coefficients: 0.565 for full sample; 0.547 AE; 0.526 EM).
  - Inflation uncertainty:
    - Full sample: coefficient -0.183 (p<0.10).
    - AE: coefficient 2.963 (p<0.01).
    - EM: coefficient -0.350 (p<0.01).
  - Policy rate:
    - Full sample: coefficient -0.059 (p<0.10).
    - AE: coefficient -0.801 (p<0.01).
    - EM: coefficient -0.009 (not significant).
  - Lag financial volatility:
    - Full sample: 0.005 (not significant).
    - AE: -0.104 (p<0.01).
    - EM: 0.072 (not significant).
  - Economic importance (1SD effects reported):
    - AE: 1 standard deviation increase in inflation uncertainty increases net forward-looking language by 3.41.
    - AE: 1 standard deviation change effects: policy rate -4.5; financial volatility -0.86.
  - Observations and fit:
    - Observations: 1512 (full), 756 (AE), 756 (EM).
    - Adj. R-squared: 0.757 (full), 0.860 (AE), 0.450 (EM).
- Rolling-window analysis:
  - AE central banks’ response to inflation uncertainty changed over time:
    - 2015–2020: AE communication less sensitive to inflation uncertainty and sometimes more backward-looking when inflation uncertainty rose.
    - Post-pandemic and 2020s: AE central banks increasingly adopt more forward-looking messaging in response to inflation uncertainty and emphasize conditionality.
  - Financial volatility term negative mainly in 2020–2024 period.
- Expanded model (Table 2) adding GDP uncertainty and economic policy uncertainty (EPU):
  - Inclusion of additional uncertainty measures does not alter main results; inflation uncertainty remains primary driver.
  - GDP uncertainty insignificant; EPU statistically significant in AE communication (positive), suggesting economic policy uncertainty increases forward communication in AEs.
  - Coefficients (selected):
    - Lag EPU: AE coefficient 0.007 (p<0.05) in column (2) of Table 2.
    - Lag GDP uncertainty: not significant.
- Decomposition into forward- and backward-looking indices (Table 3):
  - Forward-looking index:
    - Lag forward index highly persistent (coefficients ~0.593 full; 0.601 AE; 0.550 EM).
    - Lag financial volatility: negative and significant across samples (e.g., -0.102 full; -0.177 AE; -0.072 EM).
    - Lag inflation uncertainty: positive and significant for AE (2.731 p<0.01); negative and significant for EM (-0.338 p<0.01).
  - Backward-looking index:
    - Lag backward index highly persistent (coefficients ~0.531 full; 0.475 AE; 0.536 EM).
    - Financial volatility associated with reduced backward-looking communication in full sample and particularly in EMs (e.g., -0.142 EM, p<0.01).
  - Interpretation:
    - AE central banks: inflation uncertainty → more forward-looking communication.
    - EM central banks: inflation uncertainty → less forward-looking (more backward-looking) communication.
    - Financial market volatility generally reduces both forward- and backward-looking language (communication intensity declines when market volatility increases).
- Anchoring hypothesis (ECB case study):
  - Market-based monetary policy uncertainty measured as monthly standard deviation of daily market pricing for the policy rate two months ahead of a policy decision.
  - OLS regression (Table 4, column 1) finds net forward-looking index positively associated with monetary policy uncertainty (possible reverse causality).
  - MM-estimator (robust to outliers) yields net forward-looking index coefficient -0.450 (p<0.01), suggesting increased forward-looking communication is associated with lower market interest-rate expectation volatility after accounting for extreme observations and reverse causality concerns.
  - Controls in Table 4 include policy rate, inflation, financial volatility, federal funds rate; sample observations: 78.
  - Interpretation: more forward-looking language could help anchor market expectations, but causality concerns exist.

### VI. How central banks communicate uncertainty (central bank uncertainty index — CBUI)
- CBUI construction:
  - Dictionary words: “uncertainty”, “uncertain”, “unclear”, “lack of clarity”, “ambiguity”, “unknown”, “unpredictability”, “volatile”, “volatility”.
  - CBUI = frequency of phrases using any dictionary word / total number of phrases in the monetary policy statement.
  - CBUI available for 11 countries (five AEs, six EMs).
- Empirical patterns:
  - CBUI spikes coincide with large shocks: Global Financial Crisis, Brexit, COVID-19, tariff announcements.
  - Increasing trend over the last decade: greater emphasis on uncertainty in statements.
  - Cross-country correlations of CBUI are limited on average; CBUI co-movements are idiosyncratic with temporary strong co-movements around global shocks.
  - Regression analysis (relative importance exercise, Figure 13):
    - Economic policy uncertainty (EPU) is the largest driver of CBUI.
    - Next most important drivers: level of inflation and its standard deviation (distantly).
    - Interpretation: central banks tend to communicate about uncertainty mostly when general economic policy uncertainty is high, rather than exclusively when uncertainty concerns their own objectives.
- Comparison with market-based monetary policy uncertainty:
  - For the Riksbank and the ECB, CBUI versus market pricing-based monetary policy uncertainty:
    - Correlation low over full sample but rose markedly in the post-pandemic period (notably during the 2022 hiking cycle), particularly for Sweden.
    - Recent divergence (after November 2024): markets perceive little near-term uncertainty about central bank actions while monetary policy statements still convey elevated economic uncertainty.
  - Suggestion: scenario analysis and other communication tools may help anchor expectations even amid elevated economic uncertainty.

### VII. Key substantive findings (bulleted summary)
- Communication toolkits are similar across European central banks, but usage intensity and transparency vary considerably between AEs and EMs.
- AE central banks more likely to use forward-looking tools (scenarios, fan charts, projected interest rate paths); EM central banks more selective.
- Forward- and backward-looking language commonly co-occur; net forward-looking index isolates relative emphasis.
- Communication indices are persistent; lagged values strongly predict current indices.
- Inflation expectations uncertainty is a primary determinant of central bank communication:
  - AE central banks: higher inflation uncertainty → more forward-looking communication (information-based).
  - EM central banks: higher inflation uncertainty → more backward-looking communication.
- Financial market volatility tends to reduce forward-looking communication (negative coefficient), especially in AE sample during 2020–2024.
- Economic policy uncertainty (EPU) increases forward communication in AE central banks.
- Evidence for anchoring:
  - After controlling for outliers, increased net forward-looking communication by the ECB is associated with lower market interest-rate expectation volatility (MM-estimator result).
- Central banks’ own uncertainty communications (CBUI) align most with global economic policy uncertainty rather than solely with inflation or financial uncertainty.

### VIII. Policy implications and recommendations (from analysis)
- Use conditional, information-based forward-looking communication to anchor expectations when inflation uncertainty rises, particularly for AE central banks with existing credibility and forward-looking tools (scenarios, projections).
- For EM central banks with more limited credibility or tools, clearer explanation of past data and data-dependent communication may be optimal; strengthening institutional credibility could expand scope for forward-looking communication.
- Scenario analysis can be a valuable communication tool to convey conditional policy responses and clarify reaction functions; AE experience suggests growing use, but empirical assessment remains preliminary due to short history.
- Avoid overuse of forward-looking commitments in periods of high financial market volatility when forecasts are noisier; instead, emphasize conditionality and clarity about how incoming data will affect policy.
- Enhance transparency and frequency of comparable documents (MPS) to facilitate cross-country comparability and effective public communication.

*Source: Excerpt from wpiea2026133-source-pdf (15. ECB’s backward-looking communication index: Robustness to dictionary).*

### Box 1: Scenario analysis in central bank communication

### Box 1: Scenario analysis in central bank communication

### Rationale and adoption
- Many AE central banks have incorporated scenario analysis to better communicate heightened uncertainty and increase the conditionality of forward guidance.
- Following the 2024 Bernanke Review, the Bank of England progressively integrated scenario analysis into policymaking and communication.
- The ECB and the Riksbank have also adopted scenario analysis, integrating it into monetary policy assessments, stress-testing frameworks, and external communications.

### Features of scenario analysis (general)
- Scenarios are more than shocks; they explore different economic trajectories and structural changes.
- Scenarios:
  - Illustrate alternative economic states, mechanisms, and structural changes.
  - Simulate macroeconomic developments using specific assumptions and shocks.
  - Can be constructed by applying shocks to the central projection and, when necessary, adjusting structural parameters or assumptions about the reaction function.
  - Are often presented symmetrically to show both upside and downside risks.
  - Are accompanied by detailed narratives with outcomes for inflation, GDP, and the policy rate path presented in charts.

### Benefits for policymaking
- Enrich MPC/Executive Board discussions by providing a common framework and discipline for “what if” analysis.
- Foster robust internal debates and test the robustness of the baseline policy by comparing outcomes across multiple plausible states.
- Ensure more robust policy decisions by forcing discussion of alternative responses and reaction functions.

### Benefits for communications
- Transparently acknowledge uncertainty and disclose the reaction function.
- Allow individual committee members to associate with different scenarios to justify their views, providing richer economic storytelling.
- Seen as superior to traditional fan charts where focus tended to be on the middle path.
- Help reconcile the trade-off between guidance and flexibility by conveying uncertainty explicitly rather than suppressing forward-looking information.

### Bank-specific practices and observations
- Bank of England
  - Progressively integrated scenario analysis into policymaking and communication following the 2024 Bernanke Review.
  - Emphasized that scenarios explore different economic trajectories rather than being mere shocks.

- European Central Bank (ECB)
  - Integrated scenario analysis into monetary policy assessments and stress-testing frameworks.
  - Scenarios are derived from macroeconomic models reflecting different economic conditions, such as varying inflation rates or geopolitical tensions.
  - European Central Bank (2025a) underlined that scenarios bring transparency and guard against false certainty.
  - The ECB communicates scenarios in its Economic Bulletin, press conferences and blogs.
  - Since the pandemic, the ECB has published scenario analysis including modeling shocks such as energy, trade, or geopolitical events.
  - European Central Bank (2025b) strategy assessment emphasized scenarios in showing how wars or pandemics shape inflation; example: Ukraine scenario predicted above 7 percent inflation vs baseline of about 5 percent, aligning with realized 8 percent.
  - Key strengths include the comprehensive nature of scenarios covering a wide range of shocks and tail risks; complexity of models implies continued need for a wide range of communication tools to provide clarity.
  - Continued efforts are needed to contextualize scenarios and better communicate on the probability of the scenarios (see Dizioli 2025).

- Riksbank
  - Long-standing tradition of developing and publishing alternative scenarios; systematically incorporated into communication since 2023 and earlier practices since 2007.
  - Explicitly notes that “it is difficult to say in advance how monetary policy will react,” signaling a wider policy reaction function and greater conditional uncertainty.
  - Typically presents two alternative scenarios constructed by applying shocks to the central projection and, when necessary, adjusting structural parameters or reaction-function assumptions.
  - Scenarios reflect the Executive Board’s views on main risks and the Board assumes responsibility for them.
  - Unlike probabilistic fan charts, scenarios provide structured insights into how policy might adapt under different conditions.

### Role in uncertainty management and implications
- Scenario analysis increases the conditionality of forward guidance and clarifies reaction functions, improving transparency and credibility under heightened uncertainty.
- Advanced Economy central banks:
  - Increase forward-looking language relative to backward-looking language when inflation expectations uncertainty rises, highlighting expectations management and use of forecasts.
  - Have moved toward more conditional forward guidance using scenario analysis.
  - May become less forward-looking under high financial market volatility, suggesting reliance on past data when current information is noisy.
- Emerging Market central banks:
  - Do not systematically shift communication styles when faced with increased uncertainty in inflation expectations.
- Credibility considerations:
  - Central banks with well-anchored expectations can provide guidance even under uncertainty; those with weaker credibility may limit forward commitments.
  - Information-based forward-looking communications can help build credibility over time.

### Practice recommendations and open issues
- Use scenario analysis to disclose reaction functions and acknowledge uncertainty explicitly.
- Complement complex model-based scenarios with a wide range of communication tools to ensure clarity for markets and the public.
- Contextualize scenarios and better communicate the probability of each scenario to avoid false certainty.
- Continue testing baseline robustness across multiple plausible states and use scenarios to enhance transparency and credibility.

*Source: Box 1: Scenario analysis in central bank communication (excerpt).*

### APPENDIX A. APPENDIX

### APPENDIX A. APPENDIX

### A.1. Data
- Table 5 lists variables and definitions used across specifications:
  - EPU Index: Measures uncertainty about future government economic policies using newspaper coverage and related data. Source: Baker et al. (2016).
  - VSTOXX: Forward-looking index measuring 30-day implied volatility of EURO STOXX 50 options. Source: STOXX Ltd.
  - GDP Uncertainty: Standard deviation of one-year ahead growth forecasts in the consensus forecast. Source: Consensus Forecast.
  - Inflation Uncertainty: Standard deviation of one-year inflation forecasts in the consensus forecast. Source: Consensus Forecast.
  - Policy Rate: The central bank policy rate. Source: ECB & NCBs.

### A.2. Robustness with static equation
- Static specification estimated (Equation 6) omits the lagged net forward-looking index and includes controls X (EPU and growth uncertainty). Standard errors clustered at the country level; country fixed effects included.
- Main findings from Table 6 (Static model - Net forward-looking index with more uncertainty measures):
  - All sample (column 1): no variable is statistically significant; coefficients include:
    - Lag financial volatility: 0.051 (0.076)
    - Lag GDP uncertainty: -0.637 (1.844)
    - Lag EPU: 0.005 (0.007)
    - Lag inflation uncertainty: -0.237 (0.452)
    - Lag policy rate: -0.110 (0.115)
    - Constant: 9.461*** (1.616)
    - Observations: 1528
    - Adj. R-squared: 0.645
  - Advanced economies (AE, column 2): inflation uncertainty positive and strongly significant; policy rate negative and significant:
    - Lag financial volatility: -0.129 (0.125)
    - Lag GDP uncertainty: -2.965 (1.625)
    - Lag EPU: 0.016 (0.009)
    - Lag inflation uncertainty: 7.251*** (0.543)
    - Lag policy rate: -0.874* (0.275)
    - Constant: 8.830** (1.570)
    - Observations: 57
    - Adj. R-squared: 0.414
  - Emerging economies (EM, column 3): no variable statistically significant; coefficients similar sign to dynamic model:
    - Lag financial volatility: 0.133 (0.086)
    - Lag GDP uncertainty: 0.311 (2.734)
    - Lag EPU: -0.004 (0.008)
    - Lag inflation uncertainty: -0.428 (0.352)
    - Lag policy rate: -0.069 (0.119)
    - Constant: 9.238** (2.445)
    - Observations: 395
    - Adj. R-squared: 0.690
- Interpretation:
  - Advanced-economy central banks: higher inflation expectations uncertainty associated with more forward-looking communication; higher policy rate associated with more backward-looking language.
  - Emerging-economy central banks: no significant inflation-uncertainty-driven increase in forward-looking communication; tendency to shift toward backward-looking, data-dependent communication under inflation uncertainty.

### A.3. Robustness with equation in 12-months difference
- 12-month difference specification estimated (Equation 7): Y_i,t − Y_i,t−12 on changes in lagged VSTOXX, EPU, UncertaintyIndex, and changes in controls (lagged inflation expectations uncertainty, policy rate, GDP growth expectations uncertainty). Country fixed effects; standard errors clustered at the country level.
- Main summary: changes in net forward-looking index are most robustly explained by changes in the uncertainty index.
- Key results from Tables 7–9:
  - All sample (Table 7):
    - Lag uncertainty index coefficients: 0.541*** (0.161); 0.489* (0.255); 0.491* (0.253); 0.492* (0.255) across specifications (columns 1–4).
    - Lag financial volatility: small and insignificant (e.g., 0.004 (0.043)).
    - Lag EPU: close to zero and insignificant (e.g., -0.002 (0.006)).
    - Lag inflation uncertainty: 0.543 (0.323); 0.571* (0.264); 0.685 (0.379) in some specs.
    - Observations across specs: 1583; 1243; 1243; 1243.
  - Advanced economies (Table 8):
    - Lag uncertainty index: 0.982** (0.212); 0.783* (0.256); 0.809* (0.224); 0.689 (0.301).
    - Lag inflation uncertainty: 4.987* (1.687); 4.947 (1.752); 5.072* (1.185) — strong positive effect.
    - Lag policy rate: -1.797** (0.351); -1.783** (0.365) — tightening linked to more backward-looking language.
    - Lag GDP uncertainty: -3.458* (1.065) in one specification.
    - Observations: 65; 45; 95; 19 (columns 1–4 respectively).
  - Emerging economies / developing (Table 9):
    - Lag uncertainty index: 0.432* (0.190); 0.421 (0.311); 0.426 (0.310); 0.420 (0.311).
    - Lag inflation uncertainty: 0.393 (0.201); 0.454** (0.152); 0.384 (0.278) — positive but generally smaller than in advanced economies.
    - Lag policy rate: 0.075 (0.054); 0.072 (0.046) — not significant or opposite sign to advanced economies.
    - Observations: 92; 97; 24; 72 across specs.
- Interpretation:
  - Across the full sample, increases in the uncertainty index (exact coefficients noted above) are associated with shifts toward more forward-looking communication.
  - Advanced economies: the change in net forward-looking index strongly associated with professional forecasters' uncertainty in inflation forecasts (large positive coefficients such as 4.987*); tightening policy rate and higher GDP uncertainty associated with shifts toward backward-looking communication.
  - Emerging economies: positive but smaller and less robust effects of the uncertainty index and inflation-forecast uncertainty on forward-looking communication.

### A.4. Robustness controlling for the zero lower bound
- A dummy for the zero lower bound (ZLB) period included; ZLB observed in sample for Czech Republic, ECB, the United Kingdom, Poland and Sweden.
- Table 10 (Dynamic model - lag net forward-looking index with more uncertainty measures) key coefficients:
  - Lag net forward-looking index: 0.558*** (0.033) all sample; 0.509*** (0.058) AE; 0.529*** (0.041) EM.
  - Lag financial volatility: 0.018 (0.033) all; -0.122*** (0.036) AE; 0.056 (0.048) EM.
  - Lag inflation uncertainty: -0.281** (0.137) all; 3.303*** (0.812) AE; -0.436*** (0.136) EM.
  - Lag policy rate: -0.043 (0.034) all; -0.992*** (0.185) AE; -0.005 (0.038) EM.
  - Interaction terms and other interactions reported (e.g., Interaction inflation uncertainty: 2.996 (2.257) all; 1.162 (2.259) AE; 60.520*** (7.326) EM).
  - Zero lower bound dummy: 3.267* (1.867) all; -0.821 (1.889) AE; 0.594 (1.752) EM.
  - Observations: 1512; 756; 756 (columns 1–3).
  - Adj. R-squared: 0.758; 0.861; 0.454.
- Finding: results do not change much when controlling for the zero lower bound.

### A.5. Robustness to different dictionary: the ECB index
- Motivation: terms related to "data dependence" may be interpreted as forward-looking or as absence of forward guidance; construction of alternative backward-looking index excludes data-dependence-related terms.
- Empirical comparison:
  - The alternative index excluding data-dependence terms is lower by construction but co-moves strongly with the original index.
  - Correlation coefficient between the original index (including data-dependence terms) and the alternative index (excluding them): 0.85.
- Interpretation:
  - ECB has increasingly used data-dependence language in recent years; original index shows a larger increase than the alternative measure.
  - Upward trend in backward-looking communication for the ECB remains evident even after excluding data-dependence terms.
  - Policy context note: in July 2022 the ECB discontinued commitment-based forward guidance and adopted a meeting-by-meeting, data-dependent approach, reflecting a strategic decision to preserve policy flexibility amid high forecast uncertainty.

*Central Bank Communication in Times of Uncertainty: AI-assisted Decoding of Recent Trends in Europe — Working Paper No. WP/2026/133*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026133-source-pdf.pdf_