Methodology

Signal design

Together with macro-economic data, bank communication is the main tool shaping forward guidance and market expectations on future monetary policy. Whether explictly or implicitly, typically, central bank officials express their views, assessment of risks in a way that may indicate which way their rate action vote will lean in future monetary policy meetings. Market participants try to take positions ahead of these events but depending on the instance, voting officials' communication could deviate from expectations. Hawkdove aims to provide an unbiased and systematic description of the former.

Here is a non-exhaustive list of questions hawkdoev tries to answer:

• Where does each governor sit today? On a -100(dovish, easier monetary policy) to +100 scale (hawkish, tighter monetary policy), hawkdove assigns a number (a score) to a given statement by a central bank official. Such score synthesises the monetary policy statement in the context of what the official said during the same event, said prior to the event, as well as relative to what other officials said about the same topic. The score balances the statement against market prices/expectations as well as current macro-economic forecasts up to three years in the future.
• Which way are they moving? A drift statistic is calculated to assess whether the speaker is moving in the hawk-dove scale, in which direction and with what intensity. Such statistic appears in the form of a time-series specific to the speaker as well as numerical value computed taking into account time decay and matching the speaker's views by topic.
• How dense, informationally, a speaker's communication typically is? A communication-activity measure captures how frequently a governor is in the record. A stance shift from a governor who rarely speaks carries different weight from routine commentary.
• How explicit is the guidance? A separate signal is extracted from text in the cases where speakers unambigously make statements about policy or policy-related topics such as neutral/terminal rates

Signals are scored at the level of the individual communication first, then aggregated to the governor and to the bank. Keeping the atomic unit at the document level means every bank-level reading can be traced back to the specific speeches and quotes that produced it, rather than being an opaque top-down index.

Data sourcing

All source material is drawn from primary, publicly available central bank publications — official speeches, on-the-record interviews, testimony, and policy statements published directly by the monetary authority. We do not score second-hand summaries, news paraphrasing, or wire reports; the object of analysis is the policymaker's own words as released for public information.

Communications are ingested close to publication, timestamped, and attributed to a named speaker before scoring. Each scored entry retains its link back to the original source document, its publication date, and the specific passages that drove its reading, so any aggregate can be audited against the underlying text.

Paternoster MACRO is an independent research platform. It is not affiliated with, endorsed by, or connected to any central bank or monetary authority.

Scope and limitations

Coverage is currently focused on the European Central Bank and its Governing Council. Additional banks are added deliberately rather than broadly, so that scoring quality and speaker attribution are validated before a bank is published.

What the signal is, and is not, designed to do:

• It is a measure of communicated stance, not a forecast of the next policy decision. Governors can and do surprise their own prior communication.
• Sentiment scoring of policy language is inherently interpretive. Readings are best used as a comparative and directional tool — across speakers, and against a speaker's own history — rather than as absolute cardinal values.
• Aggregates reflect the material available at the time of computation. Thinly-covered speakers or quiet periods produce less stable readings, which is why communication activity is surfaced alongside stance.
• Some signals — notably the thematic watch digest and the explicit rate-bias signal — are in active development and are labelled as such in the interface until their coverage stabilises.

The scoring rubric, quote-selection criteria, weighting, and calibration that sit behind these signals are proprietary. We publish the design intent of every metric — what it is meant to capture and why — but not its internal construction.

Glossary

Plain definitions of the terms used across the dashboard and chart axes.

Stance score

A governor's aggregate position on the hawkish–dovish axis, combining all of their scored communications. Positive values lean hawkish (favouring tighter policy), negative values lean dovish (favouring easier policy). Plotted on the horizontal axis of the governor scatter. Descriptive labels (e.g. Hawkish, Slightly dovish) are banded summaries of the underlying continuous value.

Communication activity

How frequently a governor appears in the scored record. Plotted on the vertical axis of the scatter. Higher activity indicates a governor who is currently communicating often; it contextualises how much weight to place on their stance reading.

Drift

The directional shift in a speaker's stance over their recent communications, relative to their own prior baseline. Drift answers "which way are they moving," independent of where they currently sit. Computed at both the individual-entry and governor level.

Drift quotes

The specific passages within a communication that drove its reading. Each is retained with its contribution and its relationship to the speaker's prior average, so an aggregate signal can be traced back to the exact language behind it.

Rate signal

An explicit forward-guidance bias on policy rates, kept separate from the general sentiment score so that stated intent and overall tone are not conflated. Populated only where a speaker gives direct guidance; otherwise unset.

Watch themes

A thematic summary of the variables and conditions policymakers are signalling attention to — for example specific data dependencies — grouped by theme and linked back to the speeches that raised them. In active development.