Central bank communication
Selected speech
Each jurisdiction contributes its latest qualifying speech inside the selected window, measured from the snapshot date. The date is when the speech was delivered, with publication as a fallback. Open a jurisdiction for its full archived history.
Both models process the full extracted English text. Policy tone averages the difference between hawkish and dovish probabilities across sentences that match a policy keyword. A speech needs at least such sentences to qualify. Economic sentiment averages positive minus negative probability across all scored sentences. The scale runs from −1 to +1. Zero means the two sides balance.
The selected passages show up to three sentences at either end of the policy scale. The downloadable review sample supports a separate manual check. These model scores describe language. A policy decision requires separate evidence.
The BIS publishes a selection of speeches. Jurisdictions follow BIS speaker metadata. The European Union category can include ECB officials. National speakers in a currency union can discuss a shared monetary policy. Some small jurisdictions have no outline at this map resolution. The date chart includes them whenever they have a qualifying speech.
Speech archive from the BIS. Sentiment model from CentralBankRoBERTa (MIT). Policy model and license details appear in the FOMC model card. Full model revisions are in the run manifest. Map boundaries are bundled from Plotly’s world geography.