evidoria

← Back to browse

Good practice

MILA — Deutsche Bundesbank's AI Agent for Auditing ECB Policy Communication

Germany · Frankfurt am Main · See the Germany profile

Bundesbank researchers built MILA, an LLM agent that classifies the tone of ECB Governing Council communications sentence-by-sentence from 2011–2024 against the prevailing macroeconomic backdrop, testing whether policy messaging is internally consistent.

MILA — Deutsche Bundesbank's AI Agent for Auditing ECB Policy Communication

Details

Promoter
Deutsche Bundesbank
Period
2025 (covering communications 2011–2024)
Keywords
large language models, monetary policy analysis, text classification, central banking, policy evaluation

Context

The Deutsche Bundesbank developed MILA (Monetary-Intelligent Language Agent), a large-language-model-based tool that classifies the tone — hawkish/dovish, positive/neutral/negative — of European Central Bank Governing Council press conferences, statements and speeches sentence by sentence, attaching a justification tied to contemporaneous macroeconomic data.

Objectives

Test whether ECB Governing Council communications from 2011 to 2024 were internally consistent with the prevailing macroeconomic environment, using automated, sentence-level tone classification.

Activities

Published as Bundesbank Technical Paper 01/2025 (author Felix Geiger), MILA was applied to the full run of ECB communications from 2011 to 2024. The paper and its methodology are openly published on the Bundesbank's own site, independently covered by trade press, and indexed in the RePEc/IDEAS academic archive.

Results

MILA's analysis concluded that ECB messaging over the 13-year period was broadly consistent with the macroeconomic environment.

Conclusions

MILA evaluates a peer institution's communications (the ECB's) rather than the Bundesbank's own policy decisions, so it functions as an external analytical and monitoring tool rather than a direct decision input; there is no evidence yet that its outputs feed into a formal internal review or policy process at the Bundesbank itself.

Implementation

Indicative cost
Low (< €50k) — Not itemised in the source; developed in-house by Bundesbank research staff as a technical-paper research project rather than a deployed operational system.
Time to results
Short (< 1 year) — Technical paper published in 2025, analysing ECB communications retrospectively from 2011 to 2024.
Staffing & skills
Bundesbank in-house researchers (author Felix Geiger)

Conditions for success

  • Access to the full historical corpus of ECB Governing Council communications
  • Open publication of methodology for external scrutiny

Common failure modes

  • No evidence yet that outputs feed into a formal internal review or policy process at the Bundesbank
  • Evaluates a peer institution's communications rather than the Bundesbank's own decisions

Where it fits

Governance type
national central bank (Eurosystem member)
Scale
single research application
Income level
high-income

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful