Sentiment Arcs Predict Monetary Policy: How Central Banks Structure Their Message Matters More Than Tone

Central Bank Press Conferences Are More Than Information — They're Narrative

Every time the European Central Bank (ECB) or the Federal Reserve (Fed) holds a press conference, markets parse every word for policy signals. Traditional text analysis methods typically focus on overall tone — whether a statement sounds "hawkish" or "dovish," optimistic or cautious about the economic outlook. But a newly published study asks a more nuanced question: what actually carries the policy signal — the average tone of a statement, or the way sentiment is organized and emphasized across the statement as a whole?

The researchers treat central bank communication as a "structured narrative" rather than a simple information release. They argue that how sentiment rises and falls, how it is sequenced and emphasized as the text progresses, is itself a core component of the policy signal — not an optional stylistic flourish.

What Are "Sentiment Arcs"?

The central methodological contribution of this study is constructing "sentiment arcs" for central bank press conferences. Unlike lexicon-based methods that compute a single average sentiment score for an entire document, sentiment arcs track the dynamic trajectory of sentiment as a statement unfolds.

The researchers build these arcs along three dimensions:

• Monetary stance: How language reflecting hawkish or dovish tendencies is distributed throughout the text.
• Economic outlook: How optimistic or cautious expressions about growth, employment, and other economic conditions are positioned.
• Uncertainty: How much emphasis is placed on risks and uncertainty at different points in the statement.

Across these three dimensions, the researchers can characterize the full "shape" of a statement — where it starts, where it intensifies, and how it concludes. This shape is precisely the kind of information that traditional average-tone metrics discard.

---

The lexicon-based method is the most classical approach to sentiment analysis in natural language processing: a pre-built vocabulary assigns positive or negative weights to individual words, and the overall sentiment score is computed as the average across the full text. This approach is computationally simple, highly interpretable, and has long been the dominant tool in central bank communication research. Representative lexicons include the Loughran-McDonald dictionary, designed for financial language, and the more general Harvard IV dictionary. Its core limitation is that it reduces a text to a single scalar, discarding all information about how sentiment is distributed across time — regardless of whether positive sentiment is concentrated at the beginning or end, or whether peaks appear in the policy outlook section or the risk assessment section, the final score may be identical. Sentiment arc methods are designed precisely to address this limitation: they preserve the sequential structure of sentiment, treating a statement as a continuous signal that evolves with position in the text rather than a number that can be collapsed.

---

Arc Shape Predicts Interest Rate Decisions

The study's central finding is that the shape of sentiment arcs robustly predicts changes in policy interest rates — and that this predictive power exceeds lexicon-based baseline methods. This holds for both the ECB and the Fed.

In other words, simply knowing that a statement "sounds hawkish on average" or "sounds optimistic about the economy" is insufficient to capture the full policy signal. What matters is how those sentiments are sequenced and emphasized across the statement. Two statements with similar average tones but different sentiment trajectories may signal different policy directions.

This finding directly challenges common practice in macroeconomic and financial market text analysis: widely used average-sentiment scoring methods may systematically discard critical signals embedded in the structure of the communication.

Receiver-Side Effects: How Forecasters Update Their Expectations

Beyond the direct policy signal, the study reveals a "receiver-side effect." Sentiment arc features matter not just for what the central bank intends to communicate, but for how professional forecasters update their own judgments in response.

The study finds that arc features affect:

• Inflation expectation updates: Professional forecasters adjust their inflation expectations based on how sentiment is organized within a statement.
• Degree of disagreement among forecasters: Arc features are also associated with the extent of divergence in opinion across forecasters.

This means the "shape" of communication is not merely a signal carrier — it also shapes how market participants interpret the message. This effect is distinct from the direct policy signal: even with identical policy intent, different communication designs may lead markets to form varying degrees of consensus or disagreement.

---

"Professional forecasters" here typically refers to economists and analysts who participate in the Fed's Survey of Professional Forecasters (SPF) or the ECB's equivalent survey. These surveys regularly collect probabilistic forecasts for macroeconomic variables such as inflation, GDP growth, and unemployment, and serve as an important benchmark for measuring market expectations. "Disagreement among forecasters" is a key metric in central bank communication research: in theory, public communication by a central bank should serve a "coordination" function, nudging market participants toward more aligned expectations. If communication instead amplifies disagreement, it suggests the information transmission contains ambiguity or noise. The finding that sentiment arc features correlate with disagreement implies that the structural design of communication may itself be a systematic source of divergent market interpretation — a result with direct implications for central bank forward guidance practice.

---

Communication Design Is a First-Order Policy Variable

The study's ultimate conclusion carries considerable weight: communication design — the sequencing and emphasis of policy language across a statement — is a first-order feature of the policy signal, not a second-order refinement.

The implication is that how a central bank organizes and presents information is as important as the content itself. For central banks, this means the structural design of press conference language should be treated as part of the monetary policy toolkit, not an afterthought. For analysts and researchers, it suggests that future text analysis tools need to move beyond "average tone" toward "sentiment trajectory" to more fully capture the signals embedded in policy communication.

---

The distinction between "first-order" and "second-order" draws on standard usage in economics and statistics: first-order effects are factors with a direct, significant impact on outcomes; second-order effects are marginal, incremental refinements. Characterizing communication design as a "first-order feature" is a claim that it belongs in the same tier of importance as the policy rate itself or economic forecasts — not as an expressive detail that can be safely ignored. This argument fits within the broader academic tradition of "central bank communication as a policy tool": since the rise of forward guidance in the 2000s, economists have increasingly recognized that central banks' statements about the future path of policy can themselves influence financial conditions, with market effects comparable to actual rate changes. This study pushes the argument further, to the level of internal communication structure — claiming that not just what is said, but how it is arranged, carries first-order significance.

---

Conclusion

This research advances the study of monetary policy communication from "tone" to "trajectory," offering a more granular analytical framework. It reminds us that in the word games of central banking, what matters is not only what is said, but in what order it is said and where the emphasis lands. As natural language processing tools find deeper applications in macroeconomic analysis, structural features like sentiment arcs may become a new standard for decoding central bank intentions.

It should be noted that this article is based on a preprint (arXiv) abstract. Details on dataset size, model specifications, and robustness checks await disclosure in the full paper.