.svg)
.svg)
.svg)
In the noisy world of online discussions, those who speak the loudest or attract the most attention are not necessarily those who provide the most accurate or relevant information. This observation holds especially true in the financial sphere, where some of the most visible public figures have, for the past two decades, tended to adopt systematically bearish outlooks that have proven consistently inaccurate.
To address this issue, Lancrey-Javal Technologies has developed a methodology grounded in the principles of opinion research and sociological sampling, as traditionally practiced by leading opinion polling institutes, in order to isolate the most competent and best-informed voices rather than those that are merely the most influential.
To achieve this, X (formerly Twitter) stands out as the platform of choice, as it has long been regarded as the primary network for finance professionals and real-time economic discourse.
The X platform has emerged as a key lever in the distribution and circulation of financial information. A Twitter study from 2021 showed that the number of financial discussions in the United Kingdom nearly doubled since the pandemic began, and that 36% of users report that the economic news they see on X influences their investment decisions.
In France, a survey conducted by the French Asset Management Association (AFG) confirms this trend:
Beyond frequency of use, recent academic research demonstrates a significant correlation between sentiment expressed on X and stock market movements, credit spreads, monetary policy expectations, and even inflation forecasts (Zeitun et al., 2023; Adams et al., 2023; Angelico et al., 2022).
This convergence of professional adoption, informational relevance, and measurable market impact makes X the most representative and analytically rich platform for studying real-time financial sentiment.
To isolate the most competent and best-informed voices, we monitor five well-defined professional groups whose interactions and discourse form the basis of our financial sentiment analysis:
The constitution of these communities follows two main criteria:
Editorial Coherence — The thematic orientation and discourse of each account are examined in detail to ensure alignment between the claimed professional identity and the content actually published. This includes not only original posts but also retweets, replies, and shared material.
Intra-Community Influence — Accounts are further evaluated by the degree of interaction and recognition they maintain with peers from the same professional category. This peer validation ensures that selected voices belong to active professional ecosystems where expertise circulates and is collectively recognized.
These criteria deliberately exclude popularity-based metrics such as follower count or engagement volume, which favor visibility over substance. The resulting datasets reflect professional legitimacy rather than algorithmic prominence.
The dataset currently used by Narval AI includes over 200 rigorously selected accounts across the five professional categories. Collectively, these accounts produce between 20 000 and 30 000 messages per month, all processed and analyzed in real time by the model.
Follower counts range from a few hundred to over 500 000, with an average around 100 000, reflecting both the diversity and representativeness of the professional spectrum.
Over time, some accounts may become inactive or private; in such cases, a predefined pool of replacements automatically substitutes the affected profiles. Replacements are non-retroactive, ensuring that past data remain historically consistent and analytically traceable.
Establishing a real-time panel of financial professionals on Twitter by rigorously applying the principles of opinion research offers several methodological advantages.
The first is a significant reduction in conversational noise, which is particularly pervasive on Twitter, where algorithmic dynamics tend to amplify the loudest or most polarizing messages at the expense of the most relevant and substantive ones. The prior construction of coherent socio-professional communities makes it possible to avoid one of the main pitfalls of conventional social listening tools — namely, the aggregation of large volumes of undifferentiated conversations, or even the weighting of messages according to engagement metrics such as likes, retweets, or replies.
The second advantage, which follows directly from the first, lies in the isolation of discourse produced by the most knowledgeable and credible professionals within each domain, whose expertise is recognized by their peers. As a result, the language model processes only information that is epistemically valid — originating from credible sources — and therefore exhibits a high degree of predictive potential.
The combination of sociological sampling and advanced linguistic modeling thus constitutes a major methodological innovation in financial data science, one that has only become technically feasible in the era of large language models.
.svg)
.svg)
.svg)
