CEU Electronic Theses and Dissertations, 2025
| Author | Szekér, Zita Gréta |
|---|---|
| Title | Media as a Signal: Evaluating the Contribution of Economic News Sentiment to Predicting EUR/HUF Movements |
| Summary | This thesis tests whether the inclusion of media sentiment features enhances the predictive accuracy of euro-forint exchange rate movements. The analysis examines the role of economic news sentiment in the context of smaller and less liquid currency markets with the case of Hungary. A multilingual, data-driven framework is developed that extracts sentiment from economic news articles from three media sources: Portfolio.hu, Euronews, and Reuters. Sentiment is extracted from headlines and lead sentences with pre-trained transformer-based models. These sentiment indicators are combined with economic variables and used as inputs to an XGBoost classification model, which is tested with a sliding-window evaluation over the years 2014–2024. The results do not reveal a statistically significant improvement in predictive power from incorporating sentiment features, challenging the strength of media sentiment as a signal in this setting. This result can be an indication of information efficiency, timing mismatch, or structural properties of Hungary’s foreign exchange market. However, feature importance analysis does reveal that sentiment, particularly domestic publications, consistently contributes to model decisions, illustrating the potential for such variables as monitoring tools. In general, the thesis provides a reproducible approach to sentiment-based prediction in smaller markets and contributes to the growing literature on alternative data in financial prediction. |
| Supervisor | Novak, Petra Kralj |
| Department | Undergraduate Studies BA |
| Full text | https://www.etd.ceu.edu/2025/szeker_zita.pdf |
Visit the CEU Library.
© 2007-2025, Central European University