Date of Award

2026

Document Type

Thesis

Degree Name

Master of Science in Artificial Intelligence

Department

Digital Engineering

Committee Chair and Members

Kewei Li

Keywords

EUR/USD, Exchange-rate forecasting, Financial news sentiment, FinBERT, Technical indicators, XGBoost

Abstract

Forecasting exchange-rate movements is a challenging task because currency prices are influenced not only by macroeconomic and financial variables but also by market sentiment reflected in financial news. This thesis examines whether financial news headlines can be used to predict the next-day directional movement of the EUR/USD exchange rate by applying finance-specific natural language processing and machine learning techniques.

The study uses a dataset of approximately 466,000 finance-related English-language news headlines collected between 2021 and 2025, aligned with daily EUR/USD closing prices. After preprocessing and temporal alignment, the data are used to construct a binary classification task in which the objective is to predict whether the EUR/USD exchange rate will move upward or downward on the following trading day. Two primary modelling approaches are evaluated. The first is FinBERT, a transformer-based language model adapted for financial text, used in a sentiment-only setting based on aggregated daily news headlines. The second is an XGBoost classifier that combines sentiment-derived variables with technical indicators extracted from the EUR/USD price series. The study also considers baseline comparisons and evaluates performance using accuracy, weighted F1-score, AUC-ROC, and statistical significance tests.

The results show that the sentiment-only FinBERT configuration achieved modest predictive performance, with an accuracy of 0.56 on both the validation and test sets, indicating that textual sentiment alone provided limited predictive power in the present setting. In contrast, the XGBoost model with technical-indicator-enhanced features achieved substantially stronger results, with validation and test accuracies of 0.816 and 0.813, respectively, together with strong class balance and an AUC-ROC of 0.875 on the test set. The findings suggest that financial news sentiment becomes significantly more informative when interpreted alongside recent market structure, volatility, and price-action features.

Overall, this thesis demonstrates that financial news contains useful predictive information for exchange-rate forecasting, but that textual signals alone are insufficient for strong next-day EUR/USD prediction. The study contributes to the literature by comparing finance-specific language modelling with feature-based machine learning on a common dataset and by showing that a hybrid approach combining sentiment and technical information provides the most effective forecasting performance.

Download

Share

COinS