A Hybrid Sentiment–Technical Framework for Forecasting Dow Jones Industrial Average Movements Using FinBERT and XGBoost

Abstract

This study proposes a hybrid machine learning framework that integrates financial sentiment analysis with traditional technical indicators to predict daily directional movements of the Dow Jones Industrial Average (DJIA). Sentiment features are extracted from financial news headlines using FinBERT, a transformer-based language model fine-tuned for the financial domain, and aggregated into daily positive, neutral, and negative sentiment scores. These sentiment features are combined with key technical indicators, including the Exponential Moving Average (EMA), Relative Strength Index (RSI), Average True Range (ATR), and lagged trading volume, to form a unified feature set. The integrated features serve as inputs to an eXtreme Gradient Boosting (XGBoost) classifier, which is trained on historical DJIA data to forecast next-day market direction. Experimental results on the evaluation dataset indicate that the hybrid model surpasses sentiment-only and technical-only baselines, achieving perfect classification accuracy. However, this unusually high performance suggests potential overfitting or data leakage, indicating the need for further validation. Overall, the findings highlight the potential of combining transformer-based natural language processing with time-series analysis to enhance the robustness of financial market forecasting models.