Dual Model Framework for Stock Price Movement Prediction Using Technical Indicators and Financial News Sentiment with Decision Level Fusion
Abstract
This In this era, due to the increasing use of fintech, the paper proposes a dual model framework for stock price movement prediction using two complementary branches: a technical branch based on sliding window price indicators and a news branch based on ticker specific financial sentiment. The news branch uses an LLM assisted extraction stage in which the full title and article text are provided to ChatGPT to isolate ticker relevant content, followed by FinBERT sentiment scoring and daily aggregation into 775 ticker date records. Final inference uses decision level fusion: when both branches agree, the shared prediction is accepted; when they disagree, each branch receives a reliability score equal to its class specific F1 multiplied by its prediction confidence. Experiments on Amazon, Apple, and Google show the viability and limitations of combining modality specific models through an interpretable fusion rule, the results demonstrate both the viability and limitations of this approach for stock price movement prediction.