Comparative Analysis of Deep Learning and Statistical Models for Sectoral Index Forecasting Using Financial Sentiment Scores

Financial forecasting models have traditionally prioritized broad market indices and often overlooked the unique dynamics of sector-specific trends. This study addresses this gap by proposing a forecasting framework that integrates sentiment analysis with deep learning models to predict the daily closing prices of four major NIFTY sectoral indices: AUTO, BANK, IT, and OIL & GAS. Sector-specific news headlines were collected from six credible financial news outlets and analyzed using FinBERT, a transformer-based model fine-tuned for financial sentiment analysis. The resulting sentiment scores were combined with historical stock price data to form a comprehensive input for model training. Four predictive models: ARIMA, LSTM, BiGRU, and BiLSTM were implemented and compared using performance metrics such as MAE, MAPE, and accuracy. The experimental results demonstrated that BiLSTM consistently outperformed other models in all sectors, particularly when enriched with sentiment features. The findings underscore the value of incorporating domain-specific sentiment data into predictive modeling, offering a robust and interpretable approach to sectoral index forecasting.