The asymmetric impact of investor sentiment on stock returns: A sectoral deep learning analysis on BIST 30 index

Investor sentiment has emerged as a powerful factor in financial forecasting, particularly in the context of real-time digital information flows. This study explores the asymmetric effect of sentiment across five sectors in the BIST 30 Index: Energy, Aviation, Defense, Banking, and Steel. Using a hybrid deep learning architecture that integrates a Turkish BERT model for sentiment extraction and LSTM networks for price prediction, we evaluate the predictive performance of sentiment-augmented models versus traditional technical-only models. Results show significant sectoral variation: sentiment improves forecast accuracy in Energy (+14.31%), Aviation (+3.81%), and Defence (+1.58%) sectors, while deteriorating performance in Banking (-1.50%) and Steel (-10.03%). These findings challenge the universal applicability of sentiment analysis and highlight the need for context-aware, sector-specific financial modeling and suggest that sentiment acts as signal in event-driven sectors but as noise in macro-driven or commodity-dependent industries.