Mary Chevel Modesto, Victorienne Tiu, Glenn Pepito
Emerging markets like the Philippine Stock Exchange (PSE) are highly volatile and prone to sudden price swings that traditional sentiment-based financial models often fail to capture, creating heightened risks for investors and the broader economy. Although existing studies highlight the strong influence of news sentiment on investor behavior and market activity, the temporal dimension, specifically how long it takes for news to affect price movements, remains largely underexplored in emerging markets and economies, particularly within the Philippine context. This study examines the temporal impact of news sentiment on stock index prediction in an empirical manner by evaluating the predictive power of lagged local news sentiment on the Philippine Stock Exchange Index (PSEi) and assessing how different lag durations (in days) influence the performance of a Bidirectional Long Short-Term Memory (Bi-LSTM) prediction model. To focus on the effect of lagged sentiment under normal market conditions, historical post-COVID PSEi data covering four full years were collected alongside more than 120,000 news articles from the finance-related sections of multiple major Philippine news outlets. Sentiment scores were generated using FinBERT, and wavelet coherence analysis was applied to decompose both sentiment and PSEi time series into frequency components and empirically identify optimal lag values through cross-correlation. The Bi-LSTM is fed and evaluated using RMSE, MDA, R2, MAE, and MAPE to measure accuracy and directional correctness across time horizons. Results show that the most effective lag value for prediction is a 1-day lag of local news sentiment. Additionally, Corporate, Banking, and Economy news demonstrate the strongest predictive power, while Entrepreneurship and IT-Telecom sectors are less influential. When combining news sentiments, the model performs best with no lag added, and predictive accuracy improves as the lag length decreases. © 2025 IEEE.
University of San Carlos Nasipit, Department of Computer, Information Sciences, and Mathematics, Talamban, Cebu, 6000, Philippines
