JIITA, Vol.10 no.1 pp.1221-1239 (2026), DOI: 10.22664/ISITA.2026.10.1.1221
Sang Suh and Aarshna Vasaya
Abstract. Financial markets are becoming increasingly vulnerable to volatility caused by political events; however, conventional predictive models often neglect to account for real-time political instability, leading to considerable prediction inaccuracies during uncertain times. This research aims to fill this void by presenting a Dynamic Political Instability Index alongside an innovative Hybrid Attention Model that combines Bidirectional Long Short-Term Memory (BiLSTM) networks with Multi-Head Attention mechanisms. By integrating 40 years of data from the Dow Jones Industrial Average (DJIA) with the Daily Trade Policy Uncertainty (TPU) index and sentiment scores processed through VADER, the proposed framework effectively captures both temporal dependencies and discrete political shocks. Experimental findings indicate that the Hybrid Attention model significantly surpasses baseline LSTM, BiLSTM, and Transformer models. Importantly, the model addresses the “persistence bias” typically found in conventional recurrent networks, enabling the prediction of market trends instead of simply responding to past price levels.
Keywords; deep learning; stock market prediction; political instability; attention mechanism; LSTM; sentiment analysis; financial forecasting
Fullpaper: