Predictive Analytics for Financial Strategy: Multi-Timeframe Bi-LSTM on the VN-Index

Abstract

Within the digital economy, leveraging data to form financial strategies is vital, particularly in dynamic fintech landscapes such as Vietnam’s. This paper introduces an advanced predictive analytics model designed to improve forecasting accuracy for the VN-Index. Our model utilizes separate Bi-LSTM encoders to capture complex dependencies from daily, weekly, and monthly data, which are then aggregated through a hierarchical attention mechanism to provide a comprehensive market view. Experimental results on historical data from 2010 to 2023 show that our proposed model significantly outperforms a range of baselines, from classical statistical models to modern deep learning architectures. Our research demonstrates the value of advanced predictive analytics in shaping modern investment strategies, with the model achieving a peak directional accuracy (DA) of 67.38% and a balanced F1-Score for trend prediction.