A Hybrid BI-LSTM Framework for Portfolio Optimization via Fundamental Analysis: Insights from NIFTY 50

Abstract

Conventional models of portfolio optimization, which rely on technical or fundamental analysis, cannot adjust to changing market conditions. On the contrary, general technical analysis models rely on past price trends and neglect short-term market trends. This study leverages the advantages of conventional models and state-of-the-art machine learning models to address this limitation. The study proposes a hybrid model that combines important financial metrics like dividend yield, P/E ratio, and Piotroski score with BI-LSTM-based stock return forecasting. The model provided a Sharpe Ratio of 8.86, its expected return of 30.44%, and its portfolio risk of 2.8%, all based on a 10-year Nifty 50 data set. By maximizing risk-adjusted returns while preserving diversification, it performs better than conventional and machine learning models. When optimizing the investment strategy, the method provides a sound way to balance long-term financial stability with short-term responsiveness. Our hybrid model shows a p-value < 0.01 improvement in Sharpe Ratio when compared to the Markowitz mean–variance benchmark when statistical significance testing is included. We also perform out-of-sample walk-forward analysis to verify robustness in volatile, bull, and bear market regimes