Integrating LSTM Forecasting with Mean-Variance Optimization for Dynamic Portfolio Management and Trading Signal Generation

This study integrates Long Short-Term Memory (LSTM) networks with mean-variance (MV) optimization to enhance dynamic portfolio management and generate trading signals. Traditional financial models, such as MV optimization, often fail to capture the non-linear dynamics of financial markets. LSTM networks, capable of learning longterm dependencies from historical price data, are applied to forecast future stock prices, offering superior predictive accuracy over conventional methods. To further optimize portfolio performance, reinforcement learning (RL) and genetic algorithms (GA) are employed for dynamic portfolio rebalancing. Additionally, technical analysis indicators such as RSI, MACD, and SMA are integrated with LSTM predictions to provide actionable buy and sell signals. The study demonstrates the potential of combining deep learning and optimization techniques for more adaptive and intelligent financial decision-making. Results indicate that this hybrid approach leads to improved risk-adjusted returns and more informed trading strategies, thus presenting a promising method for navigating the complexities of real-world financial markets.