Financial Market Momentum Prediction and Portfolio Optimization Based on LSTM-Attention Hybrid Model and Dynamic Time Warping (DTW)

Accurate financial market momentum prediction is critical for building profitable, risk-managed portfolios, especially in volatile markets where traditional models often fall short in capturing nonlinear patterns. This study proposes a hybrid deep learning model that integrates Long Short-Term Memory (LSTM) networks with attention mechanisms and Dynamic Time Warping (DTW) for enhanced momentum forecasting and portfolio optimization. The model processes multi-asset data spanning 2013–2023 across stocks, commodities, currencies, bonds, and real estate. The LSTM-Attention component captures temporal dependencies in price dynamics, while DTW clusters assets based on behavioral similarity to support robust diversification. Experimental results demonstrate that the hybrid model outperformed baseline models such as ARIMA and LSTM without attention, achieving the lowest momentum prediction error (MAPE of 5.6% for stocks), the highest Sharpe ratio (0.67), and the lowest maximum drawdown (12.4%) with improved risk-adjusted returns and downside protection. These results highlight the model’s suitability for adaptive and data-driven portfolio management in real-world financial markets.