Attention-driven financial management for dynamic portfolio optimization and asset allocation

Dynamic portfolio management for the current financial markets faces significant challenges. Nonlinear intricacies and interactions among assets have made traditional approaches irrelevant for optimal allocation. The current research introduces a novel deep learning-based approach to active management and optimization of portfolio allocation. At the heart of this framework is a hierarchical and multi-faceted attention architecture that predicts key variables and risks for each asset. The model learns the unique temporal patterns of each data set by using separate networks for each of the information aspects including market, fundamental, and macroeconomic data. Then, a mutual attention mechanism between assets models the hidden effects and dependencies in the overall market. To validate the model, statistical data of 50 companies active in the stock market over a 15-year period is used. Finally, the predicted outputs, along with the covariance matrix estimated using the Ledoit-Wolf method, are fed to a “black hole” optimization algorithm to determine the optimal weight of each asset in the portfolio by maximizing the Sharpe ratio. The proposed method developed in this study has been found to have great performance in the prediction of monthly profit and investment risk of stocks. The new approach achieved a commendable value of 1.46 and 1.12 for root mean square error (RMSE) in predicting the profit and risk variables, indicating the high precision of the model in predicting actual values. Furthermore, in the context of the measurement of the mean absolute percentage error (MAPE), this method achieved an average of 8.09% in monthly profit prediction and an average of 30.16% for the investment risk variable. Generally, these results accurately demonstrate the efficiency and predictive power of the proposed model in managing the volatility of financial data and indicate its ability to produce valid and effective estimates for making investment decisions.