A Quantitative Portfolio Management Method with Machine Learning Optimization Algorithm

In financial trading, an effective trading strategy is key to determining profit and loss. Due to the complexity and dynamics of financial markets, the automated selection of trading strategies has become the focus of modern financial research. This study utilizes deep learning and reinforcement learning methods, proposing an end-to-end deep reinforcement learning trading strategy algorithm that combines CNN and LSTM, named CLDQN. Within this framework, the CNN module is utilized to perceive dynamic market conditions of stocks and extract crucial features, while the LSTM module is responsible for learning long-term dependencies in the time series. After processing through the reinforcement learning method DQN, the algorithm makes trading decisions. To verify the effectiveness of CLDQN, we compared it with benchmark methods such as LSTM, SVM, and decision trees. Experimental results show that CLDQN's three-year cumulative return rate on four stocks is on average 1.1875%, 1.925%, and 2.3875% higher than that of LSTM, SVM, and decision trees respectively. These results not only demonstrate the superiority of the CLDQN method but also highlight its excellent scalability and robustness.