Stock movement prediction using mamba and ensemble learning

The stock market is influenced by various factors such as national policies, economic conditions, and global events. Accurately predicting stock price movements has long posed a significant challenge for investors and economists. Accurate predictions play a critical role in reducing investment risks while also enhancing return potential. However, the volatile and nonlinear structures of financial markets complicate the prediction processes. In recent years, state space models like Mamba have achieved significant success in modeling sequential data. This study utilizes the Mamba model to predict percentage changes in daily stock closing prices. The prediction problem is approached as a classification task, aiming to determine whether the stock price will increase or decrease the next day. In the validation dataset, hyper parameter optimization is performed using grid search to enhance the model's performance. Subsequently, to improve the success of the predictions, models with different combinations of hyperparameters are integrated using ensemble learning methods. Thus, the prediction accuracy is increased compared to the single Mamba model. The study, forecasts daily closing price movements for six different companies listed on the Nasdaq stock exchange. The results indicate that the single Mamba model achieves an average F1 score of 58.9%. In contrast, the ensemble Mamba model demonstrates a notable performance with an average F1 score of 60.5%. When comparing the performance of our models with a baseline model, the LSTM model is found to have an average F1 score of 52.5%. Moreover, the financial assessment alongside the classification performance of the models is conducted. According to the findings, the ensemble Mamba model generates an average profit of $4,150 with an initial capital of $10,000 over 120 test trading days. In comparison, our basic strategy, "Buy \& Hold," incurs a loss of $809. These findings provide significant guidance for investors to make more informed decisions. In this way, optimizing returns and minimizing risks can be achieved. In conclusion, this study presents an effective method to enhance prediction accuracy in the stock market and contributes to the optimization of investment decisions.