Stock Price Trend Prediction and Securities Trading Optimization Based on CNN-GAA

Abstract

This study proposes an optimization model (SSCGL) that combines the stock price trend prediction algorithm (CNN-GAA-LSTM) with the securities trading algorithm (SE Net SMCM) aimed at improving the accuracy of stock price trend prediction. This model combines convolutional neural networks with gated attention aggregators to enhance feature representation, introduces long short-term memory networks for trend prediction, and uses a combination of compression and excitation networks and sequential Monte Carlo methods to optimize trading decisions. The innovation of this model lies in the deep integration of sequential Monte Carlo algorithm with Squeeze-and-Excitation Network to dynamically evaluate market trading signals. The experimental results show that the model exhibits strong predictive performance, with a trend prediction accuracy of 95.5%, an accuracy of 0.975, and a root mean square error of only 3.93%. In trading simulations, the model performed outstandingly with a Sharpe ratio of 1.812 and a Kalmar ratio of 3.871. Empirical evaluations showed that the annualized return rate was as high as 46.11%. These data fully demonstrate that the model can 
not only accurately capture market trends, but also improve trading efficiency by effectively controlling risks and reducing losses. Overall, this plan provides stable returns and reliable technical support for investment activities, helping financial institutions achieve deep integration with intelligent algorithm systems.