Stock Price Prediction Using Moving Average Crossover with LSTM and Hybrid LSTM Models

Abstract

Stock price fluctuations pose challenges for investors in making optimal decisions. This study evaluates the performance of LSTM with two input features and hybrid models combining LSTM with CNN, ANN, and RNN, enhanced by buy–sell–hold signals derived from the moving average crossover method. Hyperparameter tuning is applied to identify optimal configurations for stock index data across multiple countries. Results show that the hybrid and two-feature models outperform the standalone LSTM in 18 of the 20 stock indices and achieve comparable performance in one additional index. Among them, LSTM–CNN emerges as the most effective model, achieving the best performance in 10 of the 20 countries across both GDP-based country groups, with improvement levels reaching up to 76% and 88% in Groups 2 and 1, respectively. These findings indicate that integrating trading signals and hybrid architectures improves stock price prediction and supports investment decision-making.