The Role of Convolutional Neural Networks (CNNs) in Forecasting Stock Prices of Green Companies: An Analysis of Time-Step Configurations

The growing emphasis on sustainability has increased the role of green companies in the global economy, underscoring the need for accurate stock price forecasting to support informed investment decisions. This study examines the predictive capabilities of Convolutional Neural Networks (CNNs) for forecasting stock prices for green companies, focusing on the impact of various time-step configurations. By exploring different historical data windows, this research evaluates the CNN model’s performance through key metrics: Mean Squared Error (MSE), Mean Absolute Error (MAE), Rsquared ($\mathrm{R}^{2}$), accuracy, precision, recall, and F1 Score. The results indicate that shorter time steps, particularly steps 5 and 15, achieve superior predictive accuracy and effectively capture short-term trends. The model demonstrated the highest performance with a timestep of 5, achieving the lowest MSE (0.001013) and MAE (0.023909), along with an $\mathrm{R^{2}}$ of 0.977890 and an accuracy of $\mathrm{9 1. 1 4 \%}$. Cross-validation further confirmed the robustness and generalizability of the model, with an average accuracy of $73.11 \%$ and balanced precision, recall, and F1 Score values. These findings validate CNNs as reliable tools for stock price prediction in a green economy and emphasize the importance of optimal time-step selection. This study offers valuable insights into enhancing predictive accuracy through the potential integration of additional market indicators and sentiment analysis, contributing to the evolving field of financial forecasting for sustainable investments.