Year: 2025
Type: article
Abstract: In modern enterprises, financial forecasting is critical in determining investment strategies, risk management, and the basis for decision-making. Both ARIMA and GARCH models are plagued by the inability to tackle the nonlinearity and volatility of markets. An LSTM, transformers, and hybrid CNN-LSTM-based deep learning framework is proposed to increase this study's predictive performance. The models are developed with historical stock and macroeconomic data and are evaluated using MAE, RMSE, and directional accuracy. The hybrid CNN-LSTM model achieved a 34% lower RMSE than ARIMA and a 74.3% directional accuracy. Results prove that deep learning surpasses the traditional methods, with the CNN-LSTM model proving to be more accurate and robust than others. However, being interpretable and computationally intensive are still important issues for enterprise adoption.
Authors: Zhaoming Dong, Li Xu
Institution: Shanghai Communications Polytechnic
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W4411308889