Deep learning approach for stock price prediction: Comparing single and hybrid models

Articles in Press

Authors

1 Assistant Professor at Department of Management and Accounting, University of Tehran (College of Farabi), Qom, Iran

2 MSc., Department of Financial Management, Faculty of Management and Accounting, Farabi Campus of University of Tehran, Qom, Iran

10.22059/aie.2024.372470.1889

Abstract

In this research, deep-learning models have been employed for stock price prediction. For this research, we utilize data from five companies listed on the Tehran Stock Exchange. The time span considered is from the year 2001 to 2022. Five deep-learning models have been employed here, consisting of two hybrid models and three single models. The hybrid CNN-LSTM model is introduced as the main model, and its performance in terms of prediction accuracy is compared with four other models. The findings indicate that the combined CNN-LSTM model has demonstrated favorable performance compared to other models. However, the performance of another hybrid model, namely the CNN-GRU model, has also been satisfactory. Regarding the performance of single models, despite initial predictions, the results show that the CNN model has outperformed both LSTM and GRU models. The presence of factors such as volatility range significantly influences the accuracy of the models and tends to increase model uncertainty. However, in this research, which exclusively relies on technical variables, the results indicate that achieving desirable outcomes not only depends on selecting the type of neural network but also on the number of layers in each model. In general, according to the results of the research, in the four symbols used, the CNN-LSTM model has achieved the best performance and only in one of the symbols, the CNN-GRU model performs better than the CNN-LSTM model with a very slight difference. Among the single models, the CNN model has performed much better than the other two single models

Keywords

Main Subjects

Advances in Industrial Engineering

Articles in Press, Accepted Manuscript
Available Online from 26 June 2024

Files

Share

How to cite

Statistics