Original scientific paper
https://doi.org/10.15179/ces.28.1.1
Deep Learning in Financial Time Series: A Comparative Analysis of RNN, GRU, LSTM, and Hybrid Models
Yasin Büyükkör
; Karamanoğlu Mehmetbey University, Department of Business Administration, Karaman, Turkey
*
* Corresponding author.
Abstract
Accurate forecasting of financial time series plays a critical role in understanding market dynamics and informing investment decisions. However, the inherent volatility, nonlinearity, and noise of financial data make accurate prediction highly challenging. This study compares five deep learning architectures, including recurrent neural network (RNN), gated recurrent unit (GRU), long short-term memory (LSTM), and two hybrid models, RNN-LSTM and GRU-LSTM, to evaluate their forecasting performance on the S&P 500 Index between January 2022 and January 2025. Daily open, high, low, and volume values are used as input features, while the closing price serves as the target variable. The hyperparameters of all models are optimized using the random search procedure, and their predictive accuracy is assessed based on RMSE, MAE, and MAPE. The Diebold–Mariano test is further applied to examine whether differences in predictive accuracy among models are statistically significant. The empirical results indicate that hybrid architectures, particularly the GRU-LSTM model, outperform the standalone models by effectively capturing both short-term fluctuations and long-term dependencies in financial time series. The GRU-LSTM model achieves an approximately ten percent lower prediction error compared to the LSTM model, demonstrating the effectiveness of hybridization in improving deep learning forecasting performance. These findings confirm the robustness of hybrid deep learning architectures for financial time series forecasting and provide valuable insights for researchers and practitioners in quantitative finance.
Keywords
financial time series; recurrent neural network; gated recurrent unit; long short-term memory; hybrid deep learning models
Hrčak ID:
348453
URI
Publication date:
30.6.2026.
Visits: 0 *