Temporal Fusion Transformers for Enhanced Multivariate Time Series Forecasting of Indonesian Stock Prices

The stock market represents the financial pulse of economies and is an important part of the global financial system. It allows people to buy and sell shares in publicly held corporations. It serves as a platform for investors to trade ownership in businesses, enabling companies to raise capital for expansion and operations. However, the stock market can be very risky for any investor because of the fluctuating prices and uncertainties of the market. Integrating deep learning into stock market analysis enables researchers and practitioners to gain a deeper understanding of the trends and variations that will improve investment decisions. Recent advancements in the area of deep learning, more specifically with the invention of transformer-based models, have revolutionized research in stock market prediction. The Temporal Fusion Transformer (TFT) was introduced as a model that uses self-attention mechanisms to capture complex temporal dynamics across multiple time-series sequences. This study investigates feature engineering and technical data integrated into the TFT models to improve short-term stock market prediction. The Variance Inflation Factor (VIF) was used to quantify the severity of multicollinearity in the dataset. Evaluation metrics were used to evaluate TFT models’ effectiveness in improving the accuracy of stock market forecasting compared to other transformer models and traditional statistical Naïve models used as baselines. The results prove that TFT models excel in forecasting by effectively identifying multiple patterns, resulting in better predictive accuracy. Furthermore, considering the unique patterns of individual stocks, TFT obtained a remarkable SMAPE of 0.0022.

@article{Hartanto2024,
  title     = {Temporal Fusion Transformers for Enhanced Multivariate Time Series Forecasting of Indonesian Stock Prices},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {7},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Standy Hartanto and Alexander Agung Santoso Gunawan},
  doi       = {10.14569/IJACSA.2024.0150713},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150713}
}