Reviews on Transformer-based Models for Financial Time Series Forecasting
Heyi Lin · Applied and Computational Engineering · 2024
The emergence of competitive deep learning models has increasing attached attention from both the academia and industry. Thus, as one of the fields that tend to chase the state-of-art and fashion technological trend, some previous work in financial time series forecasting has turned to deep learning models, including transformer-based models. While an examination work questioning the effectiveness of transformers for general time series forecasting (TSF) in 2022, researchers are keen to work on the creative design of transformer-based neural network architectures and related improvements. On the other hand, since the success of ChatGPT in 2023 as the milestone of transformers and Large Language Models (LLMs), an alternative method is put forward that implements domain-specific LLM in financial text to obtain sentiment information or generate trading signals, which does not solve the forecasting problem but provide support in decision making in investment. This review will scan through the history of the above models and methodologies in financial time series forecasting.