Interpretable fuzzy transformer-based network (IFTN) with applications in portfolio rebalancing

• Main Author: Chai, Fion Xin Yi
• Other Authors: Quek Hiok Chai
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Computer Science and Engineering; Machine learning
• Online Access: https://hdl.handle.net/10356/175260
• Institution: Nanyang Technological University

Financial forecasting techniques are used to project financial future trends using historical data. Traditional financial forecasting techniques, such as autoregressive integrated moving average (ARIMA) and generalized autoregressive conditional heteroskedasticity (GARCH) models, have been long used in the field. In the era of big data, machine learning models have gained popularity over traditional financial forecasting techniques for their ability to process data with larger scale and higher dimensions. Although deep learning models have high prediction accuracy, these models are essentially black boxes, which means their inner logic cannot be comprehended by humans directly. Having an interpretable model is crucial as users need to understand how a model makes predictions in order to make necessary adjustments and informed decisions. However, a majority of the existing studies on using deep learning models for financial forecasting have mainly focused on increasing the accuracy of predictions, neglecting the importance of making the models interpretable. To address this research gap, this project proposed the Interpretable Fuzzy Transformer-based Network (IFTN) model that combines the interpretability of a fuzzy inference system with the forecasting ability of a transformer-based network model. The proposed IFTN model can be utilised to perform portfolio rebalancing, with the objectives of penalising underperforming markets and capitalising on well-performing markets. A two-component portfolio reallocation strategy is proposed and the reallocation process is facilitated by reinforcement learning models.