Anticipative portfolio optimisation using an interpretable evolving fuzzy transformer network

The accurate prediction of stock market prices remains one of the most challenging and studied problems within the field of financial engineering. However, the inherently uncertain and fuzzy nature of financial markets has often limited the effectiveness of conventional crisp computational models. F...

Full description

Saved in:
Bibliographic Details
Main Author: Ong, Bryan Shao En
Other Authors: Quek Hiok Chai
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2024
Subjects:
Online Access:https://hdl.handle.net/10356/175135
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:The accurate prediction of stock market prices remains one of the most challenging and studied problems within the field of financial engineering. However, the inherently uncertain and fuzzy nature of financial markets has often limited the effectiveness of conventional crisp computational models. Furthermore, as financial institutions deploy increasingly complex machine learning models, the opacity of such 'black box' systems raises significant issues. The imperative for explainable AI becomes all the more pressing in an industry that requires trust, transparency, and compliance with regulatory standards. Another challenge is optimal portfolio allocation. Classically, indicators driven by historical data is used which limits its applicability. This paper proposes a novel Evolving Fuzzy Transformer Network that utilises a fuzzy system to model the underlying uncertainty in a market. Through the creation of this neuro-fuzzy system, a quantitative explanation can be generated, allowing for interpretation of the system. Finally, we use the predictive model to generate forecast indicators to dynamically balance a portfolio with a deep reinforcement learning model.