TDRL$: Extracting and Evaluating the Fixed Causal Factors of Financial Markets
Mukhituly, Nurdaulet
Mukhituly, Nurdaulet
Author
Supervisor
Department
Machine Learning
Embargo End Date
2026-05-21
Type
Thesis
Date
2024
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Stock market analysis has traditionally been driven by the pursuit of profit, with substantial investments made into research seeking lucrative opportunities. Despite a predominant focus on financial gains, a shift towards a deeper understanding of market dynamics is also emerging. This paper aims to contribute to understanding Financial markets by exploring the fixed causal processes within financial market data. In this work, we adapted the existing TDRL framework (a non-linear ICA-based VAE model) to work with financial stock price data and improved its out-of-sample performance compared to the classical Stationary TDRL model. Furthermore, we assess the model's effectiveness in feature reconstruction, predictability, and profitability. This evaluation method will help us not only understand how well the latent factors reconstruct the data but also if these factors have predictive power or profitable potential. Lastly, this work will serve as a framework for future works in this domain.
Citation
N. Mukhituly, "TDRL$: Extracting and Evaluating the Fixed Causal Factors of Financial Markets" , MS. Thesis, Machine Learning, MBZUAI, Abu Dhabi, UAE, 2024