Author ORCID Identifier
0000-0001-8010-8326
Document Type
Dissertation
Date of Award
8-31-2025
Degree Name
Doctor of Philosophy in Computing Sciences - (Ph.D.)
Department
Computer Science
First Advisor
Guiling Wang
Second Advisor
Zhi Wei
Third Advisor
Ioannis Koutis
Fourth Advisor
Tucker Balch
Fifth Advisor
Zhaodong Zhong
Abstract
Financial markets are inherently uncertain and dynamic, driven by complex factors such as macroeconomic signals, investor sentiment, and evolving inter-asset relationships. While machine learning has advanced financial modeling, existing approaches often fall short in addressing the real-world intricacies of finance. This dissertation confronts two critical challenges, human-driven stochasticity and risk-intensive decision-making under real-world trading constraints, while seizing a pivotal opportunity, the structural dynamics of evolving financial systems. These elements are foundational to advancing robust and practical financial intelligence.
To this end, this dissertation develops a unified framework for robust financial modeling and decision-making. The framework is architected as a progressive, modular pipeline that mirrors the real-world flow of financial intelligence, starting from macro-level index forecasting, advancing to individual asset pricing, and culmi-nating in actionable trading decisions under real-world constraints. (1) At the macro level, this framework captures stochasticity driven by human behaviors. Exploratory studies investigate the weekend effect of news on asset prices and generative modeling for market behavior, providing foundational insights into stochastic market dynamics. Building on this, RAGIC, a novel hybrid framework that fuses generative models with statistical inference, is established to construct risk-aware intervals that quantify forecast uncertainty and risk exposure under volatility regimes. (2) At the asset level, this framework leverages structural dynamics in financial markets and develops DySTAGE, a dynamic graph learning framework that models assets as nodes in a time-varying market graph, accommodating changing asset composition and evolving asset relationships. By incorporating spatio-temporal attention and graph encodings, DySTAGE enables structure-aware asset pricing and reveals hidden financial patterns. (3) Lastly, to support actionable trading under realistic constraints, the framework formulates Margin Trader, the first public reinforcement learning framework that incorporates realistic margin requirements and leverage rules to learn long-short trading strategies. It is further extended with large language models to integrate external signals and guide explainable and adaptive portfolio reallocation decisions.
Together, these innovations deliver a robust, scalable, and deployable financial AI framework with broad impact: it empowers individual investors with accessible, open-source tools for uncertainty-aware forecasting; supports academic researchers with a testbed for hybrid AI-finance exploration; and offers institutional stakeholders with data-driven, risk-sensitive systems for compliant and adaptive portfolio management in high-stakes, fast-evolving markets.
Recommended Citation
Gu, Jingyi, "Robust AI solutions for financial markets through generative modeling, dynamic graph learning, and reinforcement-based portfolio optimization" (2025). Dissertations. 1852.
https://digitalcommons.njit.edu/dissertations/1852
