Microstructural Financial Modelling: Point Processes and Reinforcement Learning
A thesis advances high-frequency financial market modeling by creating realistic Limit Order Book (LOB) simulators using Hawkes processes, notably the Meta-Queue Hawkes model. This approach captures complex trading patterns and adapts to various asset types. Integrating impulse control theory with reinforcement learning, the research develops trading agents that outperform traditional methods, achieving high Sharpe ratios. Using tick-by-tick data from 15 US stocks, the study highlights calibration challenges and the need for more explainability and real-world constraints.
What it examines
This thesis develops realistic models of financial markets at the micro-level, using point processes and reinforcement learning to simulate and optimize trading in electronic limit order books. It aims to create data-driven, adaptive trading agents that can make smart decisions in highly unpredictable and complex market environments.
What it concludes
The research shows that combining advanced mathematical models with reinforcement learning leads to robust, adaptive trading strategies. These methods can be used for safer, more efficient algorithmic trading, risk management, and market simulation. Future work includes handling more complex market scenarios and improving model calibration for real-world applications.
Evidence objects
A breakthrough thesis develops realistic, data-driven Limit Order Book models using advanced point processes, notably the Hawkes process, enabling high-fidelity simulation across diverse market regimes and matching real-world 'stylized facts.'.
key_findings bullet 1 · key_findings · validation V0
The novel Meta-Queue Hawkes (MQH) model captures complex microstructural features like sparsity and multi-tick price moves efficiently, while integrating impulse control theory with reinforcement learning for robust, risk-aware trading agent design.
key_findings bullet 2 · key_findings · validation V0
Rigorous analysis on tick-by-tick data from 15 major US stocks shows the RL agent achieves high Sharpe ratios and adapts to market feedback, outperforming traditional methods, despite calibration and explainability challenges.
key_findings bullet 3 · key_findings · validation V0
This thesis introduces original frameworks for modeling Limit Order Books, notably applying compound Hawkes processes to order size, generalizing methods for small tick LOBs, and proposing impulse control for market making. Its novelty lies in integrating advanced point processes and reinforcement learning, offering compelling, impactful insights at the forefront of Market Microstructure research.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
- … of high-frequency market microstructure modelling, … decision making in electronic trading. Specifically, it … for algorithmic trading tasks in a pure jump-driven market, and …
Source row: 1335 · abstract type: snippet