DescriptionTrade execution is an ongoing optimisation problem in finance, focussing on attaining the best prices through managing price impact and forecasting future price movements and liquidity. This is especially relevant to institutional investors and to retail investors operating high-frequency, low-margin trading strategies.
This presentation introduces a framework developed to compare both the established mathematical models and the emerging deep reinforcement learning approaches to this problem. The mathematical models are purpose-built and efficient, but often assume that the underlying model dynamics are known, which is a point of contention. In contrast, the deep reinforcement learning techniques are more computationally intensive, but are very flexible, and do not require an assumption to be made regarding the underlying model dynamics.
Results from a suite of benchmarks and models implemented into the developed framework are provided, along with key areas for future research.
|Period||31 Aug 2023|
|Held at||Bangor University, United Kingdom|
|Degree of Recognition||International|