Get Real: Realism Metrics for Robust Limit Order#
2. Affiliation:#
The first author’s affiliation is J.P. Morgan AI Research, New York, NY, USA.
3. Keywords:#
reinforcement learning, trading, simulation, limit order book, stylized facts
4. Url:#
5. Summary:#
(1): The research background of this paper is to address the challenge of ensuring that market simulation is realistic enough to develop and validate trading strategies in the context of reinforcement learning.
(2): Past methods for simulating markets were lacking in fidelity, and metrics for assessing the realism of the simulation were not well-defined. The proposed approach of this paper is well-motivated in that it aims to collect a set of reference metrics for assessing the realism of simulations based on stylized facts of limit order books in real markets.
(3): The research methodology of this paper consists of collecting a set of reference metrics for assessing the realism of market simulations based on stylized facts of limit order books. The metrics are applied to real market data and simulation output to identify discrepancies between them.
(4): The paper demonstrates that there are significant discrepancies between simulated markets and real ones, but presents a comprehensive catalog of reference metrics that can serve as a benchmark for measuring future improvement. The proposed approach can help to develop more realistic market simulations and improve the robustness of reinforcement learning-based trading strategies.
6. Conclusion:#
(1): The significance of this piece of work lies in addressing the challenge of developing and validating trading strategies through realistic market simulations in the context of reinforcement learning. The proposed approach provides a comprehensive catalog of reference metrics for assessing the realism of market simulations based on stylized facts of limit order books, which can help to develop more realistic simulations and improve the robustness of reinforcement learning-based trading strategies.
(2): Innovation point: The paper proposes a well-motivated approach of collecting reference metrics for assessing the realism of market simulations based on stylized facts of limit order books in real markets, which is a novel contribution to the field of reinforcement learning-based trading strategies. Performance: The paper demonstrates significant discrepancies between simulated markets and real ones, but presents a comprehensive catalog of reference metrics that can serve as a benchmark for measuring future improvement. Workload: The research methodology involves collecting a large set of reference metrics and applying them to both real market data and simulation output, which may require additional workload compared to simpler assessment methods. Overall, the paper provides valuable insights and sets a new direction for future research in this field.