Paper: arXiv 2306.12806
Authors: Andrea Coletta, Joseph Jerome, Rahul Savani, Svitlana Vyetrenko
Abstract
Limit order books are a fundamental and widespread market mechanism. This paper investigates the use of conditional generative models for order book simulation. For developing a trading agent, this approach has drawn recent attention as an alternative to traditional backtesting due to its ability to react to the presence of the trading agent. Using a state-of-the-art CGAN (from Coletta et al. (2022)), we explore its dependence upon input features, which highlights both strengths and weaknesses. To do this, we use “adversarial attacks” on the model’s features and its mechanism. We then show how these insights can be used to improve the CGAN, both in terms of its realism and robustness. We finish by laying out a roadmap for future work.
Complexity vs Empirical Score
- Math Complexity: 6.5/10
- Empirical Rigor: 6.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper involves advanced generative models (GANs), feature dependence analysis, and adversarial training, reflecting high mathematical complexity. It uses historical order data, focuses on model robustness and realism, and discusses implementation for trading agents, indicating strong empirical rigor.
Research Flowchart
flowchart TD
A["Research Goal"] --> B["Model: Conditional GAN<br>(CGAN) for LOB Simulation"]
B --> C["Adversarial Attacks<br>on Features & Mechanism"]
C --> D["Analysis of Strengths<br>& Weaknesses"]
D --> E{"Key Outcomes"}
E --> F["Improved CGAN Realism<br>& Robustness"]
E --> G["Roadmap for<br>Future Work"]
subgraph "Data & Input"
H["Limit Order Book Data"]
H --> B
end
style A fill:#e1f5fe
style E fill:#fff3e0
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes new LOBGAN models that improve both realism and robustness based on analysis of feature dependence and adversarial strategies.
Source sentence
Using the insights from our analysis and adversarial strategies, we designed new LOBGAN models that are better in terms of both realism and robustness.
Assumes 3
- Market replay simulators assume the financial market does not react to the presence of the trading agent being tested. market
Source sentence
Thus, one implicitly makes the assumption that the financial market does not react to the presence of the trading agent which is being tested.
- The adversarial attacks presented are not considered market manipulation. other
Source sentence
2 Remark: our adversarial attacks are not “market manipulation”.
- Altering input features may produce unrealistic order‑book and trade combinations, but the CGAN output remains relatively well‑behaved. model
Source sentence
We note that, in this approach the input to the CGAN after having been altered could correspond to an unrealistic orderbook and historical trade combination.
Data
Costs
Unclear
Finds 4
- LOBGAN generates realistic price‑impact paths but shows a greater impact of limit orders over market orders, which is unexpected. statistical
Source sentence
While we find that LOBGAN can generate realistic price impact paths, we also find that the paths show a greater impact of limit orders over market orders, which is unexpected, and was not studied in the earlier papers on LOBGAN.
- The third version of LOBGAN (v3) improves realism for volume and spread and shows enhanced robustness to both BookImbalance1 adversarial and market‑making strategies. performance
Source sentence
The figure clearly shows an improvement on real-ism for both volume and spread with respect to the previous attempt (the middle row in Figure 18). As shown in Figures 19 and 20, we can see that LOBGAN v3 improves upon all of the previous versions of LOBGAN in terms of robustness to both the BookImbalance1 adversarial strategy and the market-making strategy.
- Adversarial attacks that manipulate features and the model’s mechanism expose weaknesses in LOBGAN’s feature conditioning. negative
Source sentence
As such, we design adversarial attacks to exploit and show weaknesses of such models by manipulating the features and mechanism of the models.
- Interactive realism—realistic reactivity to external agents—was explored in depth, demonstrating LOBGAN’s behavior when interacting with various trading agents. theoretical
Source sentence
In Sec-tion 6, we explore interactive realism in depth – for the first time. We analyse LOBGAN’s outputs when it interacts with a range of trading agents.
Checked by others
Contradicted by
- Many learning agents interacting with an agent-based market model
Compare the findings
This paper: LOBGAN generates realistic price‑impact paths but shows a greater impact of limit orders over market orders, which is unexpected.
That paper: Agents that can use limit orders achieve lower price impact and faster decay of absolute returns.
23 papers in the archive : Painting the market: generative diffusion models for … ; Limit Order Book Simulation and Trade Evaluation with … ; Limit Order Book Simulations: A Review ; Controllable Financial Market Generation with Diffusion … ; Financial Wind Tunnel: A Retrieval-Augmented Market …
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.