Paper: arXiv 2409.06514

Authors: Michael Giegrich, Roel Oomen, Christoph Reisinger

Abstract

In this paper, we show how \(K\)-nearest neighbor (\(K\)-NN) resampling, an off-policy evaluation method proposed in \cite{“giegrich2023k”}, can be applied to simulate limit order book (LOB) markets and how it can be used to evaluate and calibrate trading strategies. Using historical LOB data, we demonstrate that our simulation method is capable of recreating realistic LOB dynamics and that synthetic trading within the simulation leads to a market impact in line with the corresponding literature. Compared to other statistical LOB simulation methods, our algorithm has theoretical convergence guarantees under general conditions, does not require optimization, is easy to implement and computationally efficient. Furthermore, we show that in a benchmark comparison our method outperforms a deep learning-based algorithm for several key statistics. In the context of a LOB with pro-rata type matching, we demonstrate how our algorithm can calibrate the size of limit orders for a liquidation strategy. Finally, we describe how \(K\)-NN resampling can be modified for choices of higher dimensional state spaces.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper utilizes advanced mathematical concepts from stochastic calculus and metric space theory for K-NN resampling with convergence proofs, while demonstrating rigorous empirical validation using historical LOB data, outperforming deep learning benchmarks and providing specific calibration examples for trading strategies.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Simulate LOB & Evaluate/Calibrate Trading<br>using K-NN Resampling"] --> B["Data & Inputs"]
  B --> C["Methodology: K-NN Resampling"]
  
  subgraph C ["Methodology: K-NN Resampling"]
      direction LR
      C1["Identify State Space"] --> C2["Find K-Nearest<br>Historical Neighbors"] --> C3["Resample Future<br>Transitions/Outcomes"]
  end

  C --> D["Computational Process"]
  
  subgraph D ["Computational Process"]
      direction LR
      D1["Market Simulation<br>Recreate LOB Dynamics"] --> D2["Strategy Evaluation<br>Simulate Synthetic Trading"]
  end

  D --> E["Key Findings & Outcomes"]
  
  subgraph E ["Key Findings & Outcomes"]
      E1["Validated Market Impact<br>Matches Literature"]
      E2["Outperformed Deep Learning<br>in Benchmark Stats"]
      E3["Calibrated Liquidation<br>Strategy Order Sizes"]
      E4["Extensible to<br>Higher Dimensional States"]
  end

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 a K‑NN resampling framework for simulating limit order books and evaluating trading strategies, including calibration of limit order sizes in a pro‑rata matching LOB.

Source sentence
We provide a framework how K-NN resampling can be adapted for simulating LOBs and how to incorporate trades within these simulations.

Assumes 6

  • The LOB evolves as a discrete random dynamical system with transitions defined by functions fZ and fp. model
    Source sentence
    In particular, we assume that the LOB evolves as a discrete random dynamical system transitioning from one LOB state-price pair (Zt, p∗t) to the next, (Zt+∆t, p∗t+∆t), where ∆t ∈R>0 is the time increment and Zt the state representation for the LOB.
  • The system noise ε is i.i.d., implying a Markovian LOB dynamics. model
    Source sentence
    We assume that the noise is identically and independently distributed (iid) for all s ∈N
  • Price level does not impact the shape of the subsequent order book. model
    Source sentence
    We choose to model the price explicitly in order to facilitate the assignment of empty tick sizes in the LOB snapshots and we assume that the price level does not have an impact on the shape of the subsequent order book.
  • The LOB state is given by centered LOB snapshots for simplicity. model
    Source sentence
    For simplicity, we assume that the LOB state is given by centered LOB snapshots.
Show 2 more
  • Nearest‑neighbor parameter K is set heuristically to 20. model
    Source sentence
    In our experiments, we choose the nearest neighbor parameter K = 20 heuristically, which is common practice in machine learning.
  • Euclidean distance is used as the metric for nearest‑neighbor search. model
    Source sentence
    As metric for the nearest neighbor search, we use Euclidean distances to keep the implementation as simple as possible.

Data

  • historical LOB data order-book unknown
    Source sentence
    Using historical LOB data, we demonstrate that our simulation method is capable of recreating realistic LOB dynamics

Costs

Modeled Costs are part of the reported headline results.

Execution costs are measured via average cash earnings per asset, showing negative earnings (cost) around half a tick size, varying with order multiplier.

Source sentence
First, note that for all order multipliers the relative average cash earnings are negative, implying a cost of execution on average. The scale of the execution costs varies around half a tick size.

Finds 3

  • Algorithm KNNR consistently outperforms deep‑learning and naive benchmarks on KS statistics for most LOB variables, especially return‑related features. performance
    Source sentence
    Table 1 shows that Algorithm KNNR consistently outperforms our benchmarks for most variables. In particular, we highlight that all features related to returns outperform both the CGAN and the naive benchmark.
  • Synthetic trading in the simulation produces market impact consistent with literature, with execution costs around half a tick and optimal order multiplier k=1.25. performance
    Source sentence
    Highest earnings (or lowest execution costs) are reached at k = 1.25, while they decrease consistently for smaller and larger values in k.
  • The K‑NN resampling method can realistically simulate LOB dynamics both with and without trade interventions. statistical
    Source sentence
    We provide empirical evidence that our method is capable of realistically simulating LOB dynamics with and without trade interventions.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 18 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 1 paper in the archive (1 by independent authors)

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.

Cited References (49)

  1. Limit Order Book Simulation with Generative Adversarial Networks (2023)
  2. Inferring term rates from SOFR futures prices (2019)
  3. Three-Month SOFR (2024)
  4. Generative AI for End-to-End Limit Order Book Modelling: A Token-Level Autoregressive Generative Model of Message Flow Using a Deep State Space Network — → 2309.00638
  5. Combining parametric and nonparametric models for off-policy evaluation (2019)
  6. Handbook of Price Impact Modeling (2023)
  7. No-dynamic-arbitrage and market impact (2010)
  8. Market impacts and the life cycle of investors orders (2015)
and 41 more
  1. Market impact: A systematic study of limit orders (2017)
  2. $ K $-Nearest-Neighbor Resampling for Off-Policy Evaluation in Stochastic Control (2023) — link
  3. Universal features of price formation in financial markets: perspectives from deep learning (2019)
  4. Market orders and market efficiency (1997)
  5. Limit Order Book Simulations: A Review — → 2402.17359
  6. Recent advances in reinforcement learning in finance (2023)
  7. A stochastic model for order book dynamics (2010)
  8. A mathematical approach to order book modeling (2013)
  9. Price dynamics in a Markovian limit order market (2013)
  10. A law of large numbers for limit order books (2017)
  11. The role of volume in order book dynamics: a multivariate Hawkes process analysis (2017)
  12. Conditional Generators for Limit Order Book Environments: Explainability, Challenges, and Robustness — → 2306.12806
  13. A generative model of a limit order book using recurrent neural networks (2023)
  14. Enhancing trading strategies with order book signals (2018)
  15. Pro-rata matching and one-tick markets (2008)
  16. Optimal high-frequency trading in a pro rata microstructure with predictive information (2015)
  17. A distribution-free theory of nonparametric regression (2002)
  18. Statistical properties of stock order books: empirical results and models (2002)
  19. Hawkes processes in finance (2015)
  20. Order-book modeling and market making strategies (2018)
  21. State-dependent Hawkes processes and their application to limit order book modelling (2022)
  22. Simulating and analyzing order book data: The queue-reactive model (2015)
  23. An agent based model of the E-Mini S&P 500 applied to Flash Crash analysis (2012)
  24. Abides: Towards high-fidelity market simulation for AI research (2019) — link
  25. Fast agent-based simulation framework of limit order books with applications to pro-rata markets and the study of latency effects (2020) — link
  26. Neural Stochastic Agent-Based Limit Order Book Simulation: A Hybrid Methodology — → 2303.00080
  27. Modeling high-frequency order flow imbalance by functional limit theorems for two-sided risk processes (2015)
  28. Ergodicity and diffusivity of Markovian order book models: a general framework (2017)
  29. A stochastic partial differential equation model for limit order book dynamics (2021)
  30. Intra-Day Price Simulation with Generative Adversarial Modelling of the Order Flow (2021)
  31. Generating realistic stock market order streams (2020)
  32. The Limit Order Book Recreation Model (LOBRM): An Extended Analysis (2021)
  33. Model-based gym environments for limit order book trading (2022) — link
  34. The unbalanced G romov W asserstein distance: Conic formulation and relaxation (2021)
  35. Anomalous price impact and the critical nature of liquidity in financial markets (2011)
  36. Active portfolio management (2000)
  37. Market impact and trading profile of hidden orders in stock markets (2009)
  38. The Elements of Statistical Learning (2009)
  39. The market impact of a limit order (2012)
  40. The price impact of order book events: market orders, limit orders and cancellations (2012)
  41. Neural networks for option pricing and hedging: a literature review (2019) — link

4 of 49 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.