Paper: arXiv 2504.20349

Authors: Yichi Zhang, Mihai Cucuringu, Alexander Y. Shestopaloff, Stefan Zohren

Abstract

In the rapidly evolving world of financial markets, understanding the dynamics of limit order book (LOB) is crucial for unraveling market microstructure and participant behavior. We introduce ClusterLOB as a method to cluster individual market events in a stream of market-by-order (MBO) data into different groups. To do so, each market event is augmented with six time-dependent features. By applying the K-means++ clustering algorithm to the resulting order features, we are then able to assign each new order to one of three distinct clusters, which we identify as directional, opportunistic, and market-making participants, each capturing unique trading behaviors. Our experimental results are performed on one year of MBO data containing small-tick, medium-tick, and large-tick stocks from NASDAQ. To validate the usefulness of our clustering, we compute order flow imbalances across each cluster within 30-minute buckets during the trading day. We treat each cluster’s imbalance as a signal that provides insights into trading strategies and participants’ responses to varying market conditions. To assess the effectiveness of these signals, we identify the trading strategy with the highest Sharpe ratio in the training dataset, and demonstrate that its performance in the test dataset is superior to benchmark trading strategies that do not incorporate clustering. We also evaluate trading strategies based on order flow imbalance decompositions across different market event types, including add, cancel, and trade events, to assess their robustness in various market conditions. This work establishes a robust framework for clustering market participant behavior, which helps us to better understand market microstructure, and inform the development of more effective predictive trading signals with practical applications in algorithmic trading and quantitative finance.

Complexity vs Empirical Score

  • Math Complexity: 4.0/10
  • Empirical Rigor: 6.0/10 (capped from 8.5: transaction costs not modelled; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 8.5)

Why this score: The paper employs standard statistical methods (K-means clustering) and simple signals (order flow imbalance) without advanced mathematical derivations, resulting in moderate math complexity; however, it demonstrates high empirical rigor through the use of one year of MBO data from NASDAQ, explicit backtesting protocols, Sharpe ratio optimization, and the provision of source code for reproducibility.

Research Flowchart

  flowchart TD
  A["Research Goal"] --> B["Data Preparation"]
  B --> C["Feature Engineering"]
  C --> D["Clustering K-means++"]
  D --> E["Signal Generation"]
  E --> F["Strategy Evaluation"]
  F --> G["Key Findings"]

  subgraph A ["Research Goal"]
      A1["How can clustering orders in<br>Limit Order Books enhance<br>trading strategies?"]
  end

  subgraph B ["Data Inputs"]
      B1["1 Year NASDAQ MBO Data"]
      B2["Small/Medium/Large-Tick Stocks"]
  end

  subgraph C ["Methodology"]
      C1["Augment Orders with<br>6 Time-Dependent Features"]
  end

  subgraph D ["Computational Process"]
      D1["K-means++ Clustering<br>3 Clusters Identified:"]
      D2["Directional<br>Opportunistic<br>Market-Making"]
  end

  subgraph E ["Signal Generation"]
      E1["Order Flow Imbalance<br>per Cluster (30-min buckets)"]
  end

  subgraph F ["Validation"]
      F1["Training: Optimize Sharpe Ratio"]
      F2["Testing: Benchmark Comparison<br>(vs Non-Clustered Strategies)"]
  end

  subgraph G ["Key Outcomes"]
      G1["Superior Performance with Clustering"]
      G2["Robust Framework for<br>Market Microstructure Analysis"]
      G3["Practical Application in<br>Algorithmic Trading"]
  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 ClusterLOB, a K-means++ based method that clusters market‑by‑order events using six time‑dependent features into three trader types to generate order‑flow imbalance signals for trading.

Source sentence
We introduce ClusterLOB as a method to cluster individual market events in a stream of market-by-order (MBO) data into different groups. To do so, each market event is augmented with six time-dependent features. By applying the K-means++ clustering algorithm to the resulting order features, we are then able to assign e

Assumes 6

  • The tick size on Nasdaq is $0.01 and is used as the price resolution for all orders. market
    Source sentence
    In this paper, we focus on the Nasdaq exchange, where the tick size π is $0.01.
  • Three clusters (K=3) correspond to directional, opportunistic, and market‑making trader types. model
    Source sentence
    We set K = 3 as the number of clusters, which we retrospectively interpret as three trader types: directional, opportunistic, and market-making.
  • Six time‑dependent features (V, Tm, T1, T′, SBS, OBS) capture the essential order‑flow characteristics. data
    Source sentence
    In the training dataset for each tick size groups, we create six time-dependent features for each stock, denoted as V , T m, T 1, T ′, SBS, and OBS, as described in Section 3.2.1.1
  • Cluster labels are made consistent across stocks by initializing K‑means++ with centroids from a reference stock. model
    Source sentence
    For all remaining stocks, initialize the K‑means++ clustering using the centroids obtained from Step 2 rather than using random initialization.
Show 2 more
  • Order‑flow imbalances are measured in 30‑minute buckets and used as trading signals. execution
    Source sentence
    order flow imbalances across different clusters within 30‑minute buckets, we extract alpha signals that provide actionable
  • Equal‑weighted PnL is rescaled to a target annualized volatility of 0.15 for risk‑adjusted comparison. statistical
    Source sentence
    Then, we rescale the equal‑weighted PnL by their volatility to target equal risk assignment, and set our annualized volatility target σtgt to be 0.15.

Data

Backtest sample: NASDAQ small‑tick stocks · equities · intraday

  • MBO data 2021 · NASDAQ order-book public
    2021 · 15 NASDAQ stocks (CHTR, GOOG, GS, IBM, MCD, NVDA, AAPL, ABBV, PM, CMCSA, CSCO, INTC, MSFT,
    Source sentence
    The small‑tick group comprises CHTR, GOOG, GS, IBM, MCD, and NVDA

Recognised sources: LOBSTER Nasdaq TotalView-ITCH

Costs

Not modeled A trading or execution result is reported without costs.

Finds 3

Reported performance (Opportunistic cluster (ϕ2) trading strategy): Sharpe 1.34 · max drawdown 7.1%

  • The best clustering‑driven trading strategy outperforms benchmark strategies that do not use clustering, achieving higher Sharpe ratios. performance
    Source sentence
    The results show that the best clustering‑driven trading strategy outperforms benchmark trading strategies, leading to improved trading performance and highlighting its predictive power.
  • Signals derived from cluster‑specific order‑flow imbalances consistently beat benchmarks across all event types (add, cancel, trade) for small‑, medium‑ and large‑tick stocks. performance
    Source sentence
    Empirical results demonstrate that signals derived from clusters consistently outperform benchmark models without clustering in all event types, especially in terms of SR and other risk‑adjusted metrics.
  • Clustering reveals three distinct trader behaviors (directional, opportunistic, market‑making) that have interpretable correlations with price movements. other
    Source sentence
    We classify market activity into three clusters: directional, opportunistic, and market‑making traders. These clusters reflect distinct behavioral patterns and interactions with the market microstructure

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 26 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive (1 by independent authors): Neural Estimation of Irreversibility in Real and Simulated … (uses its method or data)

Code

https://github.com/YichiZhang-Oxford/ClusterLOB — the authors' own repository

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (24)

  1. The price impact of order book events (2014)
  2. Deep order flow imbalance: Extracting alpha at multiple horizons from the limit order book (2023)
  3. Limit order books (2013)
  4. Cross-impact of order flow imbalance in equity markets (2023)
  5. Improving Deep Learning of Alpha Term Structures from the Order Book (2024)
  6. Order Flow Decomposition for Price Impact Analysis in Equity Limit Order Books (2023)
  7. Lobster: Limit order book reconstruction system (2011)
  8. The price impact of order book events: market orders, limit orders and cancellations (2012)
and 16 more
  1. Deep Limit Order Book Forecasting — → 2403.09267
  2. HLOB–Information Persistence and Structure in Limit Order Books — → 2405.18938
  3. Enhancing time-series momentum strategies using deep neural networks (2019)
  4. Slow momentum with fast reversion: A trading strategy using deep learning and changepoint detection (2022)
  5. Trading with the Momentum Transformer: An Intelligent and Interpretable Architecture (2021) — link
  6. Slow momentum with fast reversion: A trading strategy using deep learning and changepoint detection (2021) — link
  7. Multi-Factor Inception: What to Do with All of These Features? — → 2307.13832
  8. Deep Inception Networks: A General End-to-End Framework for Multi-asset Quantitative Strategies — → 2307.05522
  9. Spatio-Temporal Momentum: Jointly Learning Time-Series and Cross-Sectional Strategies — → 2302.10175
  10. Transfer ranking in finance: applications to cross-sectional momentum with data scarcity (2022) — link
  11. Deeplob: Deep convolutional neural networks for limit order books (2019)
  12. Deep Learning for Market by Order Data (2021)
  13. Statistical predictions of trading strategies in electronic markets (2025)
  14. k-means++: The advantages of careful seeding (2006)
  15. Multi-level order-flow imbalance in a limit order book (2018)
  16. Analysis and modeling of client order flow in limit order markets (2023)

5 of 24 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.