Paper: arXiv 2306.05479

Authors: Alvaro Arroyo, Alvaro Cartea, Fernando Moreno-Pino, Stefan Zohren

Abstract

One of the key decisions in execution strategies is the choice between a passive (liquidity providing) or an aggressive (liquidity taking) order to execute a trade in a limit order book (LOB). Essential to this choice is the fill probability of a passive limit order placed in the LOB. This paper proposes a deep learning method to estimate the filltimes of limit orders posted in different levels of the LOB. We develop a novel model for survival analysis that maps time-varying features of the LOB to the distribution of filltimes of limit orders. Our method is based on a convolutional-Transformer encoder and a monotonic neural network decoder. We use proper scoring rules to compare our method with other approaches in survival analysis, and perform an interpretability analysis to understand the informativeness of features used to compute fill probabilities. Our method significantly outperforms those typically used in survival analysis literature. Finally, we carry out a statistical analysis of the fill probability of orders placed in the order book (e.g., within the bid-ask spread) for assets with different queue dynamics and trading activity.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematical concepts including survival analysis, Transformer architectures, convolutional networks, and proper scoring rules, indicating high complexity, while it is grounded in real-world limit order book data from Nasdaq, uses statistical analysis, and compares performance with benchmarks, demonstrating high empirical rigor.

Research Flowchart

  flowchart TD
  Start["Research Goal<br>Estimate Fill Probabilities of Limit Orders in LOB"] --> Data["Input: Time-Varying LOB Features<br>(Price/Volume Levels, Order Flow)"]
  Data --> Methodology["Methodology:<br>Convolutional-Transformer Encoder"]
  Methodology --> Processing["Computational Process:<br>Map Features to Filltime Distribution"]
  Processing --> Model["Model Structure:<br>Monotonic Neural Network Decoder"]
  Model --> Results["Outcomes:<br>Superior Fill Probability Estimates<br>(vs. Traditional Survival Analysis)"]
  Results --> Analysis["Final Analysis:<br>Interpretability &<br>Statistical Fill Probabilities by Asset"]
  Analysis --> End["Key Insight<br>Informed Passive vs. Aggressive Order Decisions"]

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 deep learning model that uses a convolutional‑Transformer encoder and a monotonic neural network decoder to estimate fill‑time survival functions for limit orders.

Source sentence
Our method is based on a convolutional‑Transformer encoder and a monotonic neural network decoder.

Assumes 5

  • The fill time T and cancellation time C are independent. statistical
    Source sentence
    (assuming T is independent of C)
  • Only the best levels of the order book are tracked, ignoring deeper levels. data
    Source sentence
    For simplicity, we also focus on the best levels with tracked orders, as they also contain a large amount of censoring which results in higher computational cost to obtain data to train the model.
  • A market order executes against any of the orders being tracked, and then the hypothetical order is considered filled. execution
    Source sentence
    market order executes against any of the orders being tracked, then our hypothetical order is considered filled.
  • It is sufficient to track only messages associated with orders in front of the hypothetical limit order in the queue. data
    Source sentence
    It suffices to track the messages associated to orders in front of the hypothetical limit order in the queue and market orders arriving above/below the limit price of the hypothetical order, depending on the side it was placed in the book.
Show 1 more
  • Right‑censored log‑likelihood is a proper scoring rule for training the model. model
    Source sentence
    To train and evaluate the model, we use right‑censored log‑likelihood, which is a proper scoring rule, unlike other scoring rules commonly used in the literature.

Data

  • LOBSTER message data · Lobster order-book public
    October 2022 · NASDAQ stocks (e.g., AAPL, AMZN, BIDU, etc.)
    Source sentence
    we use Lobster message data, which provides information on events that update the state of the order book in the Nasdaq stock exchange.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • The proposed monotonic encoder‑decoder convolutional‑Transformer model significantly outperforms state‑of‑the‑art survival analysis benchmarks on real LOB data. performance
    Source sentence
    Our method significantly outperforms those typically used in survival analysis literature.
  • Across multiple stocks, the MN‑Conv‑Trans model achieves up to ~90% improvement over the MN‑MLP baseline for observed limit orders. predictive-accuracy
    Source sentence
    MN‑Conv‑Trans ... 90.31 ... 92.47 ... 79.10 ... 91.04 ... 80.99 ... 87.20 ... 85.00 ... 77.43 ... 90.59
  • Placing orders inside the spread makes them 3 to 6 times more likely to be filled and reduces time‑to‑fill proportionally. statistical
    Source sentence
    it is approximately 3 to 6 times more likely that an order is filled when it improves the best quotes and reduces the time to fill by a similar proportion

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 17 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 10 papers in the archive (9 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.