Paper: arXiv 2509.22985

Authors: Haochuan, Wang

Abstract

Liquidity withdrawal is a critical indicator of market fragility. In this project, I test a framework for forecasting liquidity withdrawal at the individual-stock level, ranging from less liquid stocks to highly liquid large-cap tickers, and evaluate the relative performance of competing model classes in predicting short-horizon order book stress. We introduce the Liquidity Withdrawal Index (LWI) – defined as the ratio of order cancellations to the sum of standing depth and new additions at the best quotes – as a bounded, interpretable measure of transient liquidity removal. Using Nasdaq market-by-order (MBO) data, we compare a spectrum of approaches: linear benchmarks (AR, HAR), and non-linear tree ensembles (XGBoost), across horizons ranging from 250,ms to 5,s. Beyond predictive accuracy, our results provide insights into order placement and cancellation dynamics, identify regimes where linear versus non-linear signals dominate, and highlight how early-warning indicators of liquidity withdrawal can inform both market surveillance and execution.

Complexity vs Empirical Score

  • Math Complexity: 6.0/10
  • Empirical Rigor: 8.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced econometric models (AR/HAR, XGBoost) and rigorous feature selection (MI, LASSO) on high-frequency Nasdaq MBO data, with detailed methodology and backtest-ready evaluation using walk-forward cross-validation.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Forecast Liquidity Withdrawal"] --> B["Methodology<br>Liquidity Withdrawal Index LWI"]
  A --> C["Data Input<br>Nasdaq Market-by-Order MBO"]
  B --> D["Model Comparison"]
  C --> D
  D --> E["Linear Benchmarks<br>AR / HAR"]
  D --> F["Non-Linear Ensemble<br>XGBoost"]
  E & F --> G["Computational Process<br>Forecast Horizons 250ms to 5s"]
  G --> H{"Key Findings & Outcomes"}
  H --> I["Predictive Accuracy<br>Model Class Dominance"]
  H --> J["Market Insights<br>Order Cancellation Dynamics"]
  H --> K["Applications<br>Surveillance & Execution"]

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 and evaluates a framework for forecasting liquidity withdrawal using a Liquidity Withdrawal Index and machine learning models.

Source sentence
We propose and evaluate a framework for forecasting liquidity withdrawal using Nasdaq MBO data and a Liquidity Withdrawal Index (LWI).

Assumes 5

  • The Liquidity Withdrawal Index (LWI) is stationary, allowing autoregressive modeling without differencing. statistical
    Source sentence
    Across all four tickers, the ADF statistics are strongly negative with p-values near zero (Table 2), rejecting the unit root null. Thus, LWI is stationary, making it valid for autoregressive modeling without
  • Liquidity dynamics exhibit short‑memory, so only the first one or two lags are predictive. statistical
    Source sentence
    Autocorrelation plots reveal significant dependence at the first one or two lags, followed by rapid decay. This pattern is consistent with a stationary but short-memory process.
  • Linear AR(5) and HAR models capture short‑term linear dependence in liquidity withdrawal. model
    Source sentence
    We use AR(5), which provides a one-second lookback to capture short-term linear dependence in liquidity withdrawal.
  • Walk‑forward cross‑validation with an embargo period prevents information leakage and mimics live deployment. data
    Source sentence
    We apply walk-forward cross-validation with an embargo period to prevent information leakage.
Show 1 more
  • The study focuses on a one‑hour window of order‑by‑order data, assuming that this slice is representative of broader market behavior. data
    Source sentence
    For tractability, our study focuses on July 30, 2025, 14:00–15:00 ET, covering several actively traded symbols.

Data

Backtest sample: Nasdaq stocks · equities · intraday

  • Nasdaq Market‑by‑Order (MBO) data · Databento order-book commercial
    2025-07-30 14:00–15:00 ET · several actively traded U.S. equities (e.g., HIMS, NBIS, RKLB, SNAP)
    Source sentence
    MBO provides the full order flow — every add, cancel, replace, and trade message from the NASDAQ matching engine — at nanosecond resolution.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • XGBoost significantly outperforms AR/HAR at the 5‑second horizon, achieving R² above 0.90. predictive-accuracy
    Source sentence
    XGBoost significantly outperforms AR/HAR at 5 s horizons, consistent with the growing role of non-linear structure in driving liquidity withdrawal.
  • Linear AR and HAR models are competitive at 1–2 s horizons, while no model provides robust predictability at 250 ms. performance
    Source sentence
    Our comparisons suggest that AR/HAR are adequate at 1–2 s horizons, while boosted trees dominate at 5 s, and no model delivers robust predictability at 250 ms.
  • The Liquidity Withdrawal Index is stationary across all four tickers, validating autoregressive approaches. statistical
    Source sentence
    Across all four tickers, the ADF statistics are strongly negative with p-values near zero (Table 2), rejecting the unit root null. Thus, LWI is stationary.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 25 related papers). Coverage is partial, so this is neither replication nor refutation.

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.