Robust trend prediction method models heavy-tailed limit order book data
StochLOB: Heavy-Tailed Score Matching for Robust Limit Order Book Trend Prediction
Machine Learning
Summary
Market trading data, like buy and sell orders, often shows sudden jumps and heavy swings that are hard to predict with traditional methods relying on normal noise assumptions. The authors propose a new training approach called STOCHLOB that uses noise better suited to these sudden jumps and heavy tails during model training, but keeps predictions fast and deterministic afterward. This helps the model focus more on extreme but important events, improving trend prediction accuracy on cryptocurrency trade data. They also tested the method on stock market data, exploring practical use under trading costs.
What this means in practice
- •For quantitative traders: Improve trend prediction models for cryptocurrency order flows by incorporating heavy-tailed noise during training to better capture sudden market jumps.
- •For financial data engineers: Develop more robust time series models for limit order book data using novel noise corruptions that emphasize rare but impactful market events.
Authors
Arshia Abolghasemi, Seyed Reza Tavakoli, Parsa Naderi, Nazanin Mirzaei, Behrad Jahani, Saba Yousefzadeh, Tahereh Javaheri, Mohammad Hosein Rohban, Gholamali Aminian, Ali Habibnia
Abstract
Score-based diffusion methods usually perturb data with Brownian Gaussian noise, corresponding to a light-tailed forward stochastic differential equation (SDE). For market time series this SDE can be a poor abstraction: limit order book (LOB) order flow is heavy-tailed, bursty, and occasionally jump-like. We study STOCHLOB, a training-only score-matching framework that changes the forward diffusion corruption law while leaving the deployed predictor deterministic. Instead of Brownian-only Gaussian corruption, the noising process uses two non-Gaussian SDE-inspired alternatives: jump-Lévy noise through a compound-Poisson variance mixture, and clipped $α$-stable noise through a positive-stable Gaussian scale mixture. These corruptions inject finite jumps and heavy-tailed residuals into training without requiring reverse-time diffusion sampling at inference. The resulting score target has a useful interpretation: its radial factor $h_t(r)$ is the posterior expected corruption precision, so large residuals are weighted according to the latent corruption scale that most plausibly generated them. On chronologically held-out Coinbase LOB data, the proposed variants remain competitive on uncorrupted BTC and altcoin test horizons and improve several realized tail-window stress diagnostics. An exploratory Feishu A-share transfer study further tests whether the probability outputs remain usable under transaction costs and trading-rule constraints. Code and configuration files are available at https://github.com/ArshiAbolghasemi/Penny.