StochLOB: Heavy-Tailed Score Matching for Robust Limit Order Book Trend Prediction

By Arshia Abolghasemi, Seyed Reza Tavakoli, Parsa Naderi, Nazanin Mirzaei, Behrad Jahani, Saba Yousefzadeh, Tahereh Javaheri, Mohammad Hosein Rohban, Gholamali Aminian, Ali Habibnia

Rating

1678

Battle Count: 54

Relevance

9/10

Highly relevant for high-frequency trading and LOB prediction. Addresses the critical issue of heavy tails in financial data which standard Gaussian-based deep learning models often ignore, leading to poor performance in stress regimes.

Implementation Complexity

7/10

Requires implementing custom noise samplers (Jump-Lévy, Alpha-Stable) and a score table construction via Monte Carlo integration. The architecture itself is standard (BiGRU + Attention), but the training loop involves dual heads and specific loss weighting.

Reproducibility

4/5

Code and configuration files are available on GitHub. Coinbase data is public, but Feishu A-share data details are less clear regarding public availability, though the protocol is described.

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.

Links

arXiv:2610.21081 · Code · Paper rankings

The interactive Everscope explorer (charts, battles, favorites) loads below.