Driving · 2026-10-07

STOCK-JEPA: Prior-Anchored Latent Revision Representation Learning in Equity Markets

Yizhi Luo, Jiahe Yi, Jianhui Zhang, Shuo Sun

arXiv:2610.07006PDF

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

Stock-JEPA is a joint-embedding predictive framework for equity markets that learns predictable incremental latent “revisions” relative to a point-in-time financial prior. The method builds an anchor from fixed, multi-horizon return/risk prior statistics, predicts context-conditioned revisions to the target latent, freezes representation modules, and trains a downstream return readout for cross-sectional ranking and portfolio performance.

Why it matters

The authors aim to improve equity representation learning under low signal-to-noise by separating what a structured financial prior can predict from what additional historical context adds. They also provide a theoretical risk-reduction guarantee for optimal revision, and evaluate downstream ranking/portfolio utility from frozen Stock-JEPA representations.

Method

  • Construct fixed point-in-time prior statistics by multi-output Ridge regression predicting return/volatility/drawdown/equal-weighted market-relative return over 5- and 21-trading-day horizons; map them via a prior projector into the target latent space as an anchor.
  • Encode 126 trading-day historical context with a shared Transformer; a context-conditioned revision predictor estimates the future latent representation’s predictable displacement from the anchor, trained with separate anchor and revision losses.
  • Freeze all representation modules after training and fit a downstream MLP readout on 21-day log-return labels to evaluate cross-sectional ranking (Pearson IC, Spearman RankIC) and portfolio performance (net annualized return, net Sharpe, MDD).
Abstract (from arXiv)

Learning effective representations helps characterize the structure and dynamics of equity markets from financial data with a low signal-to-noise ratio. Black-box deep models can capture complex patterns but may overfit sample noise and lack explicit economic structure. Meanwhile, classic linear financial models provide interpretable references, but their oversimplified assumptions leave non-linear signals uncaptured. To combine the strengths of these two directions, we propose Stock-JEPA, a joint-embedding predictive framework that learns predictable incremental revisions relative to a point-in-time financial prior. First, we leverage a low-complexity financial model to produce fixed statistics summarizing multi-horizon return and risk. A prior projector then maps these statistics into the target encoder's latent space as an anchor. Second, we design a context-conditioned revision predictor to estimate the future representation's predictable displacement from the anchor. Separate losses update the two branches: the anchor learns from prior statistics, while the revision captures additional predictable information from historical context. Third, we freeze all representation modules and train a downstream readout, evaluating its forecasts through cross-sectional ranking and portfolio performance. Theoretically, we prove that optimal revision reduces the prior anchor's expected squared error for the same future representation by exactly $\mathbb{E}[\|\boldsymbol{\Delta}\|_2^2]$. This non-negative gain is the expected squared magnitude of the additional signal predictable from historical context. Experimentally, Stock-JEPA outperforms 13 strong baselines across large-scale China and U.S. equity universes on 5 key evaluation metrics. Ablation studies and representation analysis further demonstrate the value of the learned revisions for representation learning in equity markets.

Related papers

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First public release: 2026-10-04 · Indexed by Pub-AI: 2026-10-07 UTC

preprint

STOCK-JEPA: Prior-Anchored Latent Revision Representation Learning in Equity Markets

Yizhi Luo, Jiahe Yi, Jianhui Zhang, Shuo Sun

Preprint posted: 2026-10-04 · Original record ↗

arxiv
2610.07006
Explicit source versions (1)
  • v1 · 2026-10-04
  1. Preprint available

    Source date: 2026-10-04 · First detected: 2026-10-25 UTC · Historical baseline

    Source evidence (1)
Abstract · arxiv

Learning effective representations helps characterize the structure and dynamics of equity markets from financial data with a low signal-to-noise ratio. Black-box deep models can capture complex patterns but may overfit sample noise and lack explicit economic structure. Meanwhile, classic linear financial models provide interpretable references, but their oversimplified assumptions leave non-linear signals uncaptured. To combine the strengths of these two directions, we propose Stock-JEPA, a joint-embedding predictive framework that learns predictable incremental revisions relative to a point-in-time financial prior. First, we leverage a low-complexity financial model to produce fixed statistics summarizing multi-horizon return and risk. A prior projector then maps these statistics into the target encoder's latent space as an anchor. Second, we design a context-conditioned revision predictor to estimate the future representation's predictable displacement from the anchor. Separate losses update the two branches: the anchor learns from prior statistics, while the revision captures additional predictable information from historical context. Third, we freeze all representation modules and train a downstream readout, evaluating its forecasts through cross-sectional ranking and portfolio performance. Theoretically, we prove that optimal revision reduces the prior anchor's expected squared error for the same future representation by exactly $\mathbb{E}[\|\boldsymbolΔ\|_2^2]$. This non-negative gain is the expected squared magnitude of the additional signal predictable from historical context. Experimentally, Stock-JEPA outperforms 13 strong baselines across large-scale China and U.S. equity universes on 5 key evaluation metrics. Ablation studies and representation analysis further demonstrate the value of the learned revisions for representation learning in equity markets.