JEPA · 2026-11-11

Market-Conditioned JEPA: Learning from Unpredictable Market Shocks

Yuqi Wang, Fengyuan Liu, Mukai Li, Haochen Luo, Lei Li, Qi Liu

arXiv:2611.13572PDF

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

TL;DR

The authors introduce Market-Conditioned JEPA (MC-JEPA), a joint-embedding predictive model for financial time series that conditions training prediction on a market factor constructed from other assets’ future embeddings. In a latent factor model, conditioning makes exposure necessary for optimal prediction and reduces noise in learning asset-specific signals. On 11 years of minute-level Chinese A-share data, MC-JEPA improves ridge-probe RankIC and trading backtest metrics over per-asset JEPA.

Why it matters

MC-JEPA targets a key issue in per-asset self-supervised objectives: unpredictable market shocks become irreducible noise, providing limited supervision for market exposure. By using a leave-one-out market condition during training and a factor-structured predictor, the method learns per-asset representations that can encode market exposure without cross-sectional inference.

Method

  • Introduce MC-JEPA: a per-asset JEPA encoder plus a factor-structured predictor conditioned during training on a leave-one-out market factor from other assets’ future target embeddings.
  • Train the predictor so market information enters through loadings derived from the asset’s own representation (single-index factor structure), while keeping the deployed encoder per-asset and independent of cross-sectional inputs.
  • Use ridge probes on frozen representations (and a historical backtest) to evaluate improvements across returns, market exposure, risk, and liquidity; include ablations and label-free exposure interpretability checks.

Limitation

The paper states that even on real data, the market condition removes only a modest fraction of unconditional error (4.5 ± 0.9% over all horizons), suggesting denoising is limited and does not fully explain return gains.

Abstract (from arXiv)

Much of the variation in financial time series is driven by market-wide shocks that cannot be predicted from an asset's own history. Per-asset self-supervised learning treats this variation as irreducible noise. Under a latent factor model, we show that conditioning prediction on the realized market factor makes the asset's market exposure necessary for optimal prediction, while reducing the noise against which the asset-specific component is learned. Building on this insight, we introduce Market-Conditioned JEPA (MC-JEPA), a joint-embedding predictive architecture for financial time series. During training, we construct a leave-one-out market condition for each asset from the future target embeddings of other assets and incorporate it through factor loadings derived from the asset's own representation. On 11 years of minute-level Chinese A-share data, MC-JEPA improves over per-asset JEPA across nine targets spanning returns, market exposure, risk, and liquidity, with the strongest and most statistically significant gains in exposure, risk, and liquidity. It achieves the best RankIC on all liquidity targets and on return horizons from two hours to five days, outperforming time-series foundation models and self-supervised baselines, and delivers the best portfolio performance in a historical backtest. Ablations show that the return and exposure gains require the condition to be aligned with the realized market movement. After restoring the volatility scale removed by normalization, the predictor's factor loadings rank stocks by market beta nearly as well as a supervised probe, without labels. A synthetic factor-model study with known parameters shows that conditioning improves the recovery of both market exposure and the asset-specific component.

Related papers

Publication record

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

First public release: 2026-11-10 · Indexed by Pub-AI: 2026-11-11 UTC

preprint

Market-Conditioned JEPA: Learning from Unpredictable Market Shocks

Yuqi Wang, Fengyuan Liu, Mukai Li, Haochen Luo, Lei Li, Qi Liu

Preprint posted: 2026-11-10 · Original record ↗

arxiv
2611.13572
Explicit source versions (1)
  • v1 · 2026-11-10
  1. Preprint available

    Source date: 2026-11-10 · First detected: 2026-11-11 UTC · Historical baseline

    Source evidence (2)
Abstract · arxiv

Much of the variation in financial time series is driven by market-wide shocks that cannot be predicted from an asset's own history. Per-asset self-supervised learning treats this variation as irreducible noise. Under a latent factor model, we show that conditioning prediction on the realized market factor makes the asset's market exposure necessary for optimal prediction, while reducing the noise against which the asset-specific component is learned. Building on this insight, we introduce Market-Conditioned JEPA (MC-JEPA), a joint-embedding predictive architecture for financial time series. During training, we construct a leave-one-out market condition for each asset from the future target embeddings of other assets and incorporate it through factor loadings derived from the asset's own representation. On 11 years of minute-level Chinese A-share data, MC-JEPA improves over per-asset JEPA across nine targets spanning returns, market exposure, risk, and liquidity, with the strongest and most statistically significant gains in exposure, risk, and liquidity. It achieves the best RankIC on all liquidity targets and on return horizons from two hours to five days, outperforming time-series foundation models and self-supervised baselines, and delivers the best portfolio performance in a historical backtest. Ablations show that the return and exposure gains require the condition to be aligned with the realized market movement. After restoring the volatility scale removed by normalization, the predictor's factor loadings rank stocks by market beta nearly as well as a supervised probe, without labels. A synthetic factor-model study with known parameters shows that conditioning improves the recovery of both market exposure and the asset-specific component.