Research map / Probabilistic and scientific modeling / Molecules and proteins
Joint-Embedding Predictive Learning of Latent Market States in U.S. Equities
Simon Mahns, Randall Balestriero, Mido Assran
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use JEPA latent prediction instead of MAE reconstruction (ablation) | market-state embedding · kNN retrieval, structure descriptors | R2: 0.25 -> 0.34 | Improved |
| use JEPA instead of MAE (ablation) | market-state embedding · kNN retrieval, PC1 share | R2: 0.12 -> 0.29 | Improved |
| use JEPA instead of MAE (ablation) | market-state embedding · kNN retrieval, direction descriptors | R2: 0.04 -> 0.05 | No clear effect |
| use structured contiguous-time masking instead of random i.i.d. masking (ablation) | JEPA · kNN retrieval, direction descriptors | Spearman rho: 0.19 -> 0.14 | Improved |
| use structured masking instead of random i.i.d. masking (ablation) | JEPA · kNN retrieval, structure descriptors | Spearman rho: 0.55 -> 0.56 | No clear effect |
| keep explicit volatility input channels (ablation) | JEPA · Mantel geometry alignment | Mantel rho: 0.32 -> 0.35 | Improved |
| keep explicit volatility input channels (ablation) | JEPA · Recovery prediction (eff rank up) | delta AUC: +0.14 -> +0.24 | Improved |
| add JEPA embedding to geometry-scalar logistic regression | logistic regression · Recovery prediction (eff rank up) | ROC-AUC: 0.690 -> 0.800 | Improved |
| add JEPA embedding to geometry scalars for stress-onset prediction | logistic regression · Stress-onset prediction | delta AUC: +0.11, +0.02 | No clear effect |
| compare JEPA with volatility-only representation for news alignment | JEPA · GDELT news alignment W=63 (train+val) | Delta(W): 0.0024 -> 0.0053 | Improved |
| remove explicit volatility channels for news alignment (ablation) | JEPA · GDELT news alignment W=63 (train+val) | Delta(W): 0.0053 -> 0.0080 | No clear effect |
| evaluate FOMC announcement days | embedding displacement · Embedding displacement | Cohen's d: d = 0.06, p = .63 | No clear effect |
Closest papers by meaning
- Towards Financial World Modeling (NeurIPS 2026 · similarity 0.88)
- MASTER: Market-Guided Stock Transformer for Stock Price Forecasting (AAAI 2024 · similarity 0.84)
- Market-GAN: Adding Control to Financial Market Data Generation with Semantic Context (AAAI 2024 · similarity 0.84)
- Drive vs. Decay: On the Training Dynamics of Joint-Embedding Predictive Architectures (NeurIPS 2026 · similarity 0.84)
- Retrieval-Augmented Diffusion Modeling for Stochastic Discount Factor Portfolios (NeurIPS 2026 · similarity 0.83)
- Enhancing Foundation Models for Time Series Forecasting via Wavelet-based Tokenization (ICML 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
