Reproduce “Market-Conditioned JEPA: Learning from Unpredictable Market Shocks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuqi Wang, Fengyuan Liu, Mukai Li, Haochen Luo, Lei Li, Qi Liu
MC-JEPA learns stock representations by predicting future target embeddings, using other stocks’ future embeddings during training to condition predictions on a market proxy; inference uses each stock’s history alone. Using frozen representations and ridge probes on Chinese A-share 2023–2025, it scores 0.374 ± 0.003 on 20-day beta RankIC, versus 0.359 for per-asset JEPA, but does not lead 20-day returns; generalization is untested.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.