Vector-Quantized Discrete Latent Factors Meet Financial Priors: Dynamic Cross-Sectional Stock Ranking Prediction for Portfolio Construction

Researchers introduced PRISM-VQ, a dynamic factor model for stock ranking and portfolio construction.

It combines expert financial priors, vector-quantized latent factors, and a structure-conditioned Mixture-of-Experts to model time-varying factor loadings.

Tests on CSI 300 and S&P 500 data showed stronger return prediction and portfolio performance than leading baselines.

The approach aims to improve both signal quality and interpretability in cross-sectional return forecasting.