Joint Latent Topic Discovery and Expectation Modeling for Financial Markets

In the pursuit of accurate and scalable quantitative methods for financial market analysis, focus has shifted from individual stock models to those capturing interrelations between companies their stocks. However, current relational are limited by reliance on predefined relationships exclusive consideration immediate effects. To address these limitations, we present a groundbreaking framework analysis. This approach, our knowledge, is first jointly model investor expectations automatically mine latent relationships. Comprehensive experiments conducted China’s CSI 300, one world’s largest markets, demonstrate that consistently achieves an annual return exceeding 10%. performance surpasses existing benchmarks, setting new state-of-the-art standard in prediction multiyear trading simulations (i.e., backtesting).