Relational Fusion-based Stock Selection with Neural Recursive Ordinary Differential Equation Networks

Qiang Gao, Xinzhu Zhou, Li Huang, Kunpeng Zhang, Siyuan Liu, Fan Zhou

Abstract: Highlights•Introduce a flexible dynamic neural framework StockODE for stock selection.•StockODE relieves the uncertainty of stock fluctuations via the Gaussian assumption.•Present a Movement Trend Correlation to expose the time-varying relationships.•Devise an NRODE block to capture the temporal evolution of stock volatility.•Build a hierarchical hypergraph to incorporate the domain-aware dependencies.