AD-GAT Models Attribute-Sensitive Momentum Spillovers Between Stocks

Summary

The document summarizes a research paper on predicting stock returns by modeling momentum spillovers between related companies. It argues that conventional graph neural networks depend on predefined links, such as industry or ownership, and may pass influence between firms without accounting for their current conditions. In practice, a price move may have little relevance when its volume is low or the connected company's own valuation differs.

The proposed attribute-driven graph attention network combines company features through an attribute-sensitive aggregation mechanism, infers changing relationships from market signals with attention, and uses tensor-based feature extraction to capture interactions across signal types. The summary reports experiments on three years of S&P 500 stock data, with higher directional accuracy and AUC than the named comparison models; it gives improvements of at least 6.4% and 10.7%, respectively. These are reported research results, not evidence of live trading performance, and the page does not provide enough detail to assess implementation or robustness independently.

Key ideas

• The paper targets momentum spillovers between related companies as a source of stock return predictability.
• The proposed aggregator makes peer influence depend on the attributes of both connected companies.
• An attention mechanism infers dynamic relationships from observed market signals instead of relying only on preset links.
• Tensor-based feature extraction is used to model interactions among multiple market signals.
• The summary reports improved directional accuracy and AUC on three years of S&P 500 data, but does not establish live performance.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.