Modeling Momentum Spillover with Economic Links Discovered from Financial Documents

Momentum spillover is a market anomaly well-acknowledged in finance literature. This paper proposes using novel economic links discovered from financial documents as a momentum spillover channel, followed by modeling with graph attention networks. These text-based economic links are constructed using contextual embeddings extracted with pre-trained language models from various sections of company annual reports and earnings call transcripts. We examine the effectiveness of our proposed methods based on point-in-time S&P500 constituents from 2010/01/01 to 2022/12/31 in the US stock market. We compare our proposed model against the mean aggregator of peer firms' momentum baseline and Monte Carlo experiments based on randomized nodes or edges. Our results show that our proposed graph neural network model significantly outperforms the peer firm's momentum aggregation baseline. Furthermore, economic links discovered in some sections of company annual reports and earnings call transcripts are useful for modeling momentum spillover. In particular, the economic link constructed from management discussion and analysis from earnings call transcripts outperforms the industry link, which represents the well-acknowledged industry momentum.