Stock Recommendations Using Information Connections from Financial Bipartite Graph
In this study we propose an approach for individual investor to easily follow the value investors’ strategies based on the network analysis and machine learning on the bipartite investor-stock network constructed from a dataset of 13-F filings: quarterly investment information collected by SEC from institutional investors. We convert the bipartite graph into a directed investor-investor network, apply several analyses relating to Motif, PageRank and HITs, and detect secure investor communities by Louvain algorithm. We also apply JODIE, a temporal graph neural network model, on the original bipartite investor-stock graph to predict future investments of those investor communities. We then use these predictions to create a portfolio recommendation for individual investors.