Stock trends prediction by hypergraph modeling

Abstract

This paper presents a new stock price trends prediction algorithm using hypergraph model. Hypergraph modeling offers a significant advantage over traditional graph modeling in terms of triadic or higher relationship description within different stock portfolios over a certain period of time. Under the hypergraph model, each stock will be abstracted as a vertex of hypergraph; the hyperedges can be built by seeking the synchronous relationship of the stocks trends. In order to acquire more refined hyperedges and to avoid the tremendous growing quantity of hyperedges, we employ the frequent item sets to construct hyperedges. Therefore the prediction problem for stock trends is converted to hypergraph partitioning problem. Multilevel paradigm is then applied to do hypergraph partitioning instead of the traditional recursive bisection paradigm. Thus we get a series of stocks section, and the stock price trends can be concluded by analysis the whole section. Experiment result shows that our proposed scheme achieves fine stock trend prediction and the computation is significantly fast as well.