Ranking influential and influenced stocks over time using transfer entropy networks

Abstract

Influence is a concept found in nature and society and is related to the interdependence among a set of objects. In the context of a stock market, the price variation of stocks can affect the price of other stocks leading to an influence between stocks. This work leverages the notion of information flow measured by transfer entropy to build networks of stocks where directed edges indicate influence. Network centrality metrics such as Pagerank and node weight are used to rank the nodes in order to determine the top ranked influential and influenced stocks. The proposed methodology is applied to a dataset comprising of a 32-year period of the Brazilian stock market exchange. Results indicate that the top ranking of influential and influenced stocks is very dynamic under different ranking metrics, while top ranking of stocks based on financial indicators is relatively stable. Results also indicate that rankings based on financial indicators have little correlation to rankings based on influence, motivating the need for specific metrics to assess influence.