Search and Evaluation of Stock Ranking Rules Using Internet Activity Time Series and Multiobjective Genetic Programming

Abstract

Hundreds of millions of people are daily active on the internet. They view webpages, search for different terms, post their thoughts, or write blogs. Time series can be built from the popularity of different terms on webpages, search engines, or social networks. It was already shown in multiple publications that popularity of some terms on Google, Wikipedia, Twitter, or Facebook can predict moves on the stock market. We are trying to find relations between internet popularity of company names and the rank of the company’s stock. Popularity is represented by time series of Google Trends data and Wikipedia view count data. We use multiobjective genetic programming (MOGP) to find these relations. MOGP is using evolutionary operators to find tree-like solutions to multiobjective problems and is popular in financial investing in the last years. Stock rank is used in an investment strategy to find stock portfolios in our implementation, revenue and standard deviation are used as objectives. Solutions found by the MOGP algorithm show the relation between the internet popularity and stock rank. It is also shown that such data can help to achieve higher revenue with lower risk. Evaluation is done by comparing the results with different investment strategies, not only the market index.