Prediction of Stock Market Performance Based on Financial News Articles and Their Classification

Abstract

Hundreds of stock market news reach the financial markets every day. In order to benefit from these movements as an investor, this work aims at developing a system with which the direction of the price fluctuations can be predicted. This paper is about building an own dataset of financial information with thousands of financial articles, and historical stock data are used to build the foundation of the following actions. This foundational data is used to build a learning algorithm. Therefore, the articles are normalized by the methods of natural language processing and converted into a matrix based on the occurrences of the individual words in the news. This matrix then serves as an endogenous variable that predicts the likely direction of market impact. In order to make that statement, the actual impact on the markets was used as an exogenous variable to train different classification algorithms. To find the best algorithm to train, we created a confusion matrix. In the end, the best algorithm gets selected to perform the prediction and as a result, our trained algorithm achieved high accuracy.