Machine Learning in Asset Price Prediction: A Study Testing whether it is Possible to Predict Stock Price Movement based on Financial News about that Stock

Abstract

The fact that machine learning techniques are used in the field of finance and trading is well known. Furthermore, natural language processing has received a lot of attention in the last few years. But less attention has been given to the application of natural language processing in technical analysis and even less to the study of the correlation between traded assets and the news about those assets. This paper explores this gap by attempting to predict the stock price movement of a company using financial news data about that company and then trying to increase the accuracy of the machine learning models by using additional financial news data about other companies whose price movement is highly correlated with the price movement of the original company. The study collects 6294 news titles for the base case where only news for Microsoft is used, 15840 news titles in the case where news for five companies is used and 72547 news titles for the case where financial news with low or no correlation with Microsoft is used. Afterwards, the data was cleaned and used to predict the price movement of Microsoft, Apple, Facebook, Amazon and Google. In order to train and test the machine learning models, Scikit learn library was used. After the models were trained, the results for the base case for Microsoft were cleared and for the other companies were mixed. In the second case, where financial news was used to predict the price movement of Microsoft, the accuracy of the machine learning models has increased. In the third case, when financial news with low or no correlation with Microsoft was used to predict its movement, the accuracy was much lower than the base case. Using the discoveries of this research, the paper argues that it is possible to predict the price movement of an asset based on the news about that asset, though to a limited degree. Furthermore, when dealing with limited data, it is possible to increase the predictive accuracy of the machine learning models by using financial news data about correlated assets.