Combining Technical and Sentiment Analysis Under a Genetic Programming Algorithm

Throughout the years, a lot of interest has been given to algorithmic trading, due to development of the stock market and provided securities. In the field of algorithmic trading, genetic programming (GP) is a very popular algorithm, due to its ability to produce white-box models, effective global search, and good exploration and exploitation. In this paper, we propose a novel GP algorithm to combine the features of two financial techniques. Firstly, technical analysis that studies the financial market action by looking into past market data. Secondly, sentiment analysis, which is used to determine the sentiment strength from a text in order to decide its implication in the stock market. Both techniques create indicators that are used as inputs in machine learning algorithms, with both showing in past studies the ability to return profitable trading strategies. However, these techniques are rarely used together. Thus, we examine the advantages when combining technical and sentiment analysis indicators under a GP, allowing trees to contain technical and/or sentiment analysis features in the same branch. We run experiments on 60 different stocks and compare the proposed algorithm's performance to two other GP algorithms, namely a GP that uses only technical analysis features (GP-TA), and a GP that uses only sentiment analysis features (GP-SA). Results show that the GP using the combined features statistically outperforms GP-TA and GP-SA under several different financial metrics, as well as the financial benchmark of buy and hold.