Employing Machine Learning Algorithms for Stock Index Prediction

Abstract

Machine learning is a subset within artificial intelligence field. Its fundamental philosophy is learning the general structure from a large amount of data and making predictions with new data. In recent years, more and more academic researchers attempted to apply machine learning algorithms to process stock data and make predictions. This paper predicts the trends of the Chinese CSI 300 index with machine learning algorithms, including support vector machine classifier (SVM), classification and regression tree algorithm (CART), C5.0 algorithm for decision tree (C5.0), back propagation neural network(B-P neural network), and logistic model. In addition to the basic trading indicators, technical indicators, and valuation indicators, this paper constructs quantitative indicators of market awareness and investor sentiment based on Baidu Index. In the empirical results, the accuracy and precision of all models exceed 50%. The performances of CART and logistic model are excellent in the accuracy and precision, while the B-P neural network performs worst. As for the ROC curves, CART performs best, followed by the B-P neural network model. According to the results of back testing, the total returns of CART, logistic model, and C5.0 exceed the index, and logistic model performs best in the three back testing periods. In conclusion, logistic model and CART overperform the other three models in predicting the stock index movements.