Automatic prediction of stock price direction based on multivariate time series and machine learning

Abstract

The aim of this study is to predict the direction of the next closing price of Volkswagen AG. The study also concludes whether the stock price of Volkswagen, relies on the prices of crude oil as well as EUR/USD exchange rate. Therefore, a multivariate time series is used to form feature vectors that consist of the historical values of the same stock, prices of crude oil and prices of EUR/USD exchange rate. The direction of the next closing price of Volkswagen is predicted by comparing the performance of three supervised machine learning techniques namely, Support Vector Machine (SVM), Artificial Neural Network (ANN) and Learning Vector Quantization (LVQ). The study is focused on cross-validation of the three machine learning techniques to get the best performance from each technique. In the related works, a number of machine learning techniques are studied using different stock markets. Therefore, to the best of my knowledge, the approach of comparing SVM, ANN and LVQ for the prediction of the direction of a stock price has never been investigated. The system is designed to read three historical prices data and apply the data to a Rate of Change indicator to standardize the prices. The data is used by the three machine learning techniques to train a model and generate predicted labels as output. The predicted labels are applied to a simple back testing strategy that illustrate the profit or loss after five years. The predicted labels determine the direction of the closing price of the next day. This is useful to calculate the hit rate and performance of the techniques. Also, this will help investors decide whether to buy, sell or hold to their investments. The results conclude that ANN algorithm outperformed SVM and LVQ in regards to the performance of the machine learning technique, as the former achieved the highest hit rate of 52.24% during training of the model. The back testing results show that SVM, leads to the highest pro t when using the simple trading strategy created.