Comparison of Regressions and Multi-feature LSTM in Stock Price Prediction for Top Car Companies

Abstract

Machine learning (ML) has gradually permeated in quantitative finance, among which Regressions and Deep Learning are two well-known ML domains. However, as traditional prediction methods, regressions have shown their limitation in more complicated datasets with temporal sequence and non-linear relationships between features and responses. In order to compare the capability of widely-used deep learning methods SVM and LSTM with that of the methods in regressions, this study acquired historical data of 10 car companies stocks from Kaggle.com to simulate the share price in the next period based on four different regression methods and two Deep Learning methods, respectively. Moreover, the author uses RMSE and MAPE as evaluation metrics to compare those models training and testing sets fitness. According to the result, LSTM has better fitness than other regression models in predicting time series data. Although share price from Kaggle might not reflect the overall market condition, this study would contribute to mend the research gap of empirical application of different machine learning methods.