A comparative study of a recurrent neural network and support vector machine for predicting price movements of stocks of different volatilites

Recurrent neural networks like the long short-term memory (LSTM), and support vector machines (SVM) have been widely adopted to predict the price movements of stocks in China or other countries in recent years no matter the underlying stock is of low or high volatility, that is typically less or very fluctuating in its daily returns over a specific period of time. In this work, different performance measures of the LSTM and SVM are carefully analyzed and compared on a range of stocks exhibiting different volatilities. In addition, various data pre-processing techniques including the principal component analysis are utilized to enhance their overall performance. Surprisingly, the empirical results reveal that the overall performance of the original SVM excels on all stocks and also stocks of low volatility in the Shanghai Stock Exchange 50 (SSE 50) Index denoting the top 50 stocks listed on the Shanghai Exchange whereas the original LSTM consistently achieves the best overall performance of high-volatility stocks in the SSE 50 Index. More importantly, our work sheds light on numerous directions for future investigation.