A new LSTM based reversal point prediction method using upward/downward reversal point feature sets

Abstract

Abstract A novel Long-Short Term Memory (LSTM)-based prediction model of stock price reversal point was proposed by using upward/downward reversal point feature sets. (1) Based on the combinations of candlestick indicators and technical indicators, 27 sets of feature candidates were constructed, and then the feature sets suitable to each stock in terms of URP/DRP prediction were respectively extracted. (2) LSTM-based URP/DRP predictors were constructed, the results of which are combined to improve the prediction accuracy. Using this model, reversal point prediction has been conducted for 10 Chinese stocks and 10 American stocks. In results, the mean prediction accuracy (F1) was 68.6% and 55.2% for the Chinese and the American stock markets, respectively. Results show that the average prediction accuracy has been evaluated to be higher for Chinese market by 13.4% compared to American one. Comparing with Support Vector Machine (SVM), Multilayer Perceptron (MLP) and Convolutional Neural Networks (CNN) model, F1 of proposed model has been increased by 5.9%, 11.7% and 5.3%, respectively.