High-frequency trading strategies based on deep learning algorithms and their profitability

As an important algorithm, deep learning has been applied successfully to image processing, speech recognition, machine translation and other fields. Here, deep learning algorithms were applied to high-frequency trading. Convolutional neural network(CNN) and long short-term memory(LSTM) neural network were selected to build up and down classification models, respectively. Based on the models, high-frequency trading strategies were proposed. Then the data of bitumen futures contract was used for back-testing and empirically analyzing the superiority of the strategies. In back-testing, deep learning algorithms were compared with artificial neural network(ANN). The results show that both strategies based on CNN and LSTM neural network exhibit better profitability and generalization ability. In addition, the winning rates and expected returns of the two strategies are also better.