Using a Combination of Recurrent and Convolutional Neural Networks to Forecast the Direction of Financial Instrument Price Movement

Securities market forecasting has long been of interest to analysts and mathematicians due to the obvious opportunity to monetize the research if it proves to be successful. The work of these researchers has led to the creation of various trading algorithms; however, their effectiveness has not yet been proven. With the development of computing technologies that allow implementing complex mathematical machine learning systems, the attention to this direction has increased considerably, in particular because of the introduction of neural networks. The present paper focuses on describing the initial data (pairs of price and the number of transactions available at this price) and the process of data collection and preparation for the neural network training. Moreover, the reasons for choosing the combination of recurrent and convolutional neural networks and its scheme are given, and the training results and insights are presented.