RECURRENT NEURAL NETWORKS AS A MEANS OF PREDICTIVE ANALYSIS OF PRICE MOVEMENTS IN FINANCIAL MARKETS

Suchev Nikolay Evgenievich, Panteleeva Leysan Renatovna

The article explores the use of recurrent neural networks as assistants to inverters and traders for making trading decisions. To test the hypothesis, a neural network with long-term short-term memory (LSTM) was selected, and gold commodity futures Gold Aug 23 was selected as the analyzed asset. The study demonstrates the analysis and development of a methodology for building and training an LSTM model using the Python programming language and libraries for data analysis. The development plan and methodology include four main stages. At the first stage, data is being prepared on the basis of which the LSTM model will be trained: the movement of asset prices and technical indicators RSI, EMAF, EMAM, EMAS. After that, the stages begin: creating the architecture of the neural network model and training it on a test sample of data. The model is trained using the Adam optimization algorithm, which adjusts the accuracy of the model based on a sample of data for training. The final stages in the development of a neural network are to evaluate the accuracy of the model prediction on a test sample using such coefficient metrics as MSE, R2 and MAE. The results of the study demonstrate the high accuracy of the created LSTM model for predicting asset price movements in financial markets. The methodology provided in the paper can be useful in developing trading strategies and making decisions based on them.

Keywords:forecasting financial asset prices, recurrent neural networks (RNN), neural networks with long-term short-term memory (LSTM), model training, TensorFlow and Keras machine learning libraries, python, calculation of technical indicators, Adam optimization algorithm, estimation of model accuracy, coefficient metrics.