Forecasting prices of financial instruments using deep learning methods

Abstract

This study is devoted to the problem of forecasting financial instruments’ prices on the stock market, the relevance of which has increased significantly in the modern world, and which is an important component of the processes of conducting financial activities and making informed investment decisions. A review and comparative analysis of the methods proposed in the existing studies was conducted, their shortcomings and shortcomings were revealed. Based on this, a new approach to solving this problem was proposed. The proposed approach is based on taking into account a complex set of factors for forecasting, including technical indicators, fundamental analysis data and macroeconomic factors, the use of a systematic approach to the selection of predictors (factors) for forecasting and inclusion in the model, the introduction of modern feature engineering and feature selection techniques, noise removal in input data, applying NLP techniques and sentiment analysis to integrate textual data that influence market trends, thus increasing the accuracy of modeling market processes. These principles were incorporated with machine and deep learning techniques capable of taking into account time sequences of data and complex relationships and dependencies between them, and a predictive model was built for forecasting stock prices. The results of model testing and evaluation of obtained performance accuracy metrics’ values for the developed model show it’s higher accuracy in comparison with the base model chosen for comparison, and also prove the effectiveness of using the proposed approach.