Forecasting stock prices of Russian public companies using an LSTM neural network

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Subject. Methods for Short‑Term Forecasting of Stock Prices of Russian Public Companies Based on Neural Network Technologies. Objectives. To develop a model for short‑term forecasting of stock prices of Russian public companies using a neural network, and to create a comprehensive solution for scenario analysis of price movement trajectories across the entire market to enable the formation and calibration of a securities portfolio. Methods. The study used stock price data for public companies listed on the Moscow Exchange (442,674 quotes) from 2014 onwards. The batch download method of the Moexalgo service was applied. Subsequently, deep learning techniques were employed, within the framework of which a neural network with a long short‑term memory (LSTM) architecture was constructed. Results. Models were developed for 173 Russian public companies. These models enable stock prices to be forecasted with fairly high accuracy for a 30‑day horizon based on historical data series, which can improve short‑term strategy in securities portfolio management. Conclusions and Relevance. The obtained results can be applied to form a theoretical securities portfolio for the Russian market in order to develop a short‑term financial strategy. The results can be used by financial analysts, traders, and valuation specialists when modelling a portfolio of financial investments.