Применение моделей машинного обучения для многомерного среднесрочного прогнозирования стоимости акций

Управление большими системами: сборник трудов. 2025. № 118. С. 250–285.

Sizykh N., Кудрявцева Н. С.

Numerous studies in the field of forecasting securities prices, particularly stock prices, are aimed at finding more accurate and effective models. However, attention to multivariate forecasting, which allows for more accurate forecasts, remains underappreciated, as its implementation requires a significant increase in computing resources. Therefore, the selection of more simplified yet effective models is relevant. These models can achieve good results with lower computational costs, an accessible set of unambiguously estimated data, and a simplified setup, while maintaining sufficient accuracy for practical use. The research results presented in this article are aimed at solving this problem. The authors selected, developed, and tested methods for multivariate stock price forecasting models based on machine learning methods and modern neural network architectures. A comparative analysis of the results of a medium-term stock price forecast (30 days) using multivariate forecasting models was conducted. The test was conducted using stocks included in the S&P 500 index. During the study, additional data sets were selected that contribute to increased forecast accuracy and are available in open sources.

Keywords: фондовый рынок, модели прогнозирования, стоимость акций, stock price, stock market, методы машинного обучения, Machine learning methods, forecasting models, multivariate multistage forecasting, многомерное многоэтапное прогнозирование