:   .., ..
:  
:  118
:   -
:  2025
:   .., .. // . - 2025. - . 118. - .250-285.
:   , , , ,
(.):  forecasting models, machine learning methods, multivariate multistage forecasting, stock market, stock price
:   , , . , , , . , , , , . , , . , . (30 ) . , S&P 500. , .
(.):  Numerous studies in the field of forecasting stock prices, in particular shares, are aimed at finding more accurate and efficient models. It should be noted that attention to multidimensional forecasting, which allows for a more accurate forecast, often remains underestimated. This is due to the fact that it requires significant computing resources, additional data arrays and complex models, which makes it difficult to apply in real conditions. In this regard, it is relevant to form more simplified but effective models that can give good results with lower computational costs, an available set of unambiguously estimated data and simplified settings, while maintaining sufficient accuracy for practical use. The results of the work presented in the paper are aimed at studying this issue: methods for multidimensional stock price forecasting models based on machine learning methods and modern neural network architectures for constructing a multi-stage forecast for a 30-day horizon were selected, formed and tested. A comparative analysis was conducted and it was shown that the Attention + LSTM model provides the highest forecast accuracy, which confirms the effectiveness of attention mechanisms for extracting significant temporal dependencies.

PDF

: 57, : 28, : 26.


© 2007.