ARCHITECTURE OF THE INNOVATIVE AI-MODEL FOR FORECASTING THE TIME SERIES OF SIU0 FUTURES USING VECTORIZED NEWS FROM WEBSITES IN THE CONDITIONS OF DIGITAL ECONOMY
DOI:
https://doi.org/10.25806/uu3-22021495-501Статья поступила в редакцию: 27.03.2021
Статья опубликована: 13.04.2021
Keywords:
AI system, futures contract, forecast, big data, neural network.Abstract
The developed innovative AI-model designed to predict the closing price of a SIU0 futures contract on the Moscow stock exchange, which includes Scraper programs, Word2vec neural network, Perseptron neural network on the Deductor platform, as well as the QUIK trading terminal with an integrated Lua socket, is presented.
The theoretical foundations of forecasting the time series of financial instruments have been investigated.
The experience of using artificial intelligence systems for collecting and processing Big Data in order to forecast time series, including the SIU0 futures contract, is considered.
The hypothesis was put forward and proved that with the help of an innovative AI-model, it is possible to obtain a forecast of the closing price of the SIU0 futures contract on a 15-minute timeframe. An AI system is proposed to improve accuracy and reduce financial risk when forecasting the time series of a SIU0 futures contract for exchange trading. The authors have proposed an innovative AI system in which to predict the price of a futures contract SiU0. The innovation of the proposed approach lies in the fact that not only historical data - parameters of Japanese candlesticks and volume, but also digitized “news fluctuations” from websites were used to predict the time series. In addition, in order to minimize the prediction error, the Perseptron neural network, formed on the Deductor platform, was trained on two data types 1) cost - (Pclose) and 2) logarithmic - (ln).
The proposed innovative AI-model is of great practical importance, since it provides a high forecast accuracy. So, the average size of the neural network error in the first case (Pclose) was 0.000927425, while when the second neural network (Pln) was operating, the average error size was -0.051026481. The variance of error values as a percentage of the closing price in the first case was 0.304107913, while in the second 0.343654316, or 4 hundredths better.
Информация о публикации
Финансирование: Исследование выполнено без привлечения внешнего финансирования, если иное не указано авторами.
Вклад авторов: Все авторы внесли существенный вклад в подготовку статьи, ознакомились с окончательной версией рукописи и одобрили ее к публикации.
Конфликт интересов: Авторы заявляют об отсутствии конфликта интересов, если иное не указано в публикации.
Правообладатель: Издательский дом «Академический».
Лицензия: Статья распространяется на условиях лицензии Creative Commons Attribution 4.0 International (CC BY 4.0).
Машиночитаемый файл метаданных: JATS XML
References
Word2Vec: как работать с векторными представлениями слов [Электронный ресурс] – URL: https://neurohive.io/ru/osnovy-data-science/word2vec-vektornye-predstavlenija-slov-dlja-mashinnogo-obuchenija/ (дата обращения 2 мая 2020 г.)
Ruppert D. Statistics and data analysis for financial engineering // Springer. 2019.
Jensen M., Fischer V. and Myron V. The Capital Asset Pricing Model: some empirical tests // Praeger Publishers Inc. 1972.
Eugene F. Fama and James D. MacBeth Risk, Return and Equilibrium: Empirical Tests // Journal of Political Economy, 1973. Vol. 81(3)
Frazzini A. and Pedersen L.H. Betting Against Beta. // NBER Working Paper. 2010.
Kulachinskaya A., Kravchenko V., Bezdenezhnykh T. Organizational mechanisms of allocation of subsidies for public transport in St. Petersburg // Proceedings of the 31st International Business Information Management Association Conference. – 2018. – С. 4706-4711
Demidenko D.S. Malevskaia-Malevich E.D., Dubolazova Y.A., Victorova N.G. Optimization of the innovation process management at a manufacturing enterprise // Proceedings of the 31st International Business Information Management Association Conference. – 2018. – С. 996-1003.
