{"title":"Stock price prediction under multi-frequency model - based on attention state-frequency memory network","authors":"Wei Zhou, Yuting Pan, Zhaoxia Wu","doi":"10.1145/3548636.3548654","DOIUrl":null,"url":null,"abstract":"Stock price forecasting is the most typical forecasting task in financial time series. However, stock prices are influenced by a variety of factors, with short-term and long-term trading activity being among the most important factors affecting price changes. The trading activity in these different frequency patterns is eventually reflected in the stock price data. Therefore, if the model identifies information about the potential multiple frequencies of the time series data in the process of prediction, it can enable the model to better learn the features of the time series data. In addition, stock price data are non-linear and non-stationary in nature, which makes it a challenging task to predict future trends. We used the attention state-frequency memory neural network (A-SFM) to accomplish the above objectives. The A-SFM realizes the decomposition of frequency by Fourier transform, and learns the state and frequency information by the powerful learning ability of neural network. In addition, we add an attention mechanism to the structure of the model to extract and learn the important parts of the price information. In this paper, we use the A-SFM model to capture the state and frequency information inherent in the series from past financial market data and use this information to make short- and long-term forecasts.","PeriodicalId":384376,"journal":{"name":"Proceedings of the 4th International Conference on Information Technology and Computer Communications","volume":"1 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-06-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings of the 4th International Conference on Information Technology and Computer Communications","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/3548636.3548654","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1
Abstract
Stock price forecasting is the most typical forecasting task in financial time series. However, stock prices are influenced by a variety of factors, with short-term and long-term trading activity being among the most important factors affecting price changes. The trading activity in these different frequency patterns is eventually reflected in the stock price data. Therefore, if the model identifies information about the potential multiple frequencies of the time series data in the process of prediction, it can enable the model to better learn the features of the time series data. In addition, stock price data are non-linear and non-stationary in nature, which makes it a challenging task to predict future trends. We used the attention state-frequency memory neural network (A-SFM) to accomplish the above objectives. The A-SFM realizes the decomposition of frequency by Fourier transform, and learns the state and frequency information by the powerful learning ability of neural network. In addition, we add an attention mechanism to the structure of the model to extract and learn the important parts of the price information. In this paper, we use the A-SFM model to capture the state and frequency information inherent in the series from past financial market data and use this information to make short- and long-term forecasts.