A Stock Rank Prediction Method Combining Industry Attributes And Price Data Of Stocks

China’s stock market has developed rapidly since the reform and opening up.As an important means for corporate financing,the stock market has attracted a large number of investors.By the end of 2022,the number of A-share investors had reached 212 million.In order to obtain ideal returns and effectively avoid risks in stock trading,investors always hope to accurately predict the future price or trend of the stocks,and then make reasonable investment decisions,so stock prediction is a research of great significance.However,stock price changes constitute a dynamic,nonlinear,and time-varying complex system due to the existence of multiple uncertainties such as economic conditions,industry-specific variables,company prospects,investor psychology,and macro policies.This makes stock predicting a very challenging task.Among many time series prediction models,the Temporal Convolutional Network(TCN)has outperformed the recurrent neural network structure model on a variety of time series due to its good historical data receptive field.However,for stock price time series with autocorrelation,the structure of dilated causal convolution makes it difficult to effectively learn the dependence between different time data within the series.Additionally,for the multi-nature series of stock price,TCN cannot dynamically distinguish the importance of different nature series in the prediction,which limits the processing effectiveness of the model for stock price series.In addition,due to the existence of "Industry Rotation",some industries with specific attributes may become market preference for a period of time,and the stock prices belonging to these industries will have a greater probability of rising.Therefore,extracting the industry attributes of stocks and using them for prediction will help improve the accuracy of prediction.To solve the above problems,this paper first designed a channel-time dual attention module(CTAM).In conjunction with TCN,CTAM can adaptively learn the importance of multiple price nature series of stocks and model the dependencies between data at different times,so as to extract stock price features more fully.In addition,in order to mine and utilize the industry attributes of stocks,this paper design an industry-stock Pearson correlation matrix,and extract vectors that fully represent the industry attributes of stocks by matrix factorization algorithm.Furthermore,historical market preference was modeled according to the industry attributes of stocks to generate the dynamic correlation between stocks and market preference.Then the correlation is combined with the historical price features extracted by TCN to predict the stock ranking.The model proposed in this paper uses the newly designed CTAM combined with TCN to extract the features of stock price series,which is more adequate for the processing of multiattribute stock price series.In addition,due to the existence of "Industry Rotation" in the Ashare market,this paper mines the industry attributes of stocks from the correlation between stocks and industries as the new feature,and uses the feature in the form of modeling market preference for prediction,which provides a new idea for stock prediction.