Research And Implementation Of Stock Trend Prediction Model Based On Deep Neural Network

Stock trend prediction is crucial for investors to devise effective investment strategies,boost investment returns,and mitigate investment risks.The majority of existing stock trend prediction methods rely on modeling the historical price sequence of individual stocks,treating stocks as independent entities and disregarding the impact of stock correlation on stock prices,thus failing to capture the complete pattern of stock volatility.Although some studies have considered the correlation between stocks,most of them only focus on the correlation between stocks in the same industry or sector,and therefore cannot fully explore the complex correlation among a large number of stocks.Moreover,most current studies treat stock trend prediction as a regression or classification problem.However,they lack optimization with respect to investment objectives,making them suboptimal for guiding stock selection.Aiming at the problems of existing study,this thesis proposes a deep stock clustering algorithm based on auto-encoder and affinity propagation algorithm,and a stock trend ranking model based on spatio-temporal attention network.This thesis considers stock prediction as a ranking problem that is more closely related to real-world goals,fully exploits the potential correlations between stocks,and comprehensively characterizes the volatility patterns of stocks,thereby improving the prediction performance.The main work of this thesis can be divided into the following three parts:(1)A deep stock clustering algorithm based on auto-encoder and affinity propagation algorithm is proposed.By integrating dimensionality reduction and clustering into a deep learning framework,the algorithm performs target feature extraction and data compression are performed on the original stock sequences,and achieves adaptive clustering,thereby fully mining the complex trend correlations between stocks in the same and different industries.(2)A stock trend ranking prediction model based on spatio-temporal is proposed.The model aims to select the stocks with TopK expected returns from the stock pool.By incorporating the correlation between stocks into the prediction model,it can comprehensively describe the stock price volatility patterns from both temporal and spatial dimensions to improve the accuracy of the model.(3)Verify the feasibility and effectiveness of the above scheme and algorithm through experiments.Aiming at more than 3,000 stocks in the A-share market and Kaggle NIFTY 50,this thesis has done a full comparative experiment between the proposed scheme and the existing methods,which proves the superiority of the stock clustering algorithm and the stock trend prediction model proposed in this thesis.Additionaly,by devising stock selection and trading strategies,this thesis conducted backtesting analysis on JoinQuant platform,validating the proposed model’s satisfactory performance in live trading,and its potential to assist investors in making informed decisions to a certain extent.