Forecast Market Volatility Based On The Circular Neural Network With Baidu Index

In recent years,as artificial intelligence has become a national strategy,the application of artificial intelligence technology in various fields develops rapidly.The prediction of stock price change trend in the financial field has always been a hot issue,and it has great economic and social value to predict the trend of stock price effectively.At present,when artificial intelligence technology is applied in stock price prediction,stock index data and multiple factors are generally selected for the data in the market,and text data and natural language processing technology are used for the analysis and prediction of stock time series for the data outside the market.Based on the circular neural network model,this paper regards the data of Baidu index data analysis platform as the investor sentiment outside the market,that is,environmental variables.Combined with the data in part of the market,it uses the relatively representative daily volatility of CSI 300 index as the target predicted value for prediction.In this work,the theory part of this article mainly introduced the traditional machine learning method and the related theories of neural network model,and the empirical part,first of all,we do feature selection according to the market data of associated with the CSI 300 index volatility and the Baidu index data obtained by using crawler technology based on embedding method(Embedded),eliminating the invalid information outside part of the market,then using the mutual information(MI)measure to choose the best observation window size and standardized solutions,and then in the process of building a LSTM model we use the technique of Dropout to effectively prevent the overfitting problem,and debug of model parameter optimization,then reclassify the training set,verification set and test set,and the results are compared with the prediction results of XGBoost algorithm,a common machine learning method.In this paper,the predicted results of our Long Short-Term Memory model of the CSI 300 index volatility turns out pretty good,the loss function of the optimal solution is in a reasonable scope,and our LSTM neural network model predicted results performed a little better than XGBoost model predicted results,that is to say,the machine learning algorithm in the performance of the stock price forecast problem has certain practical value and development train of thought for the subsequent research work.