Research On Stock Price Prediction Based On LASSO And BP Neural Network

This paper combines the LASSO(Least Absolute Shrinkage and Selection Operator)regression model commonly used in statistics with the neural network model popular in deep learning.The LASSO regression model was used to filter several features of the data,and the BP(Back Propagation)algorithm in neural network was used to build the model.Take Kweichow Moutai as an example to fit and forecast the stock price data,and model the panel data composed of time series data and cross section data that affect the stock price.Combined with the actual analysis of the impact of various factors on stock prices,LASSO regression is used to screen the financial data of the company,and then the BP neural network model is established based on LASSO and BP neural network.At the same time,this paper explores some common problems in deep learning,such as parameter initialization,gradient disappearance and gradient explosion.In addition,it employs Recurrent Neural Network(RNN),which is better in processing time series data,to improve the Neural Network model structure and compare its performance.The results show that whether the LASSO regression model is used alone or the BP neural network model is used alone,its mean square error on the test set is higher than that based on LASSO and BP neural network model.The superiority of LASSO and BP neural network model in data prediction is proved.After the cyclic neural network is used to optimize the model structure,it is concluded that the optimized model has better fitting effect in the test set,and the unoptimized neural network model has better prediction effect in the test set.At present,researches on stock price prediction are based on semi-strong efficient market hypothesis or strong Efficient market hypothesis,which generally believe that both technical analysis and fundamental analysis of stock are invalid,that is,stock price cannot be accurately predicted.In this paper,stock price prediction is carried out based on invalid market hypothesis.This paper provides a new case with important reference value to the case base of comparative study.At the same time,some improvements are made on the basis of the original research results,and better prediction results are obtained by combining the two methods effectively.The calculations in this paper are based on Tensorflow and Sk Learn libraries in Python and other statistical analysis software.