Stock Quantitative Investment Model Based On SVM And LSTM Network Prediction

Quantitative investment is an investment method that uses statistical methods to predict stock price movements in order to obtain stable returns.In the stock market,due to the large number of stocks and the uncertainty of stock price fluctuations,how to accurately predict the stock price trend to build a portfolio is a problem in quantitative investment.This dissertation combines machine learning methods with portfolios,and uses optimized Support Vector Machines(SVM)and Long Short Term Memory(LSTM)networks to predict stocks.According to the forecasted rising stock,the corresponding quantitative investment model is constructed.The main work and conclusions of this dissertation are as follows:1.When predicting the real-time updated stock time series,the algorithm is required to have a faster iteration speed.To solve this problem,this dissertation proposes a space-shear-variable-step grid search algorithm.The search algorithm is used to find the C and ? parameters in the SVM,so that the SVM can greatly reduce the time for finding parameters and improve the training efficiency of the SVM while ensuring high prediction accuracy.Based on the stock market data,the commonly used quantitative technical indicators are selected as the input variables of the SVM to predict the rise and fall of the stock.2.Stock technical indicators are time series,and there must be a potential relationship between data with short time intervals.This dissertation uses the memory LSTM network to predict the stock price,because the special structure of the node effectively controls the historical information,so that the network can better learn the potential law of training concentration.The value of learning rate is an important part of training LSTM network process.Therefore,the exponential decay method is used to select the learning rate in each model iteration,and the training speed of LSTM network is improved by the dynamic selection of learning rate.3.The mean-variance model of LASSO regression is established.In view of the fact that the traditional mean-variance combination model does not have stability in practical applications,this dissertation uses the LASSO regression mean-variance combination model to establish a portfolio in the optimized SVM and LSTM network predictions for the rising stock set.This combined model not only has the ability to measure investment returns and risks,but also plays a role in screening stocks.Since the expected function of the combined model contains random variables,it is difficult to find a numerical solution that satisfies the condition completely in the solution process.Therefore,the expected function is approximated by the sample mean similarity method,and the selected stocks and investment weights are obtained.Based on the optimized SVM and LSTM networks,this dissertation constructs the GSVM-L and ELSTM-L quantitative investment models respectively,and selects 133 constituent stocks in the CSI 300 Index for back analysis.The comparison results show that the quantitative investment model achieves higher investment returns and has strong anti-risk ability,and fully verifies the feasibility of obtaining excess returns in the stock market.