Research On Stock Index Return Prediction Based On LSTM Deep Learning Model

Stocks are long-term credit instruments in the capital market with high returns and high risks,and are a popular form of financial management among investors.Stock indices are reference indices that reflect stock market movements and give an indication of market tendencies.Since the birth of stocks,the problem of stock market time series forecasting has been a hot research topic for investors and scholars.Traditional statistical models are widely used in stock market time series data forecasting problems,but due to the non-linear characteristics of stock time-series data,traditional statistical models have certain shortcomings in such forecasting research problems.With the development of science and technology,the advantages of forecasting research in stocks through methods such as statistical mathematical modeling and computer models are becoming more and more obvious.Machine learning models can effectively solve the nonlinear problem of time series data,and with the development of machine learning models,deep learning models such as artificial neural networks have started to be widely used in stock market time series data forecasting.Deep learning models involve artificial neural networks with nonlinear,non-limiting,non-constant,and non-convex characteristics,which have greater advantages in dealing with time-series data forecasting in financial markets.The multilayer long short-term memory model(LSTM)used in this paper can effectively solve the gradient disappearance and gradient explosion problems in the long series training process,and has advantages in the long series forecasting problem.The prediction of stock index returns is a formal long series nonlinear problem,and the framework of the multilayer long and short-term memory model(LSTM)can be used to predict index returns better.In this paper,the real historical data of the SSE Composite Index,SSE Composite A-share Index,SSE Composite B-share Index,and SZSE Composite Index are taken as research samples,and the raw time series data related to index returns are collected for two years since August 2,2018,to August 2,2021,and based on fully analyzing the correlation and influencing factors of stock index returns,we build a framework based on sliding window deep learning model LSTM framework,which combines the advantages of multi-layer long and short-term memory LSTM and sliding window data enhancement effect to improve the accuracy of the framework.This paper selects ARMA and decision tree models as comparison models to conduct comparative prediction experiments to demonstrate the advantages of the sliding window-based deep learning model LSTM framework for stock index return prediction.The results of this paper show that,first,the multilayer long and short-term memory model(LSTM)is more advantageous than statistical and machine learning models in the problem of predicting stock index returns.Second,the sliding window algorithm can play the effect of data enhancement in the experiment to improve the prediction ability of the model.Third,the new combined framework based on the sliding window algorithm and long short-term memory model(LSTM)innovatively built in this paper has a better prediction effect.The portfolio model of LSTM optimized and improved in this paper has obvious advantages in stock index return prediction.For this reason,this paper makes the following two suggestions: First,scholars who study-related content can choose the LSTM framework based on the sliding window deep learning model for experimental research.Second,investors in the stock market can apply the framework to predict stock index returns and make corresponding investment asset allocations with reference to the prediction results.