Research On Stock Index Prediction Based On Combined Model

In recent years,a variety of time series prediction models have been widely used in the economic and social fields,and have played an important role in different fields,providing valuable reference information for the future development of various industries in society.In the stock market,most of the time series of stock indexes can be used to build prediction models through statistical methods.These models can infer and predict the future changes that may occur in the stock market,because this method can avoid the complicated data acquisition process.,And can reduce the interference of redundant information,so it has been favored by many scholars.At the same time,stock index forecasting has a positive impact on helping investors to form investment tendencies and construct investment strategies.The forecasting accuracy of the model will directly affect investors' earnings expectations and risk assessment of stocks.However,the traditional economic forecasting model has its own limitations due to the single forecasting model,which leads to the poor accuracy of these single models in the actual stock index forecast.Therefore,it is of practical value to construct a stock index prediction model with high prediction accuracy and easy operation.In order to overcome the shortcomings of the traditional economic forecasting model,based on the existing research at home and abroad,this paper proposes a combined model that includes data preprocessing technology,multiple neural network sub-models and multi-objective optimization algorithms for combined weighting In the empirical analysis,the stock index is used as historical data to make a one-step,two-step and three-step forecast on the future stock index.Because time series data in the financial field is usually very complex,generally with randomness,nonlinearity,etc.,in the combination model proposed in this paper,the stock index data is first preprocessed,and empirical modal decomposition algorithms and collections are used respectively.Four methods of empirical mode decomposition algorithm,singular spectrum analysis,and variational mode decomposition algorithm are used to process the data.By comparing the error of the prediction results,the singular spectrum analysis is finally selected as the pretreatment method of the combined prediction model in this paper;Four models commonly used in stock index prediction,including differential integrated moving average autoregressive model,feedforward neural network,gate valve unit neural network,and long-term short-term memory neural network,are used as sub-models of combined prediction models;The target locust optimization algorithm determines the weight parameters of each sub-model in the combined prediction model,thereby constructing a complete combined model.The data selected in the empirical analysis is an important barometer that can represent the Chinese stock market,the US stock market,and even the global economy,that is,the Shanghai stock index of the Chinese stock market,the Nasdaq composite index and the Dow Jones index of the US stock market.The empirical research mainly draws the following conclusions:?By comparing the prediction results,it can be found that the combination model proposed in this article is the best performing of all the models in the experiment,with the highest prediction accuracy and the most stable prediction performance,so the prediction results of the model Can guide investors to invest,thereby significantly reducing investment risks;?After the sub-model after data preprocessing,the prediction accuracy in stock index prediction has been improved to varying degrees,including singular spectrum analysis based on decomposition and reconstruction The technology obtains the basic characteristics of the time series by removing high-frequency signals.Compared with other data preprocessing methods,the best prediction results are obtained in one-step,two-step and three-step prediction;?Optimized by the multi-objective locust optimization algorithm The weight coefficient of each sub-model in the combined model is obtained to obtain a combined model with high accuracy and high stability.To be sure,the application of this model will help investors manage their funds and make reasonable investment decisions.The innovation of the article mainly includes the following two aspects:First,the application of decomposition and reconstruction strategies,the use of data preprocessing methods,the main features of the original data are extracted by eliminating the influence of high-frequency signals,so that the prediction is more accurate.Decompose the original data and reconstruct it,so that the irregularities and uncertainties in the original data can be eliminated,and a better prediction effect can be achieved.Second,in the process of combining individual models,the multi-objective locust optimization algorithm is applied.The sub-models in the combined model are given optimal weighting coefficients,and a good prediction effect is achieved.The shortcomings of the article are mainly reflected in two aspects:first,the test data of the model in this article are all from the stock index data and do not use the data of a specific stock,so the model is versatile and portable The aspect is still to be tested.Second,this article only analyzes the time series of stock index historical data,and does not conduct in-depth discussions.In the future research,various economic indicators and other factors that affect stock prices should be added for correlation analysis.