Stock Index Combination Forecasting Models And Empirical Research Based On Incomplete Mixed Frequency Data

Stock index forecasting is widely considered among researchers.In forecasting Stock index or other economic issues,we usually face the following situations:(1)the data used to predict economic problems often exist different statistical time and frequencies,that is,mixing frequency data.For example,the stock price prediction in the economic field involves high frequency indexes such as price,volume and so on.It also involves low frequency indicators such as macroeconomic data,financial data,and industry policy;(2)for various reasons,there is often a lack of data for forecasting;(3)there is a complex nonlinear relationship between explanatory variables and explanatory variables.Traditional forecasting approaches are only able to address data with same frequency,which is limited in practicle application.Mixed frequency models offer new insights into mixed frequency data issues,and achieve a lot.However,there is some improvements:(1)the mixed data issues all apply complete data,and do not involve missing data;(2)the mixed data still exist limitations though they are widely used;(3)complex nonlinear relationship of dependent variables and independent variables is not fully explored.In this paper,we try to integrate the mixed data forecasting,grey relational analysis and the advanced genetic algorithm,so as to solve the complex nonlinear optimization problems involving the missing data in the economic domain.The main work are as follows:First,this paper will construct a univariate mixed frequency combination forecasting model based on genetic algorithm and apply to stock market index prediction.In the previous researches in mixed data frequency approaches,MIDAS approaches is the most common model at present,the basic MIDAS model is mainly applicable to the forecasting volatility and macroeconomic state in financial market.U-MIDAS model gain better results in the prediction of many practical problems,but it requires smaller differences in frequencies.Thus,the two kinds of models in practical application are subject to greater restrictions.Because many economic problems need to be predicted by a large number of mixed frequency data,the relationship between explanatory variables and explained variables is very complex,and often exists nonlinear relationship.In order to improve the prediction performance and application range of the mixed frequency data model,this paper will construct a univariate mixed frequency combination forecasting model.When the two basic models are combined to form a new model,the relationship complexity will further improve.Genetic algorithm is a kind of search algorithm based on biological evolution process and gradually approximates the optimal solution by continuous iteration.It has unique advantages in solving the nonlinear optimization problems with high complexity.Its search capability is not limited by specific problem models,and it can effectively search for large scale space.Therefore,it has special advantages in dealing with complex optimization problems of high-dimensional and nonlinear relations.Compared with other optimization algorithms,the genetic algorithm is more likely to obtain the global optimization solution.This paper will combine mixed frequency combination forecasting model and genetic algorithm.We firstly established a forecasting model of univariate mixed frequency combination forecasting model based on genetic algorithm.We also set the parameter setting of the genetic algorithm,the algorithm optimization method and the coefficient and weight estimation of the mixed data frequency forecasting model obtained by the genetic algorithm.At the same time,we design the parameter setting of genetic algorithm and the search method of model coefficients and weights.Finally,the model is applied to Chinese stock market prediction of Shanghai Composite Index,Shenzhen Component Index and the Growth Enterprise Index(GEI)index.We compare the performances with those of other models.The results show that the proposed model performs better.Second,this paper will construct a univariate missing mixed frequency combination forecasting model based on genetic algorithm and apply to stock market index prediction.In many practical problems,due to a variety of reasons,missing data is common.It is necessary to use the missing mixed data to directly predict the economy.In traditional,missing values are usually filled by weighted mean or probability mean.The weighted mean method is only suitable for smaller sample size of missing values,while the probability mean method is easily affected by the extreme data.At the same time,the field of mixed data frequency has not involved in cases of missing data.To address the above issues,this paper will construct a univariate missing combination mixed frequency forecasting model based on genetic algorithm.On one hand,we will focus on the influence of a single important factor on the results or the relationship between them when we forecast economic issues.On the other hand,the univariate missing combination mixed frequency forecasting model is the foundation of establishing the missing mixed data prediction model,so we first set up the univariate model.The objective of the grey relational analysis model is to identify the geometric relationship of the two sets of data sequences in the spatial relationship.The model can be used to quantify the similarity between the comparison sequence and the reference sequence.In this paper,we integrated the grey relational analysis model and univariate combination mixed frequency forecasting model to deal with missing data.Then we apply the genetic algorithm to search the global optimal solution.This model is applied to the China stock market to forecast Shanghai Composite Index,Shenzhen Component Index,the GEM index in use of missing mixed frequency data.The forecasting results show that the proposed model is effective.Third,this paper will construct a multivariate missing combination mixed frequency forecasting model based on genetic algorithm,and apply to stock market index prediction.In the actual economic forecasting problems,there are many factors that affect the prediction results,and all kinds of factors interact with each other.Therefore,univariate model is limited in many cases,and this paper focuses on building multivariate model.Based on the univariate model,to construct a multivariate missing mixed combination forecasting model,this paper comprehensively analyses the problems to be solved in building multivariate missing combination mixed frequency forecasting model based on genetic algorithm:(1)the dimension of independent variables increases,and the parameters to be estimated increase correspondingly;(2)there are mixed data between multiple independent variables.When dimension increases,there is a complex nonlinear relationship among independent variables,as well as between independent variables and dependent variables;(3)when the independent variables increase,the possibility of multiple missing values will increase.It is a problem that we need to establish forecasting models using mixed data containing multiple missing values.To address the above issues,this paper will construct a multivariate missing combination mixed frequency forecasting model based on genetic algorithm.After expanding the univariate model into multivariate models,we will face more complex problems such as the complexity of data relations,the huge increase of estimated coefficients and vectors,and the sharp increase of computing complexity.This paper will apply genetic algorithm to solve process and algorithm problems.Finally,we apply the proposed model to Chinese stock market to forecast Shanghai Composite Index,Shenzhen Component Index and the GEM index.Then we compare the performances with those of contrast models.The results show the validity of the model.