Research On Spatial Mixed Frequency Data Forecasting Models And Its Application

Spatial mixed frequency data forecasting: the phenomenon that the close relationship of mutual effects between regions(spatial correlation problem)and high-frequency data and low frequency data coexist(mixing data problems),is a new issue under economic forecasting of big data eras and is placed on increasing value.Current research includes two independent directions: spatial prediction and mixed frequency,but less for combining the two and the application.This study attempts to integrate soft spatial weights matrix which is more realistic for description of spatial relationships;the most classic MIDAS provides a new perspective for mixed frequency data;three advantages of the effective support vector machine solution solves a variety of nonlinear issues under multi-attribute and small sample.This study focuses on spatial mixed frequency data prediction model and its application,and proposes spatial mixed frequency data prediction model,and apply the new model to specific spatial mixed frequency data prediction issues.The process crosses and integrates spatial forecasting in spatial econometric,mixed frequency data forecasting,machine learning methods,soft set theory knowledge and so on.The main work can be described as the following three aspects:First,economic variables within a predetermined range of space usually a certain spatial autocorrelation.How to set the spatial correlation structure in spatial mixed data prediction model is the most important problem which builds the whole foundation for univariate and multi-variate in spatial mixed prediction model.After summary and analysis of common spatial weights matrix methods,we found that although in theory it is not able to describe an individual(such as regional)optimal space right spatial relationships or spatial correlation structure of the weight matrix,we can find when describe spatial distance and location,the relationship between the real and the geographical environment by common spatial weights matrix is not consistent.This inconsistency contains both uncertainty,ambiguity and even contradictory of the conclusions.It has an direct impact on the modelling of spatial mixed frequency data prediction.Soft set is an effective tool to deal with uncertainty.It contains three elements: attributes,parameters and mapping.The single spatial weights matrix is a special soft set.Definitions and calculation based on soft set can effectively integrate various popular weight matrix,broaden the traditional spatial weights matrix depending on the "distance" measure,take into account the effect of adjacent areas and non-adjacent areas,mutual effects between boards and radiation effects of centres,and build a soft spatial weight matrix corresponding to the real world.This study then gives main steps of construction method of soft spatial weights matrix based on soft set theory.This study also gives satisfying conditions and test methods for weights matrix according to spatial econometrics theory.Finally,in order to verify the validity of the model,we applied the soft spatial weights matrix influencing factors to the area of industrial agglomeration,and with certain test examples,we show the feasibility of the new method.Second,with the advantage that soft spatial weights is closer to the real economic variables description of the spatial correlation and the experience of processing spatial correlation from general prediction models to the spatial ones,we adopt the most classic mixed frequency model--MIDAS prediction model as a starting point to build a one-dimensional spatial mixed frequency data prediction model.The model is a basic model,and is meaningful for basic integration of MIDAS model and spatial prediction model.This study introduces the modelling idea of MIDAS and analyse the basic principles and setting modes of the model,and summarizes constructing ideas of the spatial mixed frequency model,to prepare for the introduction of space model space weights.We choose MIDAS model as basic prediction model,given frequency inconsistency between the dependent variable and single explanatory variable,and the significantly spatial correlation of explanatory variables,we introduce the soft spatial weights matrix based on soft set theory.Through the soft spatial weights matrix constructed based on soft set to correct polynomial in MIDAS prediction model,which means the coefficient in the new model is given together by mixed data distributed lag weights and soft spatial weights and coefficients.It reflects the setting mode of mixed prediction model with a single explanatory variable with spatial correlation.We build a high-frequency explanatory forecasts a low-frequency variable based on one-dimensional spatial mixed frequency model.Then we analysis deeply the main features of the new model,and proposed errors or accuracy measurements of effectiveness of the model.Finally,apply the new model to GDP forecasts.Among the 30 provinces and cities in China,the forecasting conclusions of quarterly GDP and re-analysis of feature weights,we confirmed the feasibility of model.Third,multi-variate spatial mixed frequency prediction model is existent in predicting areas,and enjoys practical and theoretical value.However,few studies have focused on this problem,and thus our study focuses on building a multi-variate spatial mixed frequency forecasting model based on support vector machine(SVM).First,consider that the prediction model is based on the framework the most classic MIDAS with preliminary improvement and application of one-dimensional spatial MIDAS prediction model is limited.Thus with it as reference,we comprehensively analysis urgent problems of building multivariate predicting model: increased estimated parameters caused by increased dimension of explanatory variables(mixed frequency in explanatory variables);increased dimension of explanatory variables(mixed frequency in explanatory variables)with small sample size problem;nonlinear model set setting issues under coexistence of spatial and mixed frequency.Then,we especially describes the basic principles of support vector machine and its major advantage-kernel extension model multivariate nonlinear small sample and calculation.We replace the weights in prediction model by kernel function,and apply soft weight matrix to represent spatial correlation between different frequencies in variable.We apply least squares support vector regression to construct multivariate spatial mixed frequency forecasting model based on support vector machine.It gives the way of seeking optimal model parameters,and in-depth analyse main features of the new model.It also proposes error or accuracy measurements of effectiveness of the model.Then we apply to predict the regional eco-efficiency in China,through conclusions of eco-efficiency prediction in 30 Chinese provinces and cities and eco-efficiency analysis of regional features,we confirm the feasibility of the model.