Model Selection Based On Unbiased Estimating Equation

Model selection research is an important economic component of metrology. Since Akaike's AIC criterion has been proposed, based on information theory model selection criteria developed very fast. This article is based on previous research, unbiased estimation equation according to the basic theory of constructing a new model selection criteria, which compared with other estimates based on maximum likelihood estimation and probabilistic model of the whole in terms of a broader range of applications.This paper reviews the development process of model selection criteria, the model selection criteria for the basic classification, based on hypothesis testing model selection criteria is the most basic test methods, standards are often used in conjunction with other models, and the basic principles have been applied to other model selection criteria; and based on the mean square error, information theory, cross-validation and other methods of model selection criteria will be for different purposes, model selection under different conditions. Under different types of models will take a different model selection methods, the most commonly used is based on the most basic information theory model under the selection criteria - AIC criterion. The basic reasoning, estimation model selection criteria so as to lay a foundation for further promotion.Unbiased estimating equations based on the model selection criteria for its main feature is an unbiased estimating equation as wide application, the process of making it in the calculation of the conditions required to meet fewer restrictions, which make the selection criteria with a strong range of applications. This model is constructed candidate selection criteria need to build the model calculated the cost of information function (the Cost information function), by calculating the function values to build CIC criteria (Cost information criteria). According to the model's parsimony, CIC guidelines calculation has two parts, the first part of the equation is based on unbiased estimates calculated from the Wald statistic, the second part of the model complexity penalty function. The CIC is built meet the general criteria for model selection criteria and model goodness of fit and complexity of the basic form. After further research, CIC model selection criteria to meet the requirements of robustness, but also has consistency.Application of this part of the application through two different models to verify the ability of the practical application of CIC criterion, the empirical process, this applies not only to the CIC criterion, also calculated the classical model selection criteria other comparison, proven effective CIC criterion and accuracy. First, the linear model applied to small samples, and its simple structure, is a classic econometrics book an example. Second, the CIC criterion is applied to the VAR model among a large sample, model complex structure, the new evidence in the case. By comparing two different models found, CIC has a strong ability to apply criteria, in different situations were able to choose the best model.Overall, the proposed unbiased estimation equation based on the model to the selection criteria - CIC criteria: computation is simple, wide range of applications, choose the high degree of accuracy, which can be applied in the empirical research which.