| Liquor,as one of Chinese unique industries,not only has a long history and unique national cultural connotation,but also has made great contribution to the development of national economy.The study to the factors affecting liquor sales volume can help us grasp the liquor market trend and adjust the sales strategy in time better.Firstly,in analyzing the factors affecting liquor sales volume,the repeated and cross measurements of the data led to a high correlation between the independent variables,so that the random error could not satisfy a set of hypothesis including zero mean and variance,and the traditional methods of estimation are out of reach.In order to solve this problem,this thesis introduces the Generalized Moments Estimation Method to estimate the parameters of the model,which makes the model closer to the actual application,and adds auxiliary information to further mine the implied information of the data.Secondly,in the case of missing data,this paper assumes that the missing data can not be ignored.By assuming that the default mechanism model is a Logistic regression model,the propensity score function is obtained using the Generalized method of moments method introduced in this paper.The inverse probability weighted method is used to deal with the missing data,and the weighted estimation equation of the non-linear model under the non-negligible missing data is established,and the parameters of the non-linear model are estimated by the Generalized method of moments method.Finally,the methods about variable selection are introduced to screen the factors affecting liquor sales volume.In order to combine with the Generalized method of moments method,three kinds of variable selection methods based on penalty function from plentiful mature methods are selected: Lasso,Alasso and Scad.When the penalty function is added to the weighted estimation equation in the previous step,the selection of variables is realized while the parameter is estimated by the Generalized method of moments method.The combination of the above methods is verified to be robust and effective by a large number of numerical simulations.Finally,combined with the results of part of machine learning method,this method was applied to analyze the influencing factors of liquor sales volume,which found that the three factors were GDP,transportation value and beer sales volume.The conclusion makes economic sense. |