Tax Revenue Prediction Model And Applied Research Based On Statistical Learning Method

As an important economic indicator affecting the livelihood of the country,the growth of tax revenue reflects the country’s economic development and the effectiveness of the government’s administrative capacity.The accuracy of tax revenue forecasting affects the effectiveness of government decision-making,so effective forecasting of tax revenue can help the government make effective decisions,thus promoting the country’s economic development and high-quality economic development.After years of research and exploration by scholars at home and abroad,the methods for tax forecasting mainly include time series forecasting models,econometric regression forecasting models and intelligent algorithm forecasting models,etc.Each model has its own scenario,scope and limitations of application.In this paper,on the basis of a single model,the construction of a combined model can not only expand the scope of application of the model,effectively offset the limitations of a single model,but also has a high prediction accuracy.In this paper,firstly,multi-dimensional and multifaceted influencing factors indicators are selected according to the four major tax types that make up tax revenue,and the correlation analysis and stepwise regression method are used for dimensionality reduction screening;secondly,the principal component regression model,PSO-SVM model and Elman neural network model are used to forecast and analyse tax revenue respectively,and the prediction effect of each single model is compared with ARIMA model to illustrate the prediction effect of each single model;then a combined forecasting model was constructed by weighting each single model according to the principle of minimising the variance of the single model forecasting error.Finally,on the basis of the GM(1,1)model for future forecasting of each influencing factor,it was substituted into the combined model to obtain the forecast value of tax revenue for the next ten years,and to realise the analysis of the forecasting trend for the next few years.The empirical results show that: firstly,after the screening of factors,the factors that have a greater impact on tax revenue are seven indicators such as industrial value added,which are different from the influencing factors such as GDP included in previous studies.Secondly,the results of the principal component regression model show that total population and energy consumption have a suppressive effect on tax revenue,while the rest of the factors have a catalytic effect on tax revenue,with industrial value added having the greatest effect and tertiary value added having the least effect.Thirdly,the accuracy of each single model is in the order of PSO-SVM model,Elman neural network model and principal component regression model from good to bad.Fourth,based on the principle of minimising the variance of the forecast error,the combined model has a higher forecast accuracy than each single model,indicating that averaging the models with appropriate methods can effectively reduce the forecast error and improve the forecast accuracy.