The Estimation Of Tax Capacity Based On Non-Parametric Method Of Mixed Data

There is a pervasive problem about the low efficiency of public sector. The tax revenue of China is increasing at an incremental growth rate, however is it an evidence of the high efficient performance of tax bureau? The research of estimation of the tax capacity and the tax collection efficiency is important for this problem.This paper attempts to estimate the tax capacity of Zhejiang province by statistics method. However, how to select variables in the model? Past researches of this problem use quality method which may lead to irrationality to selected variables, and may not choose variables for different districts objectively. Moreover, linear models in the past researches consider linear errors only.This paper applies non-parametric data-driven method which eliminates the variables that have no significant impact on tax capability in case of Zhejiang province to solve the problem mentioned above. Another advantage of this approach is that it is possible to build the tax capacity estimation model with both continuous and categorical data.A Linear model is also built by stepwise least squares regression in this paper. The result comes out that the goodness-of-fit of non-parametric model is much better than that of linear model through the mean square error (MSE). Stepwise regression considers the significant linear parametric which is the standard of choosing variables. By contrast, the non-parametric model has the advantage to can get rid of this problem.The result comes out that the relevant variables in non-parametric model are GDP per person, openness, total retail sales of consumer goods, and the percentage of industries which is different from the result of linear model whose relevant variables are GDP per person, openness, fixed investments, urbanization and the percentage of industries. This paper build the stochastic frontier analyze (FSA) model including relevant variables in non-parametric method to get the efficiency of tax capacity of every districts in Zhejiang province.The conclusion is that the fitness of non-parametric model is much better; moreover the robustness of this model is higher according to the evidence that the change of mean square error is only6.25%of that of the linear model after eliminating the irrelevant variables. The tax capacity model based on this improved variable selecting method is more reasonable.