A Personal Income Tax Prediction Model Based On Factor Analysis And NARX Fusion

As one of the key sources of national revenue,individual income tax can not only guarantee the stable national revenue,but also can play the role of income redistribution and narrowing the gap between rich and poor.Area A is a district of Zhongqing,its economic level and tax level are in the middle and low level of Chongqing,the research on the individual income tax in a area helps to provide the budget for the local Taxation Bureau,and provides a reference for the government to adjust the economic structure.At the same time,a district tax forecast model can be extended to Zhongqing other counties,and even the entire Zhongqing personal income tax forecast.However,the research of tax forecast is mainly divided into two directions of traditional time series prediction and regression prediction,both of which have some drawbacks.The time series analysis and prediction method does not consider the influence of external concrete factors,and the prediction error is large.Regression forecasts usually bring GDP,economic indicators,such as GNP,have neglected the influence of historical tax on current tax revenue,and the influence factors such as GDP and GNP need to be predicted by time series,and the error propagation is caused by the iteration of two models.In this paper,the combination of factor analysis and Narx dynamic neural network is applied to the tax prediction of a region.The goal is to explore a new predictive model that can effectively predict individual income tax.The input of the Narx neural network is the lag of the external data and the target data of the lag,and the output is the current target data.Therefore,first of all,this paper calculates the score of 12 tax points of individual income tax in a district,and then splicing the factor score with lag of 4-order and the personal income tax lag 4,as the input characteristic of the Narx neural network,the output is the current personal income tax.The lag factor 4 is obtained by constant adjustment,and the prediction error is minimal when the lag factor is 4 o'clock.This will take into account the impact of various tax items on individual income tax,as well as the impact of historical taxes.The experimental results show that the NARX prediction model with integrated factor score is lower than the traditional Arima prediction model,the average error absolute value of Narx model is 2.6%,and the average error absolute value of Arima model is 13.3%.From the comparison between Narx forecast model and Arima forecast model,we can see that the NARX model has better fitting effect to non-linear data.Since the income tax data from 2006 to 2016 is not linearly increasing in area A,the Narx prediction model can fit this fluctuation well,while the Arima model is less effective in the face of this fluctuation.