Modeling and Forecasting of Revenue of Taxes in Fifth Development Plan of Iran Based on a Special Structure of Nonlinear Neural Networks



In this paper modeling and forecasting of revenue of taxes in fifth development plan is investigated based on a special structure of nonlinear neural networks. The time series of taxes which are studied in this research are related to total tax, direct tax, indirect tax, companies’ tax, income tax, wealth tax, and import tax.
Based on the correlation dimension estimation technique, the structure of each time series with respect to linearity, nonlinearity and stochastic process are studied. The results indicate that there is chaotic behavior in tax time series generators and declare possibility of applying nonlinear modeling for mid-run forecast.
Then, the results of modeling and forecasting of time series of the taxes during 1959- 2009 using a novel multi- input multi- output artificial neural networks are presented. An upper and a lower band of prediction are also derived for each time series of taxes. The results for next 6 years prediction are very good in training stage and it is supposed to have good results in real next 6 years.
JEL Classification : G1, G11