خلاصة:
Tax is one of the main sources of financing government budget. Therefore, having a clear
picture about the attainable amount of taxes are not only necessary for optimal allocation
of scarce resources for tax collection, but also helps the government to develop precise
tax collection programs .In this article, the structural features of the tax revenues series
have first been examined in relation to linearity, chaotic nonlinearity and stochasticity,
using Lyapunov Exponent. These series are: total taxes, direct taxes, indirect taxes,
corporate taxes, income taxes, salary taxes, real estates taxes, business taxes, wealth
taxes, inheritance taxes and goods & services taxes. The results indicated the existence of
a chaos in the series of different tax resources with different weakness and severity.
Therefore, based on the results it was found that we can do more accurate short-term
predictions by applying nonlinear modeling. In the next step, using the data of the period
1963-2006,the tax revenues of different resources were forecasted for the period 2007-
2009 by applying both parallel and proposed Multiple-input Multiple-output structures of
the ANN’s.
ملخص الجهاز:
"For instance the value of for the corporate tax is positive but small for dimensions 2 and 3; therefore, the time series in question has relatively a weak chaos and nonlinear modeling can result in a relatively good short-run forecasts.
x 104 Neural Networks Estimation 9 Training set Testing set Predicted set Actual Tax 4 3 1 0 50 55 60 65 70 75 80 85 90 year 1 Corporate tax base= GDP-(value added of business and agricultural sectors) Figure 11.
8 Forecasting the Real Estate Tax by Using Artificial Neural Networks (parallel model) x 105 Neural Networks Estimation 14 Training set Testing set 12 Predicted set Actual Tax 10 8 4 2 50 55 60 65 70 75 80 85 90 year Figure 12.
9 Forecasting the Inheritance Tax by Using Artificial Neural Networks (parallel model) x 105 Neural Networks Estimation 16 Training set Testing set 14 Predicted set Actual Tax 12 10 8 4 2 50 55 60 65 70 75 80 85 90 year Figure (12)-Out-of- sample forecasting of the real estate tax in the period 2007-2009 4.
10 Forecasting the Wealth Tax by Using Multiple Input-output Neural Networks (proposed structure) The result of Lyapunov exponent test shows a relatively high chaos in the wealth and consumption tax series in comparison to the other taxes; therefore, proposed multiple input-output models is used to get a more accurate forecast of these tax revenues.
We have used the proposed neural network to forecast the consumption tax due to the existence of relatively high rate of chaos in the time series of the consumption tax according to Lyapunov exponent test."