An Evaluation of Forecasting Methods and Optimal Combination Models to Predict Tax Revenues

This article predicts tax revenues in various bases (total, corporate tax, income tax, wealth tax and, goods and services tax) for the time period (1390-91). In order to achieve more accurate predictions, firstly the nature of structural time series in respect of linear, nonlinear and stochastic and complexity of the tax system generating time series have been investigated using tests lyapanov exponent and solidarity. Lyapanov exponent results show a weak turbulence in tax time series in which legal persons enjoy the more sensitivity among the other tax sources. The test results also indicate that the system generating time series correlation on legal entities is more complex than the other sources. The second part of this paper is devoted to forecasting tax revenues but due to exchange rate changes and their impacts on Iran's economy, first the paper examines the possible effects arising from the increased tax revenues using VAR models, then VECM models, the models of low-time series like Box- Jenkin and neural network approach with the multi-input, multiple output were used.