Ahmad Ezzati shourgholi *1 
, Farzad Mohammadi2 , Habib Torbati3 , Farjad Bakhshor4 , Somaye Ghasemi5
, Farzad Mohammadi2 , Habib Torbati3 , Farjad Bakhshor4 , Somaye Ghasemi5
1- Assistant Professor, Afagh Higher Education Institute, Urmia, Iran , ahmetezzati@afagh.ac.ir
2- PhD Student in Financial Engineering, Department of Accounting and Finance, Ta.C., Islamic Azad University, Tabriz, Iran
3- M.A. in Economics
4- Assistant Professor, Afagh Higher Education Institute, Urmia, Iran
5- M.A. in Financial Engineering and Risk Management, Afagh Higher Education Institute, Urmia, Iran
2- PhD Student in Financial Engineering, Department of Accounting and Finance, Ta.C., Islamic Azad University, Tabriz, Iran
3- M.A. in Economics
4- Assistant Professor, Afagh Higher Education Institute, Urmia, Iran
5- M.A. in Financial Engineering and Risk Management, Afagh Higher Education Institute, Urmia, Iran
Abstract: (11 Views)
This This study aims to evaluate the forecasting accuracy of advanced models in predicting both the long-run trend and the stochastic component of tax revenues in West Azerbaijan Province. Given the crucial role of tax revenues in fiscal planning and macroeconomic decision-making, obtaining reliable and precise forecasts is of particular importance. The analysis is based on monthly tax revenue data for West Azerbaijan covering the period from April 2005 to December 2023. To forecast the long-term and stochastic dynamics, three advanced approaches are employed: the Markov-Switching model, the State-Space model, and the Wavelet–Artificial Neural Network (Wavelet–ANN) model. In the first stage, the stochastic volatility framework is applied to decompose the series into a long-run trend and short-term stochastic fluctuations. Compared to competing models—especially those based on conditional heteroskedasticity—the stochastic volatility approach provides a more efficient identification and modeling of random shocks. The empirical results reveal that the hybrid Wavelet–ANN model, through multi-scale decomposition of data and recognition of nonlinear patterns in the time series, delivers the highest forecasting accuracy for both components: a mean absolute percentage error (MAPE) of 1.08% for the long-run trend and 0.27% for the stochastic component. Moreover, when compared to the conventional Markov-Switching and State-Space models, the proposed hybrid model shows clear superiority in reproducing historical fluctuations and capturing the underlying dynamics of tax revenues more precisely.
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
