Forecasting Tax Revenues Using the Long Short-Term Memory (LSTM) Network

An Applied Research at the General Commission for Taxes

Abstract

This study aims to forecast tax revenues by employing the Long Short-Term Memory (LSTM) network method through building a predictive model capable of improving the accuracy of future tax revenue estimates. Such forecasting is of great importance in supporting financial stability and enhancing the efficiency of public budget preparation, particularly in light of economic fluctuations and the instability of some other sovereign resources. The study adopts descriptive–analytical and quantitative approaches by utilizing tax revenue data issued by the General Commission for Taxes for the period (2003–2025) and employing them to forecast revenues for the years (2026–2030). The MATLAB program was used to build the model by dividing the data into training and testing sets. A set of statistical accuracy indicators was applied to evaluate the model’s performance, including Mean Squared Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), in addition to the correlation coefficient between actual and estimated values. The results indicate that the LSTM model achieved a high level of predictive accuracy, as it recorded the lowest error indicator values along with a very high correlation coefficient, reflecting its efficiency in representing the temporal behavior of tax revenues. Moreover, the future forecasts for the years (2026–2030) using the LSTM method suggest that the tax system is moving toward a phase of relative stability.