Forecasting Tax Revenue Using Arima and Vector Autoregressiive (VAR) Modelling in Tanzania

This article intends to examine whether times series approaches of ARIMA and VAR are effective in forecasting tax revenue. It also compares the two approaches to evaluate which is the more effective forecasting method. Quarterly data from 1996Q1 to 2016Q4 (21 years or 84 observations) are used to forecast the tax revenue for the period 2017Q1 to 2017Q4. Five common types of taxes are selected due to their significant contributions to Tanzania’s total tax revenue collected by the Tanzania Revenue Authority (TRA). Generally, the results reveal that both time series approaches are effective and demonstrate strong predicting power in short-horizon tax revenue forecasting. However, in most cases that the VAR model outperforms ARIMA modelling, especially based on forecasting criteria. However, we suggest that both methods to be applied by the TRA in forecasting tax revenue as their forecasting errors differ only slightly.