Employing bayesian vector autoregression method as an alternative technique for forecasting tax revenue in Zambia.
Date
2024
Authors
Chileshe, Russell
Journal Title
Journal ISSN
Volume Title
Publisher
The University of Zambia
Abstract
Tax revenue forecasting is of critical importance for a government in ensuring adequacy and stability in tax and expenditure policies, as it also contributes to the budget and strategic planning of a country. Henceforth, several tax types need to be projected for the specific fiscal year using models that are statistically sound and with a smaller margin of error. This study explored the Bayesian Vector Auto-Regression (BVAR) method as an alternative technique for forecasting tax revenue in Zambia. The study forecasted Corporate Income Tax (CIT), Personal Income Tax (PIT),Value-Added Tax (VAT) and Total tax revenue (TTR) using Bayesian vector autoregression (BVAR) models with quarterly data from 2010Q1 to 2023Q4 and the results were compared with Autoregressive Moving Averages (ARIMA) and Error, Trend, Seasonal (ETS). Based on the RMSE, the results of BVAR model with the Normal-Wishart prior was the best model for measuring the accuracy forecasting of the CIT, PIT, VAT and TTR. In most cases, ETS is the second best after BVAR and was superior to ARIMA. The results suggest that the BVAR model is best suited to be used to forecast tax revenues in Zambia with the ETS model as an alternative. The study suggests that the BVAR forecasting methods may also be extended to other smaller taxes to investigate whether they will fit these taxes accurately as it does for major taxes.
Key Words: ARIMA,BVAR, ETS, RMSE, Personal income taxes, Corporate Income tax, Economic growth, Gross fixed capital formation, Inflation rate, Stock market performance, value added taxes, Zambia.
Description
Thesis of Master of Science in Statistics.