Use of Time Series Models on Forecasting of Value Added Tax Revenue in Kenya Revenue Authority
Taxation is one of the means by which governments finance their expenditure by imposing charges on citizens and corporate entities. Tax revenue forecasting plays a central role in annual budget formulation. It provides policy makers and fiscal planners with the data needed to guide borrowing, use accumulated reserves, or specify monetary measures to balance the budget. Therefore, it is necessary for a government to forecast the revenue it collects for planning purposes. Kenya Revenue Authority (KRA), is an agency of the government of Kenya that is responsible for the assessment, collection and accounting for of all revenues that are due to government, in accordance with the laws of Kenya. The main objective of this study was to fit time series models in the data series of revenue collections and establish their effectiveness as far as revenue forecasting is concerned. The study used the monthly Value Added Tax (VAT) collections data from the financial year 2009/2010 to 2015/2016 with the general objective of exploring patterns in the data such as trend, seasonal components, cycles among others and further establish a suitable forecasting model which can be used to predict the amount of VAT revenue to be collected in a certain specified period. The ﬁrst step was to check if the series was stationary by using Dickey Fuller Test, thereafter transformation by diﬀerencing if the series was not stationary. The order of the models was tentatively chosen by analyzing the ACF and PACF plots of the data. This resulted to AR(3) model, MA(1) model, ARMA(3,1) model and ARIMA(3,1,1) model which were fitted to the data series. In order to select the best potential model, diﬀerent statistics were used like BIC, AIC, AICc, and forecast accuracy measures like ME, MAE and MPE. ARIMA(3,1,1) was selected as the best model compared to the other models. Diagnostic check was made to test for correlation and normal distribution of the residuals using Box-Ljung test, Q-Q plots and Shapiro wilk test and the results showed normal distribution of the residuals with no correlations for all the models. Using the models, forecast of VAT revenue collection in the financial year 2016/2017 was made and the forecasted values compared with the revenue collections that were made in the respective financial year. ARIMA (3,1,1) produced better forecasted values as compared to the other models. Therefore ARIMA (3,1,1) was chosen as the best effective model to fit the data series and forecast the VAT revenue collections.