How long is a long run? Tax Revenue Forecasting
A case study of Tanzania
This paper intends to examine whether using long run sample size has more forecasting power than short run sample size. The sample size ranges from 1996 to 2016 and 2000 to 2015. Ordinary Least Square (OLS) method was used to forecast three components of tax revenues including total revenue (TR), Pay As You Earn (PAYE) and Value-added Tax (VAT). The results show that, both TR and PAYE forecasts are slightly better when using short run sample size.
However, for VAT, forecasting power is slightly better when using long run sample period. This reveals that, in contrast to other fields, forecasting tax revenue using the short run sample size data could be more useful. We believe that, the long run period is subjective and field oriented. Also, the nature of the tax can have different implications in selection of sample size and data frequency.