ABSTRACT

Every year, forecasting errors occur in Pakistan’s government tax revenues. From 1970 to 2020, the federal government made massive projection errors across all tax categories. Using data from 1970 to 2014, two-thirds of the time, overall federal tax collections were overstated by more than 5 percent (Qasim & Khalid, 2016). The consequences of inaccurate tax revenue projections could be deceptive tax revenue targets, the unfair distribution of resources, and an increase in the debt burden. The reasons for tax revenue forecasting errors can be administrative, political, or methodological. Unfortunately, the political and organizational issues can’t be measured accurately due to the qualitative and dispersed nature of the problems.

A significant part of these errors occurs due to wrong choices of forecasting methods, data discrepancy, and parameter issues. This study digs out the said problems in detail. The first objective is to document the current tax revenue forecasting mechanism and its impact on the economy. Forecasting errors are transferred into the economy through budget estimates. The study highlights the impacts specifically in terms of development expenditure, GDP, and debt accumulation.

The second objective is the decomposition of forecasting errors in terms of data, parameters,  and  estimation  methods.  FBR’s  default  forecasting  method  is  the Buoyancy approach. This method uses buoyancy estimates of tax revenues along with tax bases to forecast future tax revenue. FBR uses provisional data for forecasting purposes. This research has used different combinations of provisional, revised, final, and real-time data to check the effect of data discrepancies. The results show that the data regime has no significant impact on improving the forecasts.

The Buoyancy method uses the nominal GDP targets provided by the Ministry of Finance. The second part of the second objective is to check the effectiveness of the GDP targets revision. In different scenarios, the thesis has used provisional, revised and real-time targets of nominal GDP, and its components i.e., real GDP and inflation rate. The findings suggest that the real GDP target significantly improves the forecasts compared to provisional or revised targets. The inflation target regime has an insignificant impact on improving the projections.

The third part of the last objective is a check of the usefulness of the buoyancy approach  and  alternative  forecasting  methods.  The  forecasting  is  done  using alternative theoretical, statistical, and machine learning methods. The mean absolute error suggests marginal tax rate approach is the best tax revenue forecasting method. On the other hand, the root mean square error suggests that LASSO (Least Absolute Shrinkage and Selection Operator)/Elastic Net is the best forecasting method for Total Tax,  Custom  Duties  and  FED.  At  the same time,  the  Box  Jenkins  methodology provides the best forecast for Direct Tax and Sales Tax. The study also considers the feasibility of the proposed methodology in terms of budget, time, and organizational terms.