Forecasting the Government of Lesotho's budget: an AR-MIDAS approach

Abstract

This study uses an autoregressive mixed data sampling (AR-MIDAS) regression technique and monthly fiscal variable sub-components sampled from 1993Q1 to 2017Q4 to forecast quarterly aggregated key fiscal variables (i.e., total revenues and expenditures). Results of forecast evaluation criteria show that quarterly forecasts of total government revenue and total government expenditure are best performed by a PDL/Almon weight variant of the AR-MIDAS model with monthly values of indirect taxes and compensation of employees as predictor variables, respectively. The study recommends that the PDL/Almon weight variant of the AR-MIDAS model be used to complement techniques currently in place at the Central Bank of Lesotho and the Ministry of Finance for macro-fiscal forecasting. This will help support the Government of Lesotho's public financial management system as well as its medium term expenditure framework.