Econometric Forecasting of Budget Revenues: The Case of Tajikistan

Forecasting budget revenues is an important task for government agencies involved in budget planning. While preparing official forecasts, simple computations are often used. However, it might lead to inaccurate forecasts. The article is devoted to the development of a tax revenues forecasting system with high predictive power which could be used by government agencies, including those in developing countries. Our system automatically generates a large number of various econometric specifications, estimates them, and combines the forecasts they produce. The system uses time series models with exogenous regressors (causal time series models). The key exogenous variables in these models are various proxies for taxable bases, the model tax rate, as well as proxies for tax proceeds themselves. Before combining models, several of them could be selected based on their quality, i.e., how well they correspond to the presumed theoretical assumptions, how high their predicting power is, and how easy their estimates are to interpret. The system is tested on data for the Republic of Tajikistan for annual proceeds from VAT, corporate income tax, and personal income tax. Before testing, we analyze official data from open sources to obtain a list of relevant proxies for tax bases, and to develop a method for calculating model tax rates. Then we proceed to forecasting. By comparing the (out-of-sample) system’s forecast with the forecast of the Ministry of Finance of the Republic of Tajikistan, we show that for 2023, the system gives a lower forecasting error.