Forecasting State Tax Revenues: A Simple Alternative

Abstract

Recently the State of Minnesota experienced a fiscal emergency as a result, in part, of an inability to accurately forecast tax revenues. The approach used by the State is based on an econometric model that attempts to specify the linkages between the national and state economies. Such econometric models have been challenged because of their inability to forecast by those who propose a more mechanical, or statistical, approach. While econometric models estimate equations specified by theory, the statistical approach finds the best "model," or forecasting equation based on statistical criterion alone. Using a forecast error criterion, recent studies have shown that one fitting approach, vector autoregression (VAR) can outperform econometric models. Given the complexity of VAR, as opposed to some simpler statistical approaches available, a question arises as to whether VAR significantly out-forecasts such naive methods. In this paper a simple time series method, Box-Jenkins analysis, is used to forecast monthly revenues for Minnesota using data from 1970-84 and the results compare favorably to VAR and econometric models for both aggregate (total tax) and disaggregated (sales and income tax) specifications.