Panel Paper: Forecasting State Tax Revenue in the Face of the Great Recession

Saturday, November 4, 2017
New Orleans (Hyatt Regency Chicago)

*Names in bold indicate Presenter

Melissa McShea1,2 and Joseph Cordes1, (1)George Washington University, (2)University of North Dakota


Better forecasts allow states to better anticipate revenue declines in advance of a recession. The more advanced warning a state has of revenue short-falls, presumably the better able it will be to make needed budget adjustments.

Despite the general claim that the Great Recession was a surprise, prior to its onset, tax revenues in many states were signaling that a downturn was imminent. Nonetheless, once the recession hit, numerous states were caught by surprise. This paper focuses on the relative performance of several different approaches to revenue forecasting. The paper addresses the question of whether states can improve the accuracy of revenue forecasts by using more advanced time series and Bayesian vector autoregression (BVAR) forecast methods. Keeping with this year’s conference theme, we have relied on various Federal agencies to supply the data for our research. For our dependent variables, we use annual state revenue data collected by the Governments Division of the U.S. Bureau of the Census. For our independent variables, we use data series compiled by the Bureau of Economic Analysis, Energy Information Administration, Federal Housing Finance Agency, and the Bureau of Labor Statistics. Using this revenue data for Virginia, we first estimate baseline forecasts using vector autoregression (VAR) and vector error correction (VEC) techniques. We present the theoretical case for estimating a BVAR model; and then we present and compare forecasts based on the VAR and the BVAR. We find that the BVARs are especially effective in forecasting relatively more volatile revenue series such as the corporate income tax revenue. Future research entails replicating this study across other U.S. states to detect if there are patterns of forecasting reliability due to shared state characteristics.