International Journal of Forecasting 5 (1989) 321-331 North-Holland 321 An empirical examination of bias in revenue forecasts by state governments * Glenn CASSIDY, Mark S. KAMLET and Daniel S. NAGIN School of Urban and Public Affairs, Carnegie Mellon University, Pittsburgh, PA 1.5213, USA zyxwvutsrqponmlkjihgfed Abstract: This analysis examines the influence of economic forecast errors and political and institutional factors on the general fund revenue forecast errors of 23 states for the period 1978 to 1987. During this period states in the sample underestimated their revenue by only 0.5%. This modest tendency toward conservative forecasting is further reduced after controlling for economic uncertainty. Moreover, the analysis reveals no systematic relationship between revenue forecast errors and state political and institutional factors. Thus, the results cast substantial doubt on the prevailing belief as found in the literature, that state revenue forecasts have a pronounced conservative bias. Keywords: Economic uncertainty, Forecasting accuracy and bias, Political and institutional influences, State revenue forecasting. 1. Introduction With few exceptions state and local govern- ments are required to balance their budgets. As a consequence, revenue forecasts play a prominent role in their budget formulation processes. Bret- Schneider and Schroeder (1988) observe that, as the social and political costs of deficits are typi- cally greater than that of surpluses, “[local governments’] budgeters often adjust mean fore- casts downward to accommodate risk preferences” (P. 33). A review of the state and local budgeting litera- ture (e.g., Gerwin (1969), Meltsner (1971), and Larkey and Smith (1984)) reveals that most ob- servers of the process would concur with Bret- Schneider and Schroeder’s conclusion that there is a conservative bias in revenue forecasts. However, the review also suggests that empirical support for * We would like to thank seminar participants at the School of Urban and Public Affairs, Steven Garber, Wil Gorr, and anonymous referees for useful comments. This research was partially supported by the National Science Foundation under Grant No. SES-8606825. the conservative bias hypothesis is limited; conclu- sions have been based on either anecdotal evi- dence or on limited data. The purpose of this article is to examine the conservative forecast bias hypothesis using a more extensive data set than has heretofore been employed. It includes the revenue forecast errors of 23 states for varying numbers of years from 1978 to 1986. Closer scrutiny of the conservative forecast bias hypothesis is valuable for several reasons. First, as revenue forecasts are such an integral component of state and local government budgeting processes, any positive theory of resource allocation in this area of government would be seriously incomplete without an accompanying theory of revenue fore- cast behavior. A propensity to underforecast reve- nues would surely be a centerpiece of such a theory. Second, there is considerable sentiment for imposing a balanced budget requirement on the federal government. The Gramm-Rudman-Holl- ings deficit ceilings represent an initial step in that direction. An examination of the determinants of state revenue forecasts may also provide valuable perspective on the response of federal revenue forecasting behavior to a balanced budget require- ment. 0169-2070/89/$3.50 0 1989, Elsevier Science Publishers B.V. (North-Holland)