A New Approach To Forecasting the Food Stamp Caseload

Year: 2005

Research Center: The Harris School of Public Policy Studies, University of Chicago

Investigator: Grogger, Jeffrey

Institution: Harris Graduate School of Public Policy Studies, University of Chicago

Project Contact:
Jeffrey Grogger
Harris Graduate School of Public Policy Studies
University of Chicago
1155 E. 60th Street
Chicago, IL 60637
Phone: 773-834-0973


Over a span of less than 15 years, the FSP caseload has exhibited three prominent turning points. Caseload declined from 22.4 million participants in 1981 to 18.6 million in 1988. Between 1989 and 1994, it rose to an historical high of nearly 27.5 million people before plummeting to 17.1 million in 2000. Caseload then rebounded to 21.3 million people between 2000 and 2003.

Such extraordinary levels of volatility present a serious challenge to the agencies charged with operating the program. Although little can be done to reduce the volatility, the administrative challenges posed by that volatility could be reduced if more accurate caseload forecasting were possible.

This study develops an approach to forecasting based on a leading indicator suggested by Markov theory. The leading indicator is a function only of current entry and exit rates. Unlike conventional econometric forecasting, the approach developed does not require forecasts of future environmental conditions. The technique is applied to caseload data from California.

The results highlight both the advantages and limitations of the approach. On the plus side, the forecasts are fairly accurate and perform reasonably well in predicting a recent turning point. On the minus side, the horizon over which the approach produces forecasts is determined by the data, and in the case of the California data, they vary greatly between segments of the caseload. For the public aid segment, which includes households that receive cash welfare payments as well as Food Stamps, the approach yields forecasts 16 months ahead. However, for the non-aided segment, the forecast horizon is only 4 months. Thus, the approach yields relatively long-term forecasts for the aided segment but relatively short-term forecasts for the non-aided segment. The study finds that the forecast horizon appears to be related to the extent of turnover in the caseload.