Predicting National and Regional Recessions Using Probit Modeling and Interest-Rate Spreads

This study extends the work of Estrella and Mishkin (1996, 1998) to show that interest-rate spreads and probit modeling can be used to predict recessions in many states as well as the nation. State recessions are defined as two or more consecutive quarters of declining real gross state product. The yield spread, SPREAD, is defined as the difference between the 10-year Treasury bond rate and the three-month Treasury bill rate. The national results are similar to those obtained by Estrella and Mishkin. Probit models are estimated for all 50 states using SPREAD and unemployment insurance claims, UI, as alternative explanatory variables. For 34 of the 50 states, SPREAD is significant at the 0.01 level as a predictor of state recessions. Much weaker results are obtained using UI. Simulations for the 1979-2001 period are used to compute loss functions for the national and state models at probability screens of 30, 40, 50, and 60 percent. The results demonstrate that probit models based on can be useful in improving business and policy decisions in many states