Shrunken interest rate forecasts are better forecasts

Abstract

Predicted changes in interest rates are imperfectly correlated with actual changes in interest rates. One statistical consequence may be that large predicted changes are more likely to be overestimates than underestimates of the magnitude of the change. If so, the accuracy of predicted interest rate changes can be improved by shrinking them toward a prior mean of zero. The application of this idea to interest rate forecasts by the Survey of Professional Forecasters found a consistent improvement in the accuracy of their predictions.