This is a problem in financial prediction with a small number of data points and high dimensionality in which classical economic forecasting techniques do not work. Many commercial institutions including banks, department stores, and credit companies charge their customers one interest rate (termed the lending rate) while they can borrow at a lower rate, termed the borrowing rate. The spread (difference) between these two rates can be a major profit center. If a commercial institution forecasts that the spread (and hence profit) will decrease, they can "hedge" by buying insurance against that decrease thereby locking in a profit. We used a variety of techniques that trade off training error against model complexity using the concept of capacity control for dimensionality reduction. We minimized the mean squared error of prediction and confirmed statistical validity using bootstrap techniques to predict that the spread will increase and hence one should not hedge. We briefly discuss the classical economic forecasting techniques which are not correct because the data is not independent, stationary, or normally distributed. Our predictions of the spread are consistent with the actual spread subsequent to the original analysis.