FORECAST MODIFICATION BASED UPON RESIDUAL ANALYSIS: A CASE STUDY OF CHECK VOLUME ESTIMATION

The inability to identify all causal variables in a linear regression demand model may result in serial correlation which is generally considered undesirable; but it may be possible to take advantage of such an event. This case study, based upon Chemical Bank of New York, investigates the use of simple and exponential smoothing for modifying initial estimates from a regression model by using prior forecast error patterns to obtain better forecasts. The smoothing approaches are combined with a regression model to test for improved performance in predicting daily check volumes.