Training Data

The length of the training period must be specified. The length defines the total number of years available to construct the forecasting model. The length of the cross-validation window defines how many years to leave out when performing cross-validation, either in calculating the optimum number of modes, or for calculating performance statistics (discussed later). Leave-one-out cross-validation is most frequently performed, but for robust results it is recommended that a much larger window be used if sufficient data are available. The length of the window is constrained to be odd, and only the middle year of the cross-validation window is predicted at each step. The length of the cross-validation window can be changed from Options ~ Cross-Validation Options .