PDF: 5191 KB - Bureau of Infrastructure, Transport and Regional ...

Appendix D | Forecasting process details in the SAS forecasting system7. Data transformations• logarithmic• logistic• square root• Box-Cox.8. Combining or average the predictions of other forecasting models9. External (judgmental) forecasts10. Customised modelsSection 3: Statistics of fitThis section explains the goodness-of-fit statistics reported to measure how welldifferent models fit the data.The various statistics of fit reported are as follows. In these formula, n is the numberof non-missing observations and k is the number of fitted parameters in the model.Number of Non-Missing ObservationsThe number of non missing observations used to fit the model.Number of ObservationsThe total number of observations used to fit the model, including both missing andnon missing observations.Number of Missing ActualsThe number of missing actual values.Number of Missing Predicted ValuesThe number of missing predicted values.Number of Model ParametersThe number of parameters fit to the data. For combined forecast, this is the numberof forecast components.Total Sum of Squares (Uncorrected)nThe total sum of squares for the series, SST, uncorrected for the mean: ∑ t =y1Total Sum of Squares (Corrected)nThe total sum of squares for the series, SST, corrected for the mean: ( − ) 2where y is the series mean.Sum of Square ErrorsnThe sum of the squared prediction errors, SSE, ( ) 2where ŷ is the one-step predicted value.SSE = ∑ y − ˆ= ty ,t1t2t .∑ t = 1 tyMean Square Error.The mean squared prediction error, MSE, calculated from the one-step-aheadforecasts. MSE = [1/n] SSE. This formula enables you to evaluate small holdoutsamples.y ,257