Model Initialization
Purpose
Model initialization is the process by which the system determines the necessary model parameters for the chosen forecast model. These parameters are as follows:
Model | Model parameters |
Constant models | Basic value = Hist(1) where Hist(i) is the ith historical value Trend value = 0 Seasonal indices = 1 Mean absolute deviation (MAD) = 0 |
Trend models | Trend value = Basic value = Seasonal indices = 1 Mean absolute deviation (MAD) = |
Seasonal models | Basic value = Mean of the historical values of the first historical season Trend value = 0 Seasonal index for a period = Historical value for that period divided by the basic value Mean absolute deviation (MAD) = 0 |
Seasonal trend models | Trend value = Basic value = Seasonal index for a period = where S(i) is the ith seasonal index. Mean absolute deviation (MAD) = |
The system uses a fixed number of historical values to carry out model initialization. This number depends on the forecast model:
Model | Fixed number of historical values used for initialization |
Constant | 1 |
Trend | 3 |
Seasonal | 1 season |
Seasonal trend | 3 + 1 season |
Automatic model selection with selection with seasonal test | 2 seasons |
Automatic model selection with seasonal + trend tests | 3 + 2 seasons |
2nd-order exponential smoothing | 3 |