Hi Girish
Though you may know this i am just going through this for the sake of clarity
Ex-Post forecast is a forecast that is run in past periods for which actual demand history is also available
If more historical values are available than the system needs to initialize the model, an ex-post forecast is carried out automatically. history is divided into two groups. the first is used for model initialization and the second(more recent ones) is used for ex-post(i dont think you can define these periods though)
the basic value, trend, seasonal index and MAD are modified on a moving basis in each expost period. these new values are now used to calculate the forecast.
the total error of the actual history vs the ex-post in the historical periods is calculated and is used to justify the model by forecast accuracy measurements
So Expost Forecast for seasonal models should use the same formula that the regular forecasting model uses.
what specific forecasting model are you talking about ? a seasonal model takes 1 season for initialization. the basic value =the mean of the historical values of the first season with a trend=0 and MAD = 0 and seasonal index = hostorical value of the period/basic value
Hi Harish,
Thank you very much for the info. I have this information, but I would request you to help me understand the system calculation. I am providing a small portion of the actual forecast run for your ref.
Fore cast model: Seasonal (30)
Alpha: 0.3
Gamma: 0.3
Sigma: 1.25
History horizon: 36 months
Forecast horizon: 20 months
Number of periods per season: 3
M 05 M 06 M 07 M 08 M 09 M 10 M11 M12
History 200 300 400 500 600 700 800 900
Corrted Hist 200 300 400 500 600 700 403 200
Ex-Post Fcst 0 0 0 200 435 646 403 532
Seasonal 0 0 0 0.67 1.0 1.33 0.81 1.07
Basic Value 0 0 0 435 485 497 497 404
Here 800 and 900 were identified by the system as outliers and corrected to 403 and 200.
MAD calculated by the system here is 638.94
Now as per the standard formula for basic value, seasonal index does not match to the results of the system. All I want to know how the outliers/ex-post forecast/MAD is calculated and how to interpret the results to the users with the help of calculating it manually and tell them this is how the system is computing. I have tried using the standard formulas and tried to compare my results with that of system and it did not work. Please let me know how the system is calculating it. Your help is very much appreciated (It can save my head).
Regards,
Add comment