BookmarkSubscribeRSS Feed

Dear All,


I am using forecast server studio for producing next 18 months of demand of retail products. I have some challenges. Could you please help me.

1. even for products those have more than 24 months of history I am getting constant forecast. My crieteria is holdout accuracy. 

2. I want to control the trend. For some product there is decreasing trend but forecast servers makes it very quickly zero forecast. I want to control this becasue clients say the products will chnage the level but will not suddenly become zero in next 7-10 months . We have to introduce an intelligent levels




SAS Employee

Hi Lokendra,


Please share or send screen shots of your forecast project settings as well as the hierarchy you are using.  You can access the forecast settings by selecting Project from the menu and then Forecast Settings.  Feel free to add or send any other information that you feel is relevant.  If you can share the data and I can replicate your issues that may be helpful as well. Thank you.



SAS Employee

Hi Lokendra,


Regarding 1., if product sales are either highly stable, or highly erratic, then a flat line (constant) forecast can be most appropriate.


For highly stable demand this intuitively makes sense, and would be easy to explain to a client or other user of the forecast.


In your situation, since this is retail data, I suspect you are seeing highly erratic demand patterns (particularly if you are forecasting at a granular level like store/item/week).


A flat line forecast may not make intuitive sense for highly erratic demand because it doesn't appear to "fit" the history. However, if there is no underlying structure in the data -- that what you have are essentially random ups and downs -- then there is no pattern to be fit, and just forecasting the average demand makes sense. This appears to be the situation with your data because the constant forecast is performing best in the holdout sample.


A 2015 article in the Journal of Business Research ("Simple versus Complex Forecasting: The Evidence" by Green and Armstrong) found considerable evidence that simple models forecast better than complex models. Complex models often "overfit" the history -- mimicking the randomness rather than any real structure in the data and projecting it forward -- and thereby generate less accurate forecasts.






Available on demand!

Missed SAS Innovate Las Vegas? Watch all the action for free! View the keynotes, general sessions and 22 breakouts on demand.


Register now!

Multiple Linear Regression in SAS

Learn how to run multiple linear regression models with and without interactions, presented by SAS user Alex Chaplin.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 2 replies
  • 3 in conversation