Forecasting using SAS Forecast Server, SAS/ETS, and more

how can you know your forecasting is good enough?

Accepted Solution Solved
Reply
Highlighted
Frequent Contributor
Posts: 139
Accepted Solution

how can you know your forecasting is good enough?

Hi All,

 

    I have a general question about forecasting, how can we know the forecasting model is good enough for your data? Usually what parameter in forecast studio result we should look at? Thank you!

 

Thanks,

Jade


Accepted Solutions
Solution
a week ago
SAS Employee
Posts: 19

Re: how can you know your forecasting is good enough?

I always like to set a holdout period to validate the model. In forecast studio, you can set a holdout period in Project->Forecasting Settings->Forecast->Calculate statistics of fit over an out-of-sample range. See attached picture, where I set the length of holdout period to be 6.

 

In Modeling view, there are several useful tables and plots that you can view to help determine whether it is a good model or not.

For example, you can compare summary statistics of fit between in-sample period and holdout period. If holdout MAPE is much worse(bigger) than in-sample MAPE, you may want to check if there's any overfit in your model.

You can also check the Unbiasedness Test table, or view Prediction Error Histogram to see if there's any systematic bias in error.

You may also want to check data points that have extreme errors, to check for outliers, data error, or uncaptured features.

Hope this helps.

Set holdoutSet holdoutModeling ViewModeling View

View solution in original post


All Replies
Solution
a week ago
SAS Employee
Posts: 19

Re: how can you know your forecasting is good enough?

I always like to set a holdout period to validate the model. In forecast studio, you can set a holdout period in Project->Forecasting Settings->Forecast->Calculate statistics of fit over an out-of-sample range. See attached picture, where I set the length of holdout period to be 6.

 

In Modeling view, there are several useful tables and plots that you can view to help determine whether it is a good model or not.

For example, you can compare summary statistics of fit between in-sample period and holdout period. If holdout MAPE is much worse(bigger) than in-sample MAPE, you may want to check if there's any overfit in your model.

You can also check the Unbiasedness Test table, or view Prediction Error Histogram to see if there's any systematic bias in error.

You may also want to check data points that have extreme errors, to check for outliers, data error, or uncaptured features.

Hope this helps.

Set holdoutSet holdoutModeling ViewModeling View

☑ This topic is solved.

Need further help from the community? Please ask a new question.

Discussion stats
  • 1 reply
  • 92 views
  • 1 like
  • 2 in conversation