Did you miss the Ask the Expert session on using Events to improve your forecasts in SAS Viya Visual Forecasting? Not to worry, you can catch it on-demand at your leisure.
Learn how to use Events in SAS Visual Forecasting as a simple yet powerful tool to improve the accuracy of forecasting models. This webinar is intended for SAS users who have a good understanding of SAS® Viya® Visual Forecasting. During this webinar, you will learn:
- How to use Events in your models to improve forecast accuracy.
- How using Events is one of the simplest ways to improve your forecasts.
- Why the accuracy of your forecasts is important to your business.
You can use one of the attached images (or something else from your presentation).
Here are the questions from the Q&A segment held at the end of the webinar. The slides from the webinar are attached.
Given the significant event of Coronavirus or COVID-19, how do I factor that into a forecasting model?
COVID-19 is, obviously, a unique event and one that is still ongoing. So, it is hard to determine the shape of the event. Right now we can count it as a level shift, but the characteristics of the measure time series in terms of trend and cycle may/likely differ before versus after. Also, there is likely ramping before and after as the event progresses. I can only suggest to monitor your most important set of time series through this and characterize the behavior through one or more events. SAS forecasting experts have been working a lot with this and provided some analysis and papers related to forecasting in COVID-19.
Are there limitations as to the number of Events to include in a model?
There are no set limits. Theoretically, the forecasting algorithms will use any events if significant for the model and ignore if they are not. However, practically it is probably best to only include candidate events that you, in your business experience, think might realistically have an impact.
How would I determine if an Event is an actual Event versus a random outlier?
Outliers are generic points in a time series that deviate some significant amount (e.g. two standard deviations) from the value as predicted in the model. In certain modeling algorithms certain outliers can be ignored. Events “recharacterize” the associated outlier(s) with a factor (promotion, storm, energy shock) that can be incorporated into the model.
Do we create separate events variables for each outlier?
You can and, perhaps, should if and only if you can associate that outlier with a factor (internal or external) that could/should be incorporated into the model.
Is there a way to identify (in a statistical way) which events or/and attributes are statistically significant? Like we have in Forecast Server, a table with the estimates of the parameters and the relating p-values. My question was can we see the significance of the events/attributes in SAS Visual Forecast?
The current release of SAS Visual Forecasting (VF) does not create a table that records the parameter estimates of the champion model for each BY group. That functionality will be included into a new release of SAS Visual Forecasting next year. In the meantime, this BLOG post describes how to generate the OUTEST table, which contains model parameter estimates, in VF and how to use that in SAS Visual Analytics or SAS Studio for reporting and analysis.
Can we emulate an event with a variable?
You can if you change the dummy variable to go between a range of 1 to 0 (or 0 to 1). For example, say a marketing campaign has a decaying impact: set it at 1 in the first period and then have it decline to 0 when the impacts have ended.
That information is included in the Events Repository table generated from PROC HPFEVENTS. Here is associated documentation:
Can we point out that a specific event is applicable to specific products only?
In SAS Visual Forecasting, there is not a simple way to do that. The list of candidate events is defined for the project which may include many time series and different levels of a hierarchy. Automatic modeling algorithms will look at the significance of candidate events and include event(s) for each series only if significant, if that is your strategy. So, an event may be included in models for certain products but not others.
The list of promos is in an Excel table. Can we automate creating events from Excel?
You can use Excel as an import table to create an abbreviated events table for SAS Visual Forecasting. You can use an Excel table with only a _NAME_ and _STARTDATE_ column to feed into PROC HPFEVENTS to create an events table that VF can use. Also, there is a %HPF_ImportEventsFromExcel() SAS Macro. It is SAS9 based. But the SAS Data Set can be moved to a CAS Table. The macro will import directly from Excel into HPFEVENTS, and it also allows for condensed info. More info: https://go.documentation.sas.com/?docsetId=fsdskag&docsetTarget=p1bn0s3kle9qutn1wqbvsugw82ga.htm&doc...
Is there a method to determine when to change the trend from multi-pulse to shift model based upon the prior observational patterns?
Yes. You can change the characteristics of any event in the repository by using PROC HPFEVENTS to “recharacterize” the event type, ramp pattern, decay pattern etc. Theoretically, you could have two events in the repository list – one that is multi-period pulse and one that is a level shift corresponding to the same start date. The model will choose the event that is significant in the model.
Recommended Resources
SAS Visual Forecasting 8.5: User's Guide
YouTube: SAS Visual Forecasting
Want more tips? Be sure to subscribe to the Ask the Expert board to receive follow up Q&A, slides and recordings from other SAS Ask the Expert webinars.
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.
-
1758 views
-
0 likes
-
-
