I am using ARIMA tool for forecasting sales of a particular product. I have to forecast sales of 35 different products with same starting and ending period. Is there a way to forecast sales by running ARIMA or any other tool JUST ONCE instead of running it MULTIPLE TIMES for each product separately.
Any help will be appreciated!
Solved! Go to Solution.
Hi @NM,
Your problem is exactly the reason that the new Time Series Forecast tools were created! The TS Model Factory tool allows you to generate models for multiple groups at once. You just need to specify which field contains the string identifying the groups. Then you can use the TS Forecast Factory tool to create your forecasts. If you want an example of how to use the tools, you should check out the TS Factory Sample. But if you still have any more questions about the tools, feel free to ask them here! (I created them.)
Best,
Bridget
Hi Bridget,
The model works exactly as I expected it to be. Thank you for sharing this tool.
Only question I have is that with TS Model Factory tool, I am not able to customize the parameters used for automatic model creation, i.e. I can not specify order of AR or MA components like I can do in ARIMA. Due to this, my results from ARIMA and TS Model Factory tool are different.
Is there any way to customize AR or MA components in TS Model Factory tool?
Thanks!
Hi @NM,
Unfortunately the tools don't currently allow for model customization; they simply use the R auto.arima function to create an ARIMA model with "optimal" conditions. (See more about the auto.arima function here, page 12.) This choice was intentional, since we understand the typical use case of this tool to involve creating hundreds or thousands of models, which would be unreasonable to specify. However, if you do want this functionality, feel free to request it here.
Best,
Bridget
Are there any restriction of minimum number of records in a product group, I have a product with 85 weeks of sales but i am getting an error stating insufficient number of records in group to build arima model
Hi @aravindraj004,
Yes, there are restrictions on the minimum number of records needed for each group. As stated in the help, "You need to provide at least two full repetitions for each group. For example, if you select Hourly, you would need to have at least 48 measurements per group (24 hours in a day * 2 = 48)." Since you're working with weekly data and there are 52 weeks in a year, you'll need 2*52 = 104 records. However, if you have some other groups with sufficient data, the tool will skip the groups with insufficient data and generate models for the other groups that have enough data.
Best,
Bridget
Great discussion here...
By the way I guess I've found a little bug in the TS Model Factory yesterday.
Let's assume you do auto ARIMA with covariates for multiple groups using Model Factory
and then decide to turn to ETS where only univariate analysis can be handled
Somehow you may receive an error regarding ETS not being able to utilize covariates...
-Make sure you don't provide covariates to the TS Forecast Factory while changing to ETS
Solution is to go back to ARIMA de-select covariates forecasting and then getting back tor ETS...
A small bug that remembers your ARIMA selection for ETS...
-Go back to ARIMA
-Deselect the "use covariates in model estimation (ARIMA only)"
-Back to ETS and voila! it works...
Best
Hi
A quick question. Is It possible to group on multiple fields In the Ts Model Factory macro ?
For example if I am predicting sales, Is It possible to add both Product and Product Category in the grouping field in Ts Model Factory Tool .
If not is there a workaround for this ?
Thanks
You can only group on one thing within the TS Factory option. The work around is dividing out your data yourself. I was doing forecasts for 300 medical centers along 7 lines of business at each medical center. I had to filter out each data set so that there was only one thing to group by when it went thru the TS Factory icon.
Concat multiple fields into one and use this new field for grupings instead...
this should solve it...
Best