ETS Seasonal Type Multiplicative error

[LLebed]

Hi,

I am trying to use the ETS tool to forecast store sales, and judging by the TS Plot my seasonality should be multiplicative, however if i select Multiplicative it generates an error.

What issues in data can cause problems in ETS tool?

Thanks,

Lev

[HossC]

Hey @LLebed,

Would you be able to upload your workflow (Options => Export Workflow)? Also, can you attach a screenshot of the error message you are receiving?

Thanks!

[LLebed]

Hey @HossC,

I only now received a notification. I have now moved on towards ARIMA model and it appears to work better. I will try to update this query once i come back to ETS as i figured a better way of slicing data and would take time to re-apply ETS to it

Thanks

[fengyujing]

I met the similar error in ETS when specify seasonality with A or M, can anyone help?

[HossC]

Hey @fengyujing

Would you be able to upload your workflow (Options => Export Workflow)? Also, can you attach a screenshot of the error message you are receiving?

Thanks!

[fengyujing]

Here is the example, when I choose seasonality with auto, it run fine , but when I change it to Additive or multiplicative, it failed . Please help, thanks!

[fengyujing]

Have you get chance look this issues? any update? Thanks 

[HossC]

Hello @fengyujing,

The error message that I received was the following:

After enabling all Macro Messages, I received the following error:

The ETS: Frequency too high message is the most relevant for this error. After looking into this error message for ETS, I was able to find the reason for this behavior. The error message you are seeing is a result of specific limitations for the ets() function that is used for this tool. Specifically, the ets() function restricts seasonality to be a maximum period of 24. This article describes this behavior in further detail: https://robjhyndman.com/hyndsight/longseasonality/

Specifically, the following statement gives a brief description as to why this restriction was implemented in this function: "The problem is that there are  parameters to be estimated for the initial seasonal states where $m$ is the seasonal period. So for large $m$, the estimation becomes almost impossible."

Let me know if you have any additional questions.