Time Series Forecasting with differenced series

Hello,

how can I get the original values back from a forecasted differenced time series? The picture below shows my original series (Original GDP), the lagged column and finally the differenced column. I used the differenced column to forecast 2001 - 2003. The results from the forecast are shown in column E. To get the original values I used the last available original value (2000Q4) plus the forecast. But afterwards I need the “Forecasted Original” previous value to calculate the next original value. Is there any possibility to do this easily in KNIME?

Any help would be much appreciated.
Thank you in advance,
Franziska

Hello @Franziska_W

You can use the Lag Column node to get a row+1 $Forecasted Original$ column, then calculate the difference has to be quite straight forward.

BR

Hello @gonhaddock

Thank you for your fast reply!
I was already thinking of that but I don’t really understand how the Lag Column can help.

The first row I can simply calculate (11,334.544 + 99.704). Then I used Lag 1 and now I can calculate row 2 (11,434.248 + 78.397). But afterwards I need apply the Lag 1 again and this is very time-consuming. Am I missing something here?

Thank you.
BR
Franziska

I see, your solution is a ‘Recursive Loop’ as you have to assign the result of the calculation as part of the calculation for the next row.

I can try to built an example workflow as the configuration isn’t simple. Can take me some time.

BR

That would be really great! Thank you in advance!

BR
Franziska

Hi @Franziska_W

Maybe the solution looks a bit complex, and in fact it is. Should be easier to code it in a script node with R or Py; however it is a good exercise to have in KNIME Hub

You will notice some decimal differences from CHALLENGE result to TARGET column, but I just worked with the 3 decimals displayed in your picture.

I think that it should fit with your needs.

BR

2 Likes

This worked - thank you so much @gonhaddock !

BR
Franziska

1 Like

Column Expression with Moving Agg should also work

br

3 Likes

kudos @Daniel_Weikert
I didn’t even thought about this approach.

1 Like

Thanks @gonhaddock
“I didn’t even thought about this approach.” That’s what I think each day I see a new well designed solution in this forum. The community here is really unique. Thanks to you and others people can really find help here.
Thanks for your contribution.
br and take care

2 Likes

@Daniel_Weikert
also thank you for your approach!
I can totally agree with you that this community is really great.

BR
Franziska

3 Likes

Hi,
Maybe I was biased in this solution. I used this WF before for a much more complex forum topic ; that solution addressed a very complex case, with an unsolved data gathering, and double equation system for many items, implemented with a nested ‘Recursive Loop’ .

As summary, I saw a good opportunity to clean-up the workflow with this simple equation. Now, the complex part to configure the Recursive Loop is clean, and it can be scaled up for a more complex problems. And besides that, it is always a good fun to participate.

BR

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.