Time Series Models (ETS|ARIMA) - Adding more variables than just Period and Amount

[Elena_P]

Hi all, I'm new to the Time Series Models in Alteryx but since I'm working on a project for Cashflow Forecasting and for the purposes of the analysis they want to have the forecasted values within the output of ETS/ARIMA not only per period, but per Period, Account, Flow and Budget codes as in the screenshots below. Not sure if this is possible, but can we somehow add other variables to the Model or is there another workaround/suggestion in order to achieve this?

Thank you in advance!

[gabrielvilella]

Hi @Elena_P, only ARIMA will let you use covariates on the model, by the nature of it. There is a checkbox you can enable on the ARIMA configuration area. I will leave here a few useful articles on this topic:

• https://community.alteryx.com/t5/Data-Science/Back-to-the-Future-ARIMA-and-Forecasting-with-Covariat...
• https://community.alteryx.com/t5/Learning-Paths/Data-Science-Learning-Path/ta-p/504157
• https://kubicle.com/learn/alteryx/running-a-covariate-forecast

[Elena_P]

@gabrielvilella , thank you a lot for the provided information and the articles! They really helped and I managed to add covariates to the model. I have two more questions:
- Do you perhaps know if there is a limitation on the number covariates we can use within the ARIMA model? As if I add 2 it works, but if I add 3, I get errors..

- Within the output is it possible to add the covariates next to the Period, Sub_Period and the forecasted values? As for the purposes of the analysis they need this information, only period and amount are not going to be sufficient.

Thank you in advance for the help!

[gabrielvilella]

I am not aware of a limitation on the number of covariates. Maybe there is something wrong with your third covariate, you can verify this by adding just this field as covariate, for instance.

About adding the covariates next to the output, how exactly do you want this to be presented? Because there will be a lot of values for those covariates and they do not relate to the periods. 

[Elena_P]

@gabrielvilella, thank you for the prompt reply!

On the first point, yes, I will test it and see why it is not working then.

As for the covariates, what is needed is to have the forecasted value and the corresponding covariates it relates to (see example below). But then I guess I will need to do that for every unique combination of Account-Cashflow Type-Code as otherwise I will have multiple combinations possible per one day and then a forecast will not be created for each separate combination of covariates per each day, but will be a generic one based on all possible combinations of covariates for this day and then it cannot be represented in the way it is needed.

[RolandSchubert · Accepted solution]

Hi @Elena_P ,

if I got you right, you want to create a forecast for each combination of account, cashflow type and code.  In my understanding, you need a time series model for different groups (eg. ACC1 - INVL - GTT), while a covariant adds an additional "influencing factor" = predictor to your forecast (e.g. change to prices pr payment terms). 

Maybe it would make sense to have a look at TS Model Factory (TS Model Factory - Alteryx Community) to solve your problem. You could create a kind of "group identifier" as a combination of account, cashflow type and code and let the tool create a model for each combination. To calculate your forecast, you can use TS Forecast Factory (TS Forecast Factory - Alteryx Community).

What do you think?

Best,

Roland