Questions about Statistical Forecast
Hi, we are trying Monthly Stat Forecast Model.
(https://community.anaplan.com/t5/Best-Practices/Statistical-Forecasting/ta-p/40359)
And we are doing the functional verification.
I check the model, but there are somethings I don't understand.
Q1. Can we use specific history data?
ex) Using 2018-2019,2021-2022 data as history, not using 2020 data because they were influenced by specific reason.
I know we can control the start and end history date, but it is not enough.
Q2. Requirement about Periods of Training Data.
I heard that we should use 2 or 3 years history data.
If we use few history data, can we predict the forecast? Is there any restriction?
Q3. About Null
In this model, how can we handle the NA data?
Tell me Anaplan's policy.
) like replace null with 0 or mean, not use NA data.
Are there any Action Buttons that can controll it in bulk?
Answers
-
Q1: you should be able to use the outlier functionality or the override
Q2: depends what you are trying to predict. the model is made for that, load your data, look at the results, and choose where to apply stats forecast.
Q3: there is no NA in Anaplan for numbers. In which way is this a concern for you ?0 -
thanks for replay!
Q1. I can control it by using outlier.Q2. You says that some case we cannot forecast because of the data?
Are there any dashboard we can check the forecast error?Q3. OK, I forgot it. I forecast by using machine learning method, so I was careful about NA data.
When the data(forecast=0 value) is much, should we use Croston's method, right?
I'm worried about the forecast does not work because of 0 data.
Additional Question
Q4. Flat List vs each Lists
In the Monthly Model, you're forecasting by Flat List.
like, Product - Customer - Region - DC
Why you don't separate it to each List?
Is that spend so much memories?
Q5 . About Allocation.
In Monthly model, can we use this allocation function?
Example)
<Forecast>
Region : Europe = 100
Country : France = 50, Germany =50
0
