Automodels in Forecasting APO DP

[srinivas_krishnamoorthy]

I had a few starter questions in APO DP implementation.

1. What is the best Forecast Strategy to start with for CVCs to be forecasted? - Automodel 1 or 2 or is there any other better approach?

2. How do we exactly use Automodel 1? Since it can potentially do testing for trend and seasonality, do we switch to Automodel 2 if we find both of them to test positive ? Can we perenially work on Automodel 1?

3. How do we automate the evolution of Forecasting models? Say the data pattern begins to show positive test for seasonality through automodel 1, how can strategy automatically switch to seasonality model?

4. If the trend and seasonality test negative, how can we automatically switch the model to a more appropriate one?

5. In automodel 2, we find sometimes alpha to be proposed as 0.5 (which is the edge of the range) and Beta & Gamma to be 0 although the range is 0.1 to 0.5. Does this mean that automodel 2 does not fit in? If yes, how can strategy switch to a better model?

Please share best practices on how we can automate Forecast model switching.

thanks

srinivas

[Former Member · Accepted solution]

Hello Srinivas -

1. What is the best Forecast Strategy to start with for CVCs to be forecasted? - Automodel 1 or 2 or is there any other better approach?

-- > It depends on the data that you are using for forecasting purposes. if you are not able to determine whether the set of the data that you have is seasonal or trend or constant you can go aheas and select automatic model 1 or 2 and see the results for yourself.

2. How do we exactly use Automodel 1? Since it can potentially do testing for trend and seasonality, do we switch to Automodel 2 if we find both of them to test positive ? Can we perenially work on Automodel 1?

-- > its upto the business requirements and how they are satisfied, depending on the results that each of the model generates and that best suits the requirement needs shud be used.

3. How do we automate the evolution of Forecasting models? Say the data pattern begins to show positive test for seasonality through automodel 1, how can strategy automatically switch to seasonality model?

--> Gurus answer (would like to know myself if there is a way)

4. If the trend and seasonality test negative, how can we automatically switch the model to a more appropriate one?

-- > before even u want to switch to the model that gives you the results that satisfies the business reqs, you can run thru the simulation procedure and see what fits the requirements.

5. In automodel 2, we find sometimes alpha to be proposed as 0.5 (which is the edge of the range) and Beta & Gamma to be 0 although the range is 0.1 to 0.5. Does this mean that automodel 2 does not fit in? If yes, how can strategy switch to a better model?

-- > Gurus Answer...

Hope this helps.

Regards,

Suresh Garg

[srinivas_krishnamoorthy · Accepted solution]

Thanks Suresh for your reply.

I am looking for a technical solution to the problem that I mentioned. We are looking at a touchless system that is intelligent enough to take decisions based on changed data patterns. I was looking at how messages generated from say automodel 1 can be used in an automated fashion to take smart decisions. Eg. say the message is positive for season and trend test then system should switch strategy to 56. Else the solution becomes very manual intensive and prone to errors.

[srinivas_krishnamoorthy · Accepted solution]

Still looking for a possible solution to automate switching of Forecast Models. Would any DP guru in the forum think that we should have the Forecast Strategy as a characteristic as an example ?

The basic question I have is as history builds up for a CVC with time, how can Forecast model sense this buildup and evolve using a more powerful Forecast strategy automatically? What is the best practice?

[Former Member]

Did you find answer to this?

If Yes, could you please share?