1
$\begingroup$

I have an ARIMA model, I'm trying to figure out how much an external variable (exogenous covariate) could improve the forecast, so I need to "synthesize" a rate that tell me the usefulness (or impact) of a new external regressor. Does anyone know a good way to do it?

$\endgroup$
0
$\begingroup$

Do this: create a series X, which is a multiple of your target variable say 3 or 5 times or a fraction of your target variable say 10% or 15%. Multiply your target variable values with the selected multiple or fraction but don't multiply all your target values by say a single multiple or fraction value, multiply target values by say numbers in a close range say between 10-13. example below

so Step 1: if my Y (target)= [10,20,30,40..], then I create X as [10*10=100,20*13=160,30*11=330,40*11.5=460...] (created random numbers between 10-13 and multiplied them by target values) Same multiple will create perfect relation between X & Y, and different multiple values in a close range will overcome this to a good extent.

Step 2: To further ensure that X is not a perfect predictor of Y but only reasonably good one and there are other predictors at play like TIME (since you are doing ARIMA), add random noise to X by creating some random numbers and adding them to series X creating a new series say Xnew.

random noise= [25,37,31..]

Xnew= [100+25, 160+37, 330+31..so on]

thus you can now add Xnew to your ARIMAX model.

Xnew is definitely a significant predictor, that you will see in your ARIMAX results.

Best Vik

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.