1
$\begingroup$

I have to predict the number of items or order size of a customer on a monthly basis. My doubt is which will work better or provide more accurate results - ARIMA/ARIMAX or finding the distribution of the past data and building a probability model based on the said distribution.

I know until both the options are tried, one cannot comment on the accuracy. But for someone with the knowledge and experience having worked on such problems or someone who has the theoretical knowledge of both the above approaches, which is the better option.

I have only knowledge of OLS regression and ANOVA, so please excuse if the question is trivial.

Thanks, appreciate the help.

Regards

$\endgroup$
1
$\begingroup$

We always recommend using causal variables. There is always a cause and effect and if you can identify variables that explain the historical variability, you can then use the forecast of those causals to guide the forecast in a certain direction. ARIMA models implicitly capture the causal, but aren't as powerful. ARIMA is like driving a car using your rear view mirror and using causals adds the forward view using the street signs up ahead that tell you the road turns.

$\endgroup$

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.