1
$\begingroup$

I have two time-series and I need to make a model that takes the input of one of the time series to estimate the value of the other time series. I've tried a lot of different specification of ARIMA and haven't found one that works for me yet. Also, I will have to do this multiple times with different data sets and I don't want to have take time to manually specify a model each time I do this.

Below is a graph of some data that I have. I need to take the green line and model the blue line. You can ignore the grey line.

enter image description here

I have little experience with machine learning, but from what I understand, some techniques can help specify models. Any thoughts about techniques to solve this problem. FYI I have access to SAS enterprise miner.

$\endgroup$
0
$\begingroup$

It helps if you show what the ARIMA model looks like and what the problem is specifically. But just going with what you have, 1) try harmonic seasonal terms, 2) have you tried forward-chaining to find the model or are you looking at information criteria only? 3) have you tried BSTS models? 4) You can also use synthetic control methods (available in R and Stata, but I recommend the R package)

$\endgroup$
1
  • $\begingroup$ I tried a couple of models, the one that worked the best had a one day, 7 day and 30 day lag. But the residuals were showing strong patterns in everything that I tried. I need to research the three solutions you mentioned as I am not familiar with them. I'm looking at information criteria only. Thank you very much for pointing me in the right direction. $\endgroup$
    – Jarom
    Jul 13 '17 at 18:56

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.