0
$\begingroup$

I'm trying to model how one time series A (transactions in an asset in some market) leads another time series B (transactions in a related asset). In this sense, I guess I'm trying to forecast B using lagged values of B and lagged values of A.

What tool seems to fit the description of what I'm trying to do? I was thinking ARDL models fit this kind of problem quite well, but I haven't found a ready-to-go implementation in Python.

There's also ARIMAX but from the descriptions it seems that this would be trying to predict B using lagged values of B and contemporary values of A. Is it possible to predict B using only non-contemporary values of A and B (is it as simple as just shifting A?) Lastly, it's monthly data, but going in, I'm unaware of the lag factor to be used.

Although I've read a bit around this site, very new to time series and appreciate all advice, cheers.

Improve this question
$\endgroup$
1
  • $\begingroup$ Have I answered your question? You have neither commented on it nor accepted. $\endgroup$
    – Richard Hardy
    Commented Sep 22, 2016 at 11:30

1 Answer 1

Reset to default
1
$\begingroup$

Is it possible to predict B using only non-contemporary values of A and B (is it as simple as just shifting A?)

Yes, it is. If you have a model $y_t=f(y_{t-1},x_t)$ you can define $z_t:=x_{t-1}$ and run $y_t=f(y_{t-1},z_t)$ and thus effectively $y=f(y_{t-1},x_{t-1})$.

Regarding the lag factor, you will have to experiment with it to see which works best, unless there is a theory implying a definite lag order and you wish to stick to the theory.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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