I need to fit a ARIMAX model (in R, library: TSA), with variables like

$Y = $ time series that I want to predict/forecast

$X_{i} =$ exogenous quantitative variables

$D_{j} =$ qualitative variables (dummies)

As far as I know, the ARIMA parts of ARIMAX function only refers to $Y$, and adding $X$ and $D$ captures just a "one day" effect.

Is it right?

Is there any way to surpass that issue?


Imagine that

$D_{2.k} = 1$ if event $A$ happened;

$D_{2.k} = 0$ otherwise; $\forall k$.

I know that when $A$ happened, at day $t$, $Y$ could have been affected from $t$ to $t+p$; i. e., a day $t$ event splashes $(t+1), (t+2), \dots, (t+m)$.

Same thing about all $X$ and $D$.


If your predictor variable is stochastic then you can employ Transfer Function Identification in order to form and encode a lead , contemporaneous or lag effect of one or more a predictor series.

This is largely correct except for suggestions regarding model identification.

Also see https://onlinecourses.science.psu.edu/stat510/node/75/ and @forecaster's excellent comments here Transfer function in forecasting models - interpretation and a lengthy discussion here steps to time series analysis on my data

If your predictor variable is not stochastic then all I can suggest is a trial and error approach incorporating different dummies (0/1) .

| cite | improve this answer | |

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.