Tell me more ×
Cross Validated is a question and answer site for statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
up vote -1 down vote favorite

I have four variables and would like to construct a VAR model I would then like to make a VAR forecast on one of the variables using my own data for the forecasts of the other three variables. Is there an R package or process from the model fit that allows me to do this?

share|improve this question
This appears to be only a question about how to do this in R, rather than about the statistical issues associated with VAR forecasts. If this is not true, please edit your Q to clarify the statistical issues you are wondering about, if it is true, this Q is better asked, & probably answered, on Stack Overflow rather than here. Please do not cross-post (SE strongly discourages this), if you want your Q migrated faster, flag it for moderator attention. – gung Sep 5 '12 at 18:39
1  
What is the point of a VAR in this case? Why not just use a univariate ARX model that incorporates the contemporaneous and lagged values of the other variables as well? – John Sep 5 '12 at 18:41
I would like to look at a variety of scenarios in the other three variables and see how it effects the forecast. – adam.888 Sep 5 '12 at 19:15
I think the VAR is there because poster knows the series aren't really exogenous, but wants to see predictions conditional on rather specific scenarios for some of them. Whether that's a good idea is a separate question. – conjugateprior Sep 5 '12 at 20:43
Thanks for the univariate ARX suggestion. – adam.888 Sep 6 '12 at 10:17

closed as off topic by gung, Macro, Andy W, Peter Ellis, csgillespie Dec 1 '12 at 15:51

Questions on Cross Validated are expected to relate to statistics within the scope defined in the FAQ. Consider editing the question or leaving comments for improvement if you believe the question can be reworded to fit within the scope. Read more about closed questions here.

1 Answer

up vote 2 down vote accepted

R packages to do VAR modelling are listed in the Time Series View.

To see the impact of shocks to particular series, you'd normally compute impulse-response functions. All VAR implementations should do for you, e.g. in vars.

However, explicitly fixing the path of one series is a bit different, and perhaps also a bit odd because by doing so you are treating that series as exogenous, whereas one normally fits a VAR because the series are considered to be endogenous. Normally one applies VARX models for the first sort of assumption, but I think that's probably not what you have in mind for scenario planning.

That said, you could probably treat the problem as an sequence of one step prediction problems in AR rather than MA formulation with a fitted VAR. You'd have to write a function to sample a new step forward for all other series at each step, conditional on where you wanted the controlled one to be. If you do that enough times and summarise it you'll get a conditional forecast.

So, the answer to your question is: no, I don't think so. And the reason there isn't (so far as I know) is given above.

share|improve this answer
Thanks very much for your helpful answer. – adam.888 Sep 6 '12 at 10:16

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