4
$\begingroup$

Right now I am working with vector autoregressive models in order to make 3 months forecasts for a commodity good (sawlogs) y. I have several time-series of "follow-up-products" of sawlogs that should work as "predictors" for saw-log prices from a logical point of view. I encountered within the VAR-function from package "vars" (http://cran.r-project.org/web/packages/vars/vars.pdf), that one attribute called "type" has the following expressions: "const", "both", "trend", "none". I really don't know what this means from a statistical point of view.

Since neither the package-description nor other literature I've screened so far can give me an answer I actually understand I'd like to ask you guys the following:

How should I interpret/understand and use the argument "type" in R's VAR() Function?

What do those 4 different arguments really mean? "both", "none", "trend", "constant"? Could anyone explain this in a simple way and probably provide an example as well?

Does this mean that I can directly use non-stationary time series for my VAR-model since I can consider trend/season afterwards by setting the "type-argument" to both, or am I wrong here?

Improve this question
$\endgroup$
0

1 Answer 1

Reset to default
5
$\begingroup$

constant means there is an intercept included in each equation in the model. trend means there is a linear time trend included. both means both of them are included. none means neither of them is included.

Type ?VAR in R to get an explanation. There you will find the algebraic form of the model provided. Pay attention to the CD_t term. You may also see the vignette for the vars package, perhaps it could help.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Thanks for your answer Richard. I read the vignette and didn't really get smarter afterwards. Sometimes it is stated that the data has to be stationary BEFORE usig a VAR model, then I ask myself how there even could be a trend? Does this "type"argument imply that you can directly use non-stationary data within a VARestimateion and then set "type" to "both" in order to consider trend and season? $\endgroup$
    – George
    Commented Nov 19, 2014 at 15:26
  • 1
    $\begingroup$ If the non-stationarity is caused by a deterministic time trend (for example, a linear time trend), it can be treated by including a time trend regressor in the VAR model (option type="trend"). If, on the other hand, non-stationarity is due to the process being integrated (e.g. I(1) process), this cannot be cured in such a simple way. By the way, both is not about seasonal component. There is another option for that – it has a name season. $\endgroup$
    – Richard Hardy
    Commented Nov 19, 2014 at 15:38

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.