Vector ARIMA model is used in multiple time series analysis. I am just wondering if there is any software or tool can be used to build the model. Some tools,like R, can only be used to predict the value after the Vector ARIMA is already built by the user. So is there any tool can create the vector ARIMA model? Thanks in advance!
$\begingroup$
$\endgroup$
2
-
$\begingroup$ By "create the model", do you mean "estimate the model from observed data"? $\endgroup$– SilverfishCommented Dec 19, 2014 at 9:42
-
$\begingroup$ I have no experience with it, but take a look at the 'marima' R package (Multivariate Arima) on CRAN. $\endgroup$– mfrickCommented Oct 28, 2020 at 9:25
Add a comment
|
3 Answers
$\begingroup$
$\endgroup$
There are multiple R packages which can be used to calculate the model. 'dse' has multiple options included in the 'est' range of functions which given the data you are attempting to model using a VAR etc can calculate a range of models given your specific requirements.
Sorry about making this a separate answer but i cannot comment until i get more rep.
$\begingroup$
$\endgroup$
I don't know what is the industry practice, but I'd start with looking at state space model (SSM) representations of ARIMA processes, such as this paper.
SSM tools are available everywhere even in MATLAB, in R there's astsa package, which handles not only normal models.
$\begingroup$
$\endgroup$
6
You can simulate a variety of time series (VAR, VARX, VARMA, state space...) with the function simulate from the R-package dse.
answered Apr 20, 2014 at 14:40
-
$\begingroup$ I know the dse package, but it seems to me that we need to input the parameters of Vector ARIMA model by ourselves, then the function only simulate the model, but NOT find the model(fit the model,find the parameters). May I right or wrong? Thanks! $\endgroup$– Hao YuCommented Apr 20, 2014 at 14:55
-
$\begingroup$ If you want to simulate, or, create the vector ARIMA model, you have to specify the parameters. Maybe one could somehow randomize the parameter-specification, but why not just supply them? Then you get data and you can the fit model and estimate parameters, and look, if the routines find the "correct" model. $\endgroup$ Commented Apr 20, 2014 at 15:03
-
$\begingroup$ in R, we have another package--forecast, which can do ARIMA(univariate) model, and it can automatically find the model by using auto.arima() function, and then forecast the future values by forecast() function. I just want the similar functionality for Vector ARIMA model, since we don't know how to set the parameters p,d,q and the coefficients, which is normally estimated by least square estimator. $\endgroup$– Hao YuCommented Apr 20, 2014 at 15:13
-
$\begingroup$ Does it have to be VARMA or does VAR satisfy you? $\endgroup$ Commented Apr 20, 2014 at 15:15
-
$\begingroup$ without doing difference firstly, the VAR model may not be the correct model. I have tried VAR model, what I feel is I need to do difference firstly. $\endgroup$– Hao YuCommented Apr 20, 2014 at 16:17