2
$\begingroup$

I'm using time series data containing both trend and seasonality. I also have 2 endogenous predictor variables that I would like to include in my model.

In R I've used the forecast package to develop a dynamic regression model with use of auto.arima() and the xreg argument from the forecast package. I understand this procedure takes a regression and then attempts to fit the residuals with an ARMA Model.

I've also developed what seems to be an appropriate model using the forecasting Module in SPSS by specifying a Seasonal ARIMA model and including my covariates. However, one of the coefficients on one of my endogeneous predictors has a negative sign which makes no sense intuitively.

I've read Dr. Hyndman's article The ARIMAX model muddle and found it to be extremely insightful and useful. However, I have not been able to find any documentation on what type of statistical procedure SPSS uses to fit an ARIMA model with covariates, so I'm not sure how I should interpret the coefficients or how concerned I should be with a flipped sign. Any help clarifying the modelling procedure used by SPSS would be tremendously appreciated.

$\endgroup$
  • $\begingroup$ You say I understand this procedure takes a regression and then attempts to fit the residuals with a linear regression but that's probably a typo since it is ARMA model that is used to fit the residuals. $\endgroup$ – Richard Hardy Nov 4 '15 at 20:07
  • $\begingroup$ That was a typo; I've updated the question. Thank you. $\endgroup$ – Rick Arko Nov 4 '15 at 20:48
2
$\begingroup$

See the Algorithms doc available from the Help menu for computational details. If you have Statistics version 23, you might also be interested in the TCM procedure (Analyze > Forecasting > Create Temporal Causal Models)

$\endgroup$
  • $\begingroup$ Thank you. It appears as though the series of predictor variables are added to the transfer function model. (I would be interested to know if there is a way to automatically fit an ARMA process to regression residuals as the forecast package in R does.) I read over some material on Temporal Causal Models and haven't decided if I'll be able to make use of that functionality yet. Thanks for the suggestion. $\endgroup$ – Rick Arko Nov 5 '15 at 22:38

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.