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 used minitab software to fit time series model and predict. For some models I got the following error message;

"Fitted model may be nonstationary or noninvertible. Completion of computation impossible."

Using another package the relevant models were fitted.

What is the reason for that kind of error message and Is there solution to fix that error?

share|improve this question
1  
May I suggest you provide more information on your dataset and the kind of model you considered. This will avoid possible confusion as in your preceding question. – chl♦ Jul 5 '12 at 13:40
1  
5 years of daily exchange rates were used for analysis. According to ACF and PACF knowledge, mixed model (ARIMA) be the appropriate and several models were fitted to select suitable model. But, P value of Ljung-Box statistics is zero for all models. Since I didn’t get the correct model, R package was used and got the best model as ARIMA(5,1,2) using auto.arima() function. Though refit the model using minitab considering the obtained order the above mentioned error message was displayed. – Anthony Jul 5 '12 at 18:00

closed as off topic by Macro, gung, Andy W, mbq♦ Oct 15 '12 at 12:11

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 0 down vote accepted

The software that you used to identify the model form is flawed in many ways and it's conclusions have led to the minitab error. Model selection that expressly ignore Pulses, Level Shifts, Local Time Trends, possible changes in model for.parameters over time , possible changes in the variancee of the error term over time are in my opinion severely flawed. Don't blame Minitab for not being able to estimate an over-parameterized model. Model of the form (5,1,2) reflect overspecification due to the mistaken assumption that the standard error of the acf is = 1/sqrt(n). I suggest that you look elsewhere for the automatic identification. You might remember that some things (software) are worth what you pay for them.

share|improve this answer

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