Methods for evaluating partial autocorrelation for identification of ARIMA models

Question

I am trying to programmatically identify an ARIMA model for a series of data and forecast values.

Currently the problem i am facing is to find a way to evaluate partial autocorrelation. I have been looking for methods to calculate PACF for quite a long time now but in vain.

Please provide some online resources which can help me in this matter.

Thank you.

Answer 1

Use the Durbin-Levinson algorithm. It will be explained in any good time series book such as Brockwell and Davis. Here are some online explanations:

• http://www.stat.ufl.edu/~berg/sta4853/files/sta4853-4.pdf
• http://amath.colorado.edu/courses/4540/2008Spr/HandOuts/DurbinLevinson.pdf

Answer 2

You may also be interested in the 'auto.arima' function in the forecast package for r as an example of a way to programmatically identify ARIMA models. It probably doesn't find the model in the exact same way you would, but some of the code/ideas might be useful to you.