I'm going through my copy of Analysis of Financial Time Series, 2nd Edition, and I'm at the ARMA portion. One of the techniques for model selection is computation of the extended auto-correlation function, which indicates that, in a table of EACF's with MA coefficients listed across the top and AR coefficients listed down the side, the top-leftmost EACF that is less than the absolute value of 2 times the standard error of the EACF is at the position that indicates a good choice for the orders of the model. Can someone please explain to me exactly what the EACF is?
http://faculty.chicagobooth.edu/ruey.tsay/teaching/bs41910/lec9.pdf has some info. These are Tsay's lecture notes, and Tsay seems to be the co-developer of this measure.
Chapter 2 of Alan Panratz's Forecasting with Dynamic Regression Models contains a short discussion about the extended autocorrelation function. The book provides two references for the reader interested in more detail. You may like to track them down and give them a read.
Wei, W. W. S. (1990). Time Series Analysis, Redwood City, CA; Addison-Wesley.
Tsay, R. S. and G. C. Tiao (1984). "Consistent Estimates of Autoregressive Parameters and Extended Sample Autocorrelation Function for Stationary and Non-stationary ARMA models," Journal of the American Statistical Association, 79, 84-96.
By the way, the EACF is basically another tool, like the ACF and PACF, for identifying the order of an ARIMA model.