A very short question. Am I correct in thinking that when fitting an ARFIMA(p,d,q) model, one first fits an ARFIMA(0,d,0) to estimate $d$ and then fits an ARIMA(p,0,q) on the residuals of the ARFIMA fit to estimate p and q?
If p or q in ARFIMA(p,d,q) are not 0, then the residuals of the ARFIMA(0,d,0) fit should exhibit this autocorrelation because ARFIMA(0,d,0) removes the long-memory dependence only.
