1

I wonder because the rugarch library on R allows to generate an ARMA-GARCH, but the coefficients of the ARMA are totally different from those generated by Arima from the forecast package.

I thought that the addition of the GARCH was only done by adding the GARCH, and not by re-estimating the model, am I right or am I wrong?

Richard Hardy
  • 54,375
  • 10
  • 95
  • 219
quezac
  • 33
  • 3
  • By coefficients, do you mean lag orders such as $p$, $q$ in ARIMA($p$, $d$, $q$) or "slope" coefficients such as $\varphi_1$ in $x_t=c+\varphi_1 x_{t-1}+\varepsilon_t$? – Richard Hardy Feb 13 '20 at 10:01
  • I mean "slope" coefficients, they're very different, and I keep the same lag orders. – quezac Feb 13 '20 at 10:54
  • This issue has been discussed multiple times before. See ["ARMA/GARCH estimation in sequence"](https://stats.stackexchange.com/questions/87600) and the linked questions in the panel on the right. You may also search for more such questions using relevant keywords. – Richard Hardy Feb 13 '20 at 11:26
  • Thank you, I'm sorry. – quezac Feb 13 '20 at 11:30
  • No problem at all. – Richard Hardy Feb 13 '20 at 11:37

0 Answers0