GARCH Models in Python

fyjyuuidrtrst-

Overall, I enjoyed the course and found the explanations clear and the exercises helpful. However, I have one significant concern regarding the modeling approach taught in the section on ARMA-GARCH.The course demonstrates a two-step procedure: first fitting an ARMA model to the mean, then applying a GARCH model to the residuals. While this is computationally simple, it is theoretically inconsistent with standard financial econometrics literature (e.g., Ruey S. Tsay's Analysis of Financial Time Series).The correct approach should be joint maximum likelihood estimation, where the ARMA and GARCH parameters are estimated simultaneously. Two-step estimation ignores the fact that GARCH effects influence the efficiency (and sometimes consistency) of the mean parameter estimates. It can lead to biased standard errors and suboptimal parameter estimates.I understand that Python's arch package does not natively support ARMA mean models, which may have motivated this workaround. However, it would be helpful to at least mention this limitation and explain that in practice (or in other languages like R), joint estimation is the preferred method.Thanks for creating such a valuable learning platform!