Hello everyone,
I was studying the GJR-GARCH and EGARCH models and realized that the Conditional Log-Likelihood function is non-smooth, since the parameters that must be estimated will eventually branch (due to a conditional if statement in the case of a GJR-GARCH model) or lie inside a modulus (in the case of a EGARCH model).
My issue is that the Econometrics Toolbox uses fmincon with the 'sqp' algorithm, which is supposed to be designed for smooth problems. What am I missing, shouldn´t fmincon be inappropriate for the purpose of GJR-GARCH and EGARCH estimation?
Thanks for your help!
Duarte
Best Answer