A few months ago I've read somewhere that although the exponential GARCH model may lead to higher BIC values in comparison to other extensions of the GARCH family (GARCH, GJR-GARCH, TGARCH, ...), volatility forecasting under this model may lead to biased results. I'm trying to find a trustworthy source (an academic paper) which could back this, but so far I've found

  • this Matlab page which indirectly tells me that results are biased,
  • this 2010 working paper which hasn't been published in any journal, and
  • some bachelor/master theses that report this info without backing it with an academic source.

And as you may imagine, I can't (or better, I don't want to) quote other students' theses in my own thesis.

Do you know of any published work which proves this claim? Or a reliable source which suggests why the EGARCH shouldn't be used for forecasting?

  • $\begingroup$ In general, higher BIC values indicate inferior models. I think EGARCH has some poor properties; you could check the papers about log-GARCH models by Genaro Sucarrat, I think he contrasts log-GARCH with EGARCH and mentions what's wrong with EGARCH. Here and here you may find just a small remark that proving properties of QML estimator for EGARCH models is difficult and has not been done except for the most basic case. $\endgroup$ – Richard Hardy Aug 8 '16 at 19:09
  • $\begingroup$ Yeah sorry for that, I meant higher BIC values in absolute terms (being negative, of course I meant lower values). But thanks for the links here and in your answer below, I'll check them out! $\endgroup$ – Kondo Aug 8 '16 at 19:35

I found a couple of papers mentioning bias of EGARCH:

  • Demos, Antonis, and Dimitra Kyriakopoulou. "Bias correction of ML and QML estimators in the EGARCH (1, 1) model." Preprint (2010).
  • Deb, Partha. "Finite sample properties of maximum likelihood and quasi-maximum likelihood estimators of EGARCH models." Econometric Reviews 15.1 (1996): 51-68.

The first one is the same as you mention, so it is of no extra use to you. Probably the second one could be more useful. Unfortunately, I do not have access to it, so I cannot see what is inside.


This does nor address your question directly, but is somehow related

"We find no evidence that a GARCH(1,1) is outperformed by more sophisticated models in our analysis of exchange rates, whereas the GARCH(1,1) is clearly inferior to models that can accommodate a leverage effect in our analysis of IBM returns."

Peter R. Hansen, Asger Lunde, A forecast comparison of volatility models: does anything beat a GARCH(1,1)? JAE, Volume 20, Issue 7, December 2005, Pages 873–889


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.