The RATS Software Forum

Hello,
I have 5 series of data which consist of 5-year bond yields for 5 different countries. I have estimated an EGARCH(1,1) model for each as well as the GJR-GARCH(1,1). I would like to test which of these 2 models does a better job at matching the data. I found a procedure by Engle and Ng (1993) however, can't figure out how to implement it in RATS.
Thank you,

If you're talking about "Measuring and Testing the Impact of News on Volatility", Journal of Finance, 1993 (please use full citations), I don't think that has any test for GJR vs E-GARCH. It has tests for GARCH against E-GARCH or GJR or other models which allow asymmetry.

Thank you for your response,

I'm in fact referring to this exact paper. I understand how they establish the ratio and formally test whether or not the EGARCH model adds value. Basically, they test whether they can reject the null hypothesis corresponding to the usage of a GARCH(1,1) model.

However, I wondered whether there is any formal way to test my hypothesis. My series exhibit strong signs of asymmetry and as a consequence my hypothesis is that a model such as the GJR-GARCH of the EGARCH would provide a better fit to the data. However, having estimates for both of these I don't know how to choose which one is best. However, results Ljung-Box statistics are much better for the EGARCH(1,1).

Thank you very much,
Alan

When I skimmed that paper looking for a "test" for EGARCH vs GJR-GARCH, I noticed they had at least one model where they clearly favored one of the two over the other based upon log likelihood (which should be comparable) and some other characteristics. There are non-nested tests for non-linear models like the Vuong test, but I would guess that it will likely fail to reject either in favor of the other.