2. You can assume any error distribution for a model and compare the modelling/forecasting results. Choose the distribution that offers the best result (suggested by ur evaluation criteria) because the distribution discribes your data best.

3. R-square doesn't matter much to the variance model.

cheers

thanks for your help. then how about the first question?

是的，三者的和必须小于1。

你好，最近我也在做GARCH族模型，那为什么高铁梅和易丹辉中TARCH的例子三者和都是大于1的呢？

我也想知道这个问题，请问你解决了吗