From Schweser: “If coefficient a1 is statistically different from zero, the time series is ARCH(1)” “If a time-series model has been determined to contain ARCH regression, regression procedures that correct for heteroskedasticity, such as generalized least square, must be used… otherwise standard error lead to invalid conclusions” So does that mean if a1 is significant different from 0… the model is not useful? but on the next example, it has something like “we conclude that time series is ARCH(1), as such, the variance of error term of next period can be computed.” SO is it good thing to have ARCH(1)? if we correct them, we cannot predict the next period’s value…?
sorry…the question is a bit windy for me to contribute… can u press down for me to see how i can contribute
is ARCH(1) a good thing or not? --> it seems like its not bcoz we need to correct them --> but it seems to be a good thing because we can predict next period variance
ARCH is not used to correct the model. in another word, the estimated coefficient is still OK. then why ARCH? cuz you wanna know the % for Type I error, and reject the model when %>5% or 10% or sth. in that case, you cal. standard error, and derive t-test, right? But with hetero present, you can’t get the right rejection decision, because of WRONG parameter, which is hetero residual value. IF you are about to make the right decision, you need the correct se right? so then GLS comes in to help you find the right se and make the right decision whether to put the model on use. that’s it. Nothing to do with the coefficient…