Hi all, it says an autoregressive model is one which “uses a lagged dependent variable as an independent variable” but i thought in the previous section, this was one of the ways you would mis-specify a model, and therefor a big no-no? how come you can do it?
you can do this as long as there is no serial correlation, i think
^^ yep. i keep getting confused between autocorrelated and autoregressive. but, hey, i am dsylexic 
Dsylexic Wrote: ------------------------------------------------------- > ^^ yep. i keep getting confused between > autocorrelated and autoregressive. but, hey, i am > dsylexic haha if you truly are its a hell of an achievement to be writing level two.
I think autoregressive and autocorrelated are pretty close and for the CFA exams they are really close. An autoregressive model as described by the OP is one kind of model that you would fit if there is serial correlation. IMHO lots and lots of real world serial correlation is solved pretty easily by an AR(1) model. You can find a whole bunch of other kinds of models you could also fit with serial correlation including the whole class of ARIMA, (*)GARCH, spectral, etc. models. Fortunately for CFA exams you don’t have to know anything about those.
Autocorrelation is when the covariance between different error terms is different. i.e. think if you plotted the error terms they have some sort of a pattern->they are correlated That’s why in an autocorrelation model you have two OLS equations, one for the dependant variable y=b*x(t)+e(t) and e(t)=a*e(t-1)+e(t). In layman language “you’re doing two regressions”. The ARCH(x) model is similar to above except it says the error terms are squared this time up to a lag “x”. serial correlation=autocorrelation, i.e. substitute words for one meaning.
e(t)=a*e(t-1)+e(t) => a = 0…
Spotting! should be e(t)=a*e(t-1)+v(t)
GARCH models are touched on in the CFA curriculum