AR and lagged dependent variables

[hmurphy94]

hi

i want to test the day of the week effect using both OLS and GARCH

for the OLS part, i see in some papers that they simply use a lagged dependent variable such as return(-5) for 5 week days. other papers use ar(1) , however does using ar(1) mean the model is no longer OLS?

[EViews Mirza]

OLS Is just an estimation method which may or may not produce consistent estimates. Although I'm not sure of which mostly in particular you refer to in your post, I am almost certain that the inclusion of an AR(1) term in your regression will not invalidate (in terms of consistency) the OLS estimator. Note, however, that OLS may be inefficient when dealing with GARCH errors.

[hmurphy94]

ok thanks

so i should run the OLS with ar(1) then after put it into the GARCH model? or do i need to include MA terms?

also, when using the ARIMA function for my return series, it says maximum ar and maximum ma options, should they be five as there are five days or should i keep it at default four? and should i use AIC or SIC to minimise?

thanks
hugh