EViews.com

hi

i am currently doing a masters thesis on calendar effects on the shanghai stock exchange and so close to the deadline i think ive made a huge mistake. i am using OLS then using GARCH and showing how the results change. for the day of the week effect i ran:

ls return c gday return(-5)

i used the (-5) lag as i had seen it in another paper (stupid i know) so i ran it again using (-1) which worked. however my question is, do i use the ARMA model in the OLS regression and interpret the results, or do i run the OLS the way i have explained, then use ARMA to construct the GARCH model?

so in other words, is ARMA part of OLS do remove serial correlation, or is it just a way of removing serial correlation in order to make my equation suitable for GARCH?

when running the GARCH model i simply wrote into the command: arch(1,1) return c gday return(-5)/(-1)

thanks
hugh

The (-5) you saw in other papers probably is just controlling for the endogenous variable on same weekday of the previous week, and then trying to measure if, on average, the returns increase or decrease on a given weekday. The first thing you need to do is to check if your errors are serially correlated. If they are, adding an AR(1) term might be a good idea. GARCH models might be do that for you automatically, but you can certainly get to nearly stationary error terms using OLS only. Ideally, after you arrive at a stable model using OLS, you can present results of both OLS and GARCH specifications.