@Mohammad
If my understanding about Chan and Prof Shum's paper is correct, this algorithm might need a list of leveraged ETFs for re-balancing. I think the best way to implement this algorithm is to sort a list of ETFs and find the max gain/lost at 2:15pm. Then long/short the best/worse ETF in the list.
Here is my list
context.stocks = {sid(32270), sid(38533), sid(35021), sid(39215), sid(32272), sid(39214), sid(33270), sid(33263), sid(33265), sid(33268),sid(33261), sid(33264), sid(33262), sid(33266), sid(32268), sid(27846), sid(32382), sid(38532), sid(32267), sid(32384), sid(39212), sid(32265), sid(39211), sid(33278), sid(41576), sid(39369), sid(33279), sid(41574), sid(33276), sid(36386), sid(33272), sid(36385), sid(33277), sid(33275), sid(39377), sid(33274), sid(33281), sid(33271), sid(33280), sid(33273)}
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2161057