You have not presented enough evidence to support your conclusion. You really need some idea of tail stats for the distribution of results of repeated 277-trade samples drawn from the 1586 population. Building a bootstrap distribution would be appropriate but not trivial due to multiple entries, serial correlation amoung overlapping trades, heteroscedasticity (high vol periods mean more quickly completed trades so less autocorrelation), etc. A simple one-sided Chi^2 (Fisher Exact), which would overstate Harris's case, 65% vs 61%, 277 in the sample, fails to reject the null of "could have been obtained by chance" by a wide margin. I would love to see a properly constructed block bootstrap p-value estimate from you or Code7.
I'm not saying I would trade this system. Actually, its Sharpe ratio is below 1 depending on how you do position sizing. What I said is that its performance was better than the equally weighted bars and based only on that criterion it was better than chance. As you know, one can assume chance in everything if counterfactuality is questioned. Neverthless, you sound very good with statistics, much better than me, so be my guest and analyze the report from Amibroker (excel format) http://www.4shared.com/file/xq2CSUkN/Harris_aps_QQQQ_system_trades.html Edit: I have gotten much better systems from APS with Sharpe > 4 and pf > 3 so this system is of no interest to me. I did this analysis just from curiocity.