The ACD Method

Maverick74 said:
In order to test this, you have to account for some alternative. In other words, had I randomly bought that same security over the 3 year period, what were the returns and are your results statistically significantly different then the null comparison. Otherwise the results are meaningless.
More...

Glad to hear I have may have been on the right track. When I test a system out, it needs to be able to kick randoms ass over my testing period with a small standard deviation to prove consistency of edge. This is not only in pnl terms but also with regards to drawdowns and the means of MAE & MFE and there ratios. Also I like to seeing if a particular strategy can work over multiple markets with different conditions to eliminate all possibility of curve fitting.
 
deltastrike said:
A small non-sequitur... but I'm pretty proud of this little histogram as it represents a decent amount of work. It's been said that ACD helps filter out the noise, and it appears I've found one way of doing that. This is over a three year period for this particular security.
More...

Delta, curious...you've only tested this on negative number lines?
 
koolaid said:
Delta, curious...you've only tested this on negative number lines?
More...

Not exactly, I wrote research function which tests all combinations of my selected filters and returns the one with the highest hit rate (hit rate = % change from day of signal for x periods). The histogram I posted is the result of one such test for just one security (long only), so when I test the basket I intend to trade I suspect I'll get different results for every security in the basket. Any critique of my process thus far would be appreciated.
 
It sounds like typical curve fitting and I think you will find you are wasting your time. I recommend reading globalarbtrader's material online or his book.
 
You must log in or register to reply here.
Back
Top