Your Account  •  Become a Member  •  Help  •  Search
 Forums ›› Technically Speaking ›› Strategy (System) Design ›› Exits a lot more complicated than entries

 Page 1 of 2:   1  2
 ronblack   Registered: Oct 2006 Posts: 569 06-15-12 08:51 AM In the link below there is an example of EL code for long and short exits (Adaptrade forum): http://groups.google.com/group/adap...e23c393c9?hl=en The long exit condition is a true monster. Is this an indication of excessive curve-fitting in your opinion? I've never seen such a complicated exit condition before. Edit/Delete • Quote • Complain
 braincell   Registered: Jul 2011 Posts: 538 06-15-12 10:35 AM Not neccessarily. If you could simplify the math of the exit tremendously and still have almost the same exits, then it might be a valid discovery. That's not always the best way to discover curve-fits, but it's possible that in this case it is. You need statistics. Edit/Delete • Quote • Complain
 zdreg   Registered: Oct 2003 Posts: 8330 06-15-12 10:40 AM is this thread talking about exiting the euro? Edit/Delete • Quote • Complain
 jcl   Registered: Jan 2012 Posts: 407 06-15-12 10:40 AM This is a machine generated pattern, and it looks indeed like the mother of all curve fitting. I think the problem is not that it's a complicated formula, but it seems to depend on price patterns from up to 84 days in the past, which makes no sense. A price pattern algo can not work with a pattern length of more than about 3..4 bars, because the autocorrelation of price series is zero beyond that horizon. This looks like a typical example of how not to use or not to program a price pattern generator. Edit/Delete • Quote • Complain
 braincell   Registered: Jul 2011 Posts: 538 06-15-12 12:33 PM I didn't spot the 84 days in the past. If that's the case, then yes, it's definitely a curve-fit and jcl is right. Edit/Delete • Quote • Complain
 intradaybill   Registered: Feb 2008 Posts: 2962 06-15-12 01:46 PM Quote from jcl: This is a machine generated pattern, and it looks indeed like the mother of all curve fitting. I think the problem is not that it's a complicated formula, but it seems to depend on price patterns from up to 84 days in the past, which makes no sense. A price pattern algo can not work with a pattern length of more than about 3..4 bars, because the autocorrelation of price series is zero beyond that horizon. This looks like a typical example of how not to use or not to program a price pattern generator. I agree with jcl on this one. It looks to me like this program draws vectors to future high points corresponding to entries and then fits patterns than make the entry profitable. Similarly to what someone would do in hindsight looking at a chart. One way to check these exit conditions for validity is to generate random entries and apply the exits to them. You will find out that after a simulation of about 100 different runs, the returns will average 0 or even turn negative. It looks pretty much like a geek forum there. Edit/Delete • Quote • Complain
 Page 1 of 2:   1  2