Dax adaptable strategy Breackout/Mean reversion
Forums › ProRealTime English forum › ProOrder support › Dax adaptable strategy Breackout/Mean reversion
- This topic has 82 replies, 13 voices, and was last updated 3 years ago by xburrex.
-
-
04/24/2017 at 9:16 PM #3324904/24/2017 at 10:43 PM #3325304/24/2017 at 10:47 PM #3325404/25/2017 at 11:40 AM #33310
have done some more work on this code, I noticed that the strategy was poor when ATR (or VOl) was low, so I thought to make a position sizing inversely proportional to the ATR.
Furthermore I created 2 scenarios,
1 ATR is bigger than x
2 ATR is smaller than x
Now the curve in the 1hr case looks steeper and smoother.
I attach the results.
Regards
3 users thanked author for this post.
04/27/2017 at 5:59 PM #3371604/27/2017 at 6:41 PM #3372505/02/2017 at 5:17 PM #34291Hey Francesco, following on from our discussion on your strategies page.
I tried the code you pasted on 200k units but the same story as 30min. Classic curve fitted picture. I’m not a fan of the seasonal code, especially splitting the months in to two parts but will keep it in there for testing.
I will try to make some time to run an IN/OUT sample on your variables on 200k and let you know the results soon.
1 user thanked author for this post.
05/02/2017 at 5:23 PM #34294Thanks a lot Cosmic1 for your follow up. Very kind of you to do the test on 200k bars.
There are a few variables in the code so I was expecting curve fitting to a higher then necessary degree.
I hope further discussion could help in finding ideas that decrease the amount of fine tuning and maybe we could come up with a more general framework.
Regards.
Francesco
05/02/2017 at 6:57 PM #34305Avoid overfitting:
Develop, code and tune your idea on 70% of the history. If it looks ok, do a final backtest on the last 30% of the history. IS/OOS should be your best companion 🙂
EDIT: By looking twice on the screenshot Cosmic1 provided, the final look of the equity curve depends a lot of when you start the backtest. Your strategy didn’t loose so much in the past, so please don’t throw this idea to garbage! 🙂
1 user thanked author for this post.
05/02/2017 at 9:46 PM #34348I agree 70/30 IN/OUT sample is your best friend.
I also agree not to bin it as some modifications could work here. It’s certainly worth a try.
I tried a quick IN/OUT on adxval, atrmin, m, and n. This did not work too well but I have to admit this is a quick and dirty test before jumping in to the code to fully understand everything about it.
Francesco, you will have a better understanding on what will and wont work. Can you try on 100k units but IN/OUT sample? Let me know if anything works for you? Then I can run a similar approach to 200k…
1 user thanked author for this post.
05/02/2017 at 9:49 PM #34349Absolutely Cosmic1, will work on it tomorrow afternoon. At your disposal for any question about the code.
Best
Francesco
1 user thanked author for this post.
05/03/2017 at 1:51 PM #34418Cosmic1,
I have tun a wf test 70/30 for 5 windows.
the results are quite good for the 100,000 bar case as far I understand as all the WF efficiency are well above 50%
I attach the results.
Maybe this strategy just doesnt work before 2014, but in this market I think it makes sense.
Please let me know your thoughts about it.
1 user thanked author for this post.
05/03/2017 at 1:54 PM #34421and the equity line..
05/03/2017 at 3:30 PM #34443I have also tried with EURUSD, the results are less good but still accettable.
Attached.
Regards
Francesco
05/03/2017 at 3:31 PM #34447 -
AuthorPosts
Find exclusive trading pro-tools on