- This topic has 286 replies, 38 voices, and was last updated 1 year, 6 months ago by
.
-
Posts
-
Hi Petko,
Yes, you’re absolutely right. I just opened 2 browsers and ran an M1 and a D1 reactor for 24 hours. At the end of the day, the M1 reactor had calculated around 500,000 strategies and the D1 had calculated more than 11 million.
So yes, now it makes sense. Thanks for clearing it up Petko.
Hello,
This is my acceptance criteria. But I am looking forward for some feedback. I’m pretty new to EA studio (4months).
COMPLETE BACKTEST:
min back test quality: 98
min r-squared: 60
min count of trades: 200
min stag %: 30
min equity dd %: 20
min prof factor: 1.2
IN-SAMPLE:
min prof fact: 1.2
min r-squared: 60
min backtest qual: 98
OOS
min prof fact: 1.2
min r-squared: 60
min backtest qual: 98
I been focusing on GBPUSD: 15M AND H1.
This are my settings on my acceptance criteria and validation and looking for some feed backs:
Running on Reactor:
Data Horizon – 02/01/2018 to 06/11/2020 (14,706 bars) data source from FS DUKAS
S/L: 50-100(ALWAYS USE/FIXED)
TP: 10-50(ALWAYS USE)
GENERATOR SETTINGS:
OOS: 30% ( GEN STRATS W/ PRESETS and USE COMMON ACCPT CRIT)
WFO SETTINGS:
SEGMENTS: 5
OOS 30%
+- 10 STEPS
OPT PRESET IND
VALID SEG 5
MONTE-CARLO SETTING:
20 @80%
default settings for validation (im looking to change this, but i just cant think of a good setting for validation)
I hope this info help some people build a basic..
lately I have been messing around min count trades on acceptance criteria and sometimes I use it to filter and or sometime use a lower min count trade.
Im really looking forward for some feedback to improve my validations/acceptance criteria for reactors/wfo
Hello, and welcome to the Forum!
I think you are doing a pretty good job and you realize what you are doing.
However, the min count of trades are small, you need to keep at least 300 as a min.
If you tried and you fail to get any strategies it is because you have a small number of bars?
If you use the Free Forex Historical Data app, there are more bars, why would you use less than 15k?
Also, you may want to try generating EAs not just for the GBPUSD because this will bring you more diversification.
Cheers,
What I recently use for my acceptance criteria is:
PF > 1.4
A number of trades 700.
I noticed that if I use Walk Forward Optimization in the Reactor it finds more strategies with bigger Count of Trades.
Did anyone notice that? It works great by the way!
I have noticed the same. I guess because it divides the backtest period into zones, and in each one, it requires min count of trades.
However, what I wanted to share recently is that I use more and more the R-squared and the tests show that if I generate EAs with above 80 as a value, the strategies perform pretty well after that.
Recently I noticed that we don’t need to have too many bars to get more count of bars. With the default 50 000, you can still get 300 to 500 count of trades.
And this way we get more active strategies. That is what I Explain in the upcoming trading course.
Hello Petko!
I realized what you said in this last post after watching the course 🙂 Amazing!
I did many tests and right. The generator still produces strategies within my acceptance criteria of 500 trades and 1.3 PF.
You are always right! It is so evident that if we have fewer bars and we still want 300 to 500 trades as a minimum, we would have more active strategies. I didn’t think it is possible.
Now comes the question: How much should we decrease the number of bars? 🙂
Hey Andi,
It seems I missed your kind words, mate! 🙂
Well, I do not go below 100000 bars on the lower time frames. It depends on the symbol but even we have 20 000 bars if the strategy shows 500 count of trades and a good Profit Factor, passes Monte Carlo, and is not over-optimized, then all is good.
Hi,
I am Very new to EA Studio, I am unable to generate even one strategy! please find the screenshot below and anyone please correct me what is wrong?
Just my opinion….. You are using to much filtering. I would drop any optimisation and normalisation. Also Just stick with the monte Carlo for the robustness testing. I would also always use a SL and TP. After 2 years of using EA studios and probably a 1000 + hours invested in this program and with the help of Petko I have realised it is best to set the parameters lower, get lots of strats and filter them out on demo before putting them on live. Currently I have taken this to the extreme by creating about 50000 strats a week on the M1 and then pre-filtering them down to 100 / 200 with the prior weeks data and then running them on demo and doing a final filter down to 30/40 strats to run live. Again this is just what is working for me. Watch some of Petkos videos on the filtering of starts, I believe this is key…… Good luck.
Also…. Once you have got the reactor sorted. Take time to learn the validator. This is a very powerful part of the program that will allow you to input 1000’s of strats and “Run them forward” on say the last weeks / months data. I use this as a pre-filter before I take them to demo for a final filter (For example wait 3 trades and then any strats above PF of 2 transfer to live)…..
Hi richard, what default Acceptance Criteria’s do you use to get that huge amount of strats per week? How many bars do you use? Do you use the Generator or the Reactor? Many thanks for sharing your knowledge.
I only go back 8 weeks on the M1. I use the reactor. I leave out the previous week as I use this as a pre-filter. So actually start back 9 weeks. I use COT = 100, PF = 1.2, Max losses = 3. I don’t use any robustness or optimisation. I just want raw strats as I run my own form of robustness tests later by comparing many slightly different versions of my core strategy. I run 12 instances of EA studios at the same time. 6 on each PC. It normally takes 6 hours to get 300 strats on each EAS. So about 4 or 5 days to get 60000 strats. Seems like an overkill I know……. I have done many tests with different amounts of strats up to 60000 and have found that the more I have the lower the DD and the higher the PF becomes. I am aware that I may be curve fitting the strats by having that many so am watching for that. After I have the 60000 I run them forward on the previous weeks data in the validator. With much stricter criteria: COT = 10, PF = 2, Max losses = 1. This shows me how those 60000 strats would have run “Live” for that week. There is normally 300 strats that pass this criteria. I then select the top 100 / 200 and create a portfolio. I run them on demo. I use a custom made filtering app that does the final filtering and sends the ones that pass through to a live account. Normal settings are something like wait 3 trades, if 2/3 trades win with a PF of 2 and R2 above 50% they go to live. I monitor them in the live account and if they start to drop below my criteria they are removed.
I am always continuously testing. So if I have an idea that I want to test I usually run about 10 slight variations of it side by side for a month or so and then evaluate and move forward with the one that is working the best. This way I keep narrowing down what works for me.
After lots of testing with the M15 and the H1 I found I just could not get consistent results. Not to say it can’t be done, just doesn’t work for me. The markets have also been pretty crazy for the last year and there is potentially massive changes ahead to the fiat currency system as we know it….. So i want to trade a TF that can adapt quickly to changes and not look back to far in the past. Again this is just what I am comfortable with…..
I have also narrowed it down to one particular pair that performs well and I also only run my EAs on a 4 hour window each day when the markets are most active and the markets are not crossing over from one another…..
I still have a way to go but the results are improving and getting more consistent as I go…..
I am working towards opening an account with FTMO and some other prop firms in the next month or so. So I have been trying to create a system that stays with the tight requirements of the prop firms….
Good luck and just keep testing and practicing……
- You must be logged in to reply to this topic.