June 16, 2020 at 11:19 #51206
I am failing to generate strategies don’t know why? I tried on 2 assets XAUUSD & EURUSD. I keeping getting message of Acceptance Criteria too strict etc. I am following exactly the steps recommended by Petko and I list below the same:
1. Historical Data – I am taking from FS Dukas copy- adapted to my broker specification. M-15 Time frame.
2. Strategy properties – SL – always use; fixed or trailing; Range 10 – 100; TP – always use; range – 10 to 50 pips
3. Generator setting – 300 mins; Net Balance; 20% OOS. Common Acceptance Criteria – Min net balance 10; Min count of trades -300, Min profit factor – 1.2; In sample (training) profit factor 1.1; Out of Sample (trading) profit factor 1.1;
4. Back Test output – Net profit; Profit factor; R squared; Max stagnation; count of trades;
5. Collection – I have no performance filters; only resolve correlation automatically.
6. Data – Max Bars – 200.000; Min Bars – 300; Not using any date range.
I am not even using reactor because at generator level itself I am not generating strategies or sometime one or two! What mistake am I making? Help please. It seems for XAUUSD a different approach is required – Petko can guide please. But why am I failing? What other possibilities should I be experimenting with to generate strategies?
June 16, 2020 at 22:50 #51253AndiMember
I think you are missing something with the historical data. Could you please attach a screenshot so we can see better?
For the Gold, you need to place a wider range for sure. Try 10-10000. And see what values will be there in the strategies that are generated.
Usually, the gold is with different count of digits with the brokers. Usually, it would 200 pips= $20. But it might be different.
Strange you don’t get strategies for EURUSD….you can test if the issue is with the broker. Just remove the acceptance criteria, and generate. If still nothing, it is something from the Historical data. If it works, just start adding acceptance criteria slowly, one by one and you will see what stops you…
June 19, 2020 at 19:07 #51333
I followed your advice. I did a sample run of the generator for 300 mins; Net Balance and In sample without any acceptance criteria and without any filter criteria in collection. I used FS DukaCopy database for M15 and it loaded 200,000 bars from 01/06/2012 to 12/06/2020. I adjusted the specs to that of my broker for EURUSD for spread @ 20 points (although the actual spread was 16 points), Swap day to Wednesday and swap rates for long and short. I noticed something interesting – it generated very quickly 100 strategies within first 5 and 10 minutes and then stayed at that right up to the end. Of the 100 strategies, 98 strategies were correlated. If I checked the ‘resolve correlations automatically’ I am then left with only 2 strategies! Collection record stats read – Pushed total strategies -4012, Added Strategies 4021; Resolved correlations -0 ; Pruned excessive strategies -3921; Correlated strategies – 98. I am yet to check out XAUUSD as per advice and will do so. Look forward to your advice as to able to generate strategies. Somewhere something is amiss. The problem it seems that far too may correlated strategies are being throughput and why after 100 the collection doesn’t collect any more strategies I don’t know. I don’t think it is the historical data but you know better 🙂
June 20, 2020 at 14:25 #51339Desi IvanovaModerator
Actually, you need to check the box with Resolve Correlations before you run the Reactor. Not after that.
The generator generates strategies all the time and pushes the ones that pass the acceptance criteria in the collection.
The maximum is 100. When this number is reached, it keeps pushing, and in this 100 will come just better and better strategies until the end.
So they are not the same strategies staying in there.
I think what happened in this case, is that the generator found a great strategy with many correlated versions and non of the rest succeeded to show better results. When the generator was over, and you clicked to resolve the correlations, obviously you will be left just with a few.
So try again, but first, check the Resolve Correlations and after that run the reactor.
Also, check the tools if everything is fine.
June 22, 2020 at 20:54 #51649
Ok, followed your advice to obtain limited success but don’t know if it is adequate. I will present my findings. I used the Reactor with Monte Carlo validation. Details as under:
Data Source: FS DukasCopy; USDJPY; M15; 200k bars; Did not use date. Strategy properties: SL – always use; fixed or trailing; Min pips 10 – Max pips 100; TP – Always use; Min 10; Max 100 pips;
Generator setting – 600 mins; search best – Net Balance; OOs 20%; Common Acceptance Criteria – Min Net profit – 10; Min count of trade – 500; Min profit factor – 1.2; In sample (training) min profit factor of 1.1; OOS (trading) 1.1;
Optimizing strategy – None. Perform Robustness testing – Monte Carlo validation.
Collection – No performance filters; Only ticked resolve correlation automatically.
Tools – General Settings – ticked both start backtest from 100 bars and close the position at the end of the back test.
Correlation Analysis threshold at 0.98; and ticked both detect balance lines correlation and detect strategies with similar trading rules.
Settings – Account currency USD; Initial Amount – USD 3k; Leverage 1/200; Trading session – everything is at 00:00 to 24:00 and do not trade on Sunday; ( just for info – my broker does not trade on saturday either);Back test output metrics – Net profit; Profit factor; R-squared; Max stagnation and count of trades;
Data horizon – Max data bars – 200k; Min data bars – 0.3k; No dates were used
The outcome is – Generated 91,643 strategies; Passed validation – 3 (0.00%); Monte Carlo validation – Calculated strategies 3(60); Passed validation 2 (66.67%). Please advice if these results are in keeping with what is to be expected? Personally I feel they are too less. Leave me with very little scope to even have a portfolio! What I observe from Petko’s training videos is that always lot more strategies are generated. I don’t know what mistake I am doing anymore! Clearly something is amiss, or so I think! While my patience is tested, I do believe am doing something wrong – hoping someone will guide me through as patiently 🙂
June 23, 2020 at 8:41 #51783Petko AleksandrovKeymaster
That is great info that you provide. Usually, the traders ask me questions with ver small information, and I just need to guess what they are doing 🙂
OK, from what I see, first, you are doing a great job! I think you understand clearly what you are doing.
Here is my opinion what might be stopping you from having more strategies in the collection:
1. The SL with option trailing. Give it a try with fixed. I don’t know why, but for me, it works better.
2. You have to lower the acceptance criteria. For example, just OOS 20% + Profit Factor 1.2 and Count of Trader 300. 500 is a lot for some currency pairs.
The process to figure out what is stopping you is the following:
1. Run the Reactor without any acceptance criteria(no monte Carlo) Just generator.
2. Start adding some criteria. For example, Profit Facto 1.1. See how it goes for 2-3 min. Add Profit Factor 1.2…it will start filtering more and more the strategies. Keep adding until you see a decent amount of strategies.
BUT For 600 min if you see 10 strategies. That is good enough. Sometimes I run multiple times the Reactor for 600min before I get 20-30 strategies. And that is the better thing, rather than having 100s and to waste time on Demo testing.
- You must be logged in to reply to this topic.