- This topic has 156 replies, 37 voices, and was last updated 4 days, 12 hours ago by
.
- Posts
I use 🙂
Maximum equity drawdown %: 10-15% (Depends on FX 10% or Crypto 15%)
Minumum Trades: 50
and Sometimes:
Maximum stagnation %: 30%
Minimum return / drawdown: 3
@dommech-Hey there! Good question. My acceptance criteria is:
Min count of trades: 10
Max equity drawdown%: 20
Min net profit: 10
Min backtest quality: 98
Min win/loss ratio: 70
Max stagnation%: 30
and sometimes Min profit per day: 1 *Of course this depends on the lot size being used for generation of strategies.
If I can’t get the generator to produce strategies with all of these requirements, then I just run it without acceptance criteria and use these as the performance filters. I use the criteria for all assets that I’m trading. After the the strategies are all created, I then do some additional analyzing to pick out the best strategies to use such as using the optimizer, turning on the stop loss or take profit, etc.
Hello Dommech,
something important from me here:
Min count of trades:50 is too small number. Specially in Forex you should go with min 300-400 count of trades.
Basically when the generator/reactor works it looks for combinations between different entry and exit conditions, indicators with different parameters etc., so when you have strategy with only 50 count of trades this might be not a stable strategy. We want to see bigger count of trades so we know that this combination of indicators, parameters, SL, TP and so on, works on more trades(longer period of time).
I really like the win/loss ratio, but it is hard for me to get results on all pairs when I keep it on 70.
In the acceptance criteria for min count of trades, Petko is right, I even try to keep it above 600 trades.
@Petko and @Haliffa-Does the chart timeframe that we are using affect the count of trades? For instance, I can generate a lot more count of trades on the M15 chart vs. the H1 chart. But then again, I get into the issue of wanting that 70% winning performance and only getting a smaller count of trades. I wonder if this is the trade-off (higher winning percentage=smaller count of trades vs. lower winning percentage=larger count of trades).
Since I don’t want to give up on that winning percentage, I may be working with short-term strategies for quite a while.
Thx for the answers guys!
Some strategies with especially extremely low DD (0.5-2%) I’ve found that they are extremely rare.
My Backtesting data is over 3 years, always when using FX I’m thinking of adding it to 5 years, my thoughts are that market behavior changes and 3-5 years is enough? Or how do you guys think?
Usually I get 200 trades/strategy but if I set it higher I might miss the rare ones with the low DD
I understand that this affects the robustness of the strategy if it’s only 50 trades but the low DD makes me want to take the risk
Off Topic, but:
How far back do you guys go with FX, how many years or bars do you use to get enough statistic data?
@dommech-Yeah, I leave the generator more wide open and once it’s done or as I filter is when I get really picky about the strategies, just so I don’t miss those nice and very low DD strategies.
For FX, I am going back 5 years. Now that we are in 2019, I changed my data horizon to start at 1/1/2014. I do this because it gives me enough bars as I am trading H1 and M15 charts. For the lower timeframes, 200k bars should be good enough to get some good strategies.
@jacpin thx sounds good, I go usually for M15 H1 and sometimes H4 but not lower
Will change to 5 years as well 🙂
I recently heard in one of Petko’s courses that instead of demo testing the files you could (optional!) Change the data horizon and leave 1 month out of the generator and later recalculate. this gives me even more confidence that the strategy is robust. I still put many EA’s on demo so exchange the bad ones
Hello Jacpin,
You are getting more count of trades on M15 probably because you have more data on it ( I guess you have the max for EA Stduo 200k there?).
So basically the more bars we have the more count of trades we can expect to see. Of course it depends how you set the generator/reactor, acceptance criteria, etc.
Just be careful if you use small count of trades and higher win/loss ratio, not to go in over-optimized strategies. Make sure that you do the Monte Carlo test and Demo testing (I know that you do demo).
Hello Dommech,
Glad to see you are working in more details. Yes, this method of removing the last one month is really great.
This way we see if the strategy was going to be profitable if we really traded with it. It saves a lot of time, and as you said, it gives confidence.
Both strategies look fine.
@Petko and Dommech-Can you let me know which course has this method to leave off the last month in the Data Horizon? I want to understand the steps on how to do this and see if it will really save me some time or provide me with more confidence in the strategies that I generate.
Hello Jacpin,
Glad to hear from you. I show that in the The Ethereum trading in 2018 + 99 Robots every month course, but I have it in my “to do” list where I want to dedicate a whole course about it. I will show different methods on how to avoid the Demo trading.
@Petko-You are awesome! If you can please do that I would be so grateful. Sometimes, the waiting for my EAs to pass demo is frustrating to me. But I don’t want to take unnecessary risks. 🙂 As soon as you have anything that shows how to avoid demo trading, then sign me up!!!
I will use Demo as well but I will not Demo test my strategies before going live, I will still continue to demo test “backup strategies” if I need to switch out some of the ones. Testing this way on “Out-Data” work as well as demo testing. Markets change anyway so it’s not a guarantee that the strategy will work live if it did on demo.
I go with enough history data to feel confident + 80-90% Monte Carlo + the last step is the one month of “Out data” I don’t feel that I need more confirmations than that, this way really saves Weeks or even months so I’m willing to take the risks.
Hey, I’m compounding profits so lets start to compound time 😉
- You must be logged in to reply to this topic.