Optimise an EA and get the best of the optimised ones. - page 6
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Good times... has anyone thought about the "permanent" things of the market.... i.e. those "moments" - where the market does not change its conditions... for example: pullbacks, pullbacks from a trade, a flat break and things like that... For this kind of things you have to adapt ... or rather adjust your trading robots ... If a better idea occurs than the one you are already using... The main thing is to adjust for large movements - and you can forget about small ones... =)
What are the big moves? How much is that anyway?
Denis Tarasov:
And so the main task is to adjust to big movements - and you can forget about small ones... =)
You don't have to adjust!
You have to catch all the movements. That is, have a bunch of TCs, each of which - catches something different. And only look at the result, selecting the workers.
But why not gather all TCs into one, and optimise the weights for each TC?
Yes, there is such an idea. But to optimise the weights of each one, we need to optimise each one in terms of parameters first.
Soon I'm going to switch over to this task again and post test EAs representing separate types of TS and the general EA capable of working in real time and with MM containing all tested TS. Here, it will be possible to run only one TS for the time being, but the plan is to make each of them to work with its own MM - this is just the idea of "weights" of each TS within the general Expert Advisor.
Can you briefly write what your TS consist of? Entry by indicator, exit by pips, is there a trawl...
Will these EAs work on MOEX? If so, I may use them on Si or any other symbols. But I would advise you to make a general counter of settings, because the optimization will be long, then the project participants may need the power for their needs, and then the person will have to stop the optimization, and the counter allows you to optimize portions, for example, you made 10000 passes, saved the result and continued optimization at 10001, when the opportunity arises again.
Can you briefly write what your TS consist of? Entry by indicator, exit by pips, is there a trawl...
Will these EAs work on MOEX? If so, I may use them on Si or any other symbols. But I would advise to make a general counter of settings, because the optimization will be long, then the project participants may need the power for their needs, and then the person will have to stop the optimization, and the counter allows you to optimize portions, for example, you made 10000 passes, saved the result and continued optimization at 10001, when the opportunity arises again.
As I wrote above, my TS are the most "dumb" ones, with minimum parameters. The main trick is that there are a lot of them. As a result, the question from "what to do to make my TS work stably" transforms into the question "how to select the TS, that already works stably, and will not change its behavior for as long as possible". Since my TS is designed to "cover" as wide a range of market behavior as possible - there is always a TS that works at the moment.
The algorithms themselves are based on the following points:
1. trend detection. At the moment I use either crossing of price and sliding bar or touching of price-channel's border. These are two variants. (It shows the parameter - period of moving or channel).
2. Entry can be either with the trend or against the trend. For a moving average, the signal to enter is a bar-impulse ("hit" bar, "long" bar, "shift"). For a channel, the touch itself is a signal. So we have two options (for the moving bar, there is a parameter - the type of the impulse bar, and where to count its direction).
3. Trailing - I use different variants: forward trailing - the set SL "pulled up" to the current price to close; backward trailing - the set TP "pulled up" to the current price to close; fixed TP-SL; reversals. Total - four options. (appears the parameter - the size of SL or TP relative to the daily volatility, and for fixed TP/SL - also TP/SL ratio).
Total - we have 2x2x4=16 variants of TP per symbol.
To all TC add more parameters: timeframe, possible limitation by hex, moment and Breakeven level, for sliding bars - filter of entering by distance from EMA (not to enter, say, the trend, when we have gone too far away from the EMA).
As practice shows (I said above) - there is ALWAYS a TS that works at the moment. There is no question of "what to come up with". It's just a question of choosing.
About the exchange - the principles themselves are universal.
But the whole system is designed for MT5, with the possibility of working on MT4, now 28 symbols are recognised.
In principle, nothing prevents you from using any symbol available in MT5, you just need to complement the ECurrencySymbol enumeration and refine the functions that interact with it.
I would advise you to make a general counter of settings, as the optimization will be long, the power may need to participate in the project for their needs, then the person will have to stop the optimization, and the counter allows you to optimize portions, for example 10000 passes made, saved the result and 10001 will continue optimization when the opportunity arises again.
I believe the best genetic optimization is performed in a year. Backtest for 5 months, Forward 7, OHLC mode at 1M.
Such optimization takes two to four hours on quad core i5, depending on number of parameters. 20-40 hours on single core AMD Sempron LE-1200.
The total counter of settings is not required since MT5 allows you to stop optimization and further start it from the point where it was stopped. I use this quite often.
At the moment, the situation for 'favourites' is as follows:
The "quality" column is an integral estimation of the balance curve, taking into account a number of its parameters.
The curves themselves (all the TC work without MM, with a minimum lot, the y-axis is the income in the deposit currency - dollars):
You can see that the most "beautiful" curve have systems of reverse trailing (RTS), but such systems are extremely dangerous, because by reducing TP during trailing they have a very small (although regular) profit (usually 1-3% of daily volatility), but with a huge stop (usually from 3 to 5 daily volatility, there are a couple of TS in which the stop is as much as 7-day volatility). These systems are well adapted to severely flat symbols, but even a small trend pulls them out.
As I said before, anyone who optimised one of the "outsider" TS can get access to any "favourite" TS for 3 months.
The Expert Advisor itself (EALeague) and the Expert Advisors for optimizing individual TS are available on Yandex-disk.