Discussion of article "Developing a self-adapting algorithm (Part II): Improving efficiency" - page 5

New comment
 
Maxim Kuznetsov:

and I don't mean figures in 2-3-4-5 candles :-) they have too small a window no matter how you spin it.

The article correctly notes that the ratio is important, but it is not well founded that 50/50 is the norm. This conclusion is drawn from "general postulated" considerations, which it is not

---

diverting from the topic, to the equally popular topic of "trends/waves". In an impulse move, quotes are up and b/w up, in a pullback they contradict. A pullback movement is formed by a smaller number of counts.


The "pullback" is highlighted - the quote goes down significantly, but there are more counts/candles up. The whole downward movement is formed by a counting number of candles, and only in the Euro-session.

It's hard to make out such a thing in real time, but it's worth taking into account when marking history.

Yes, I have analysed more than 100 instruments and wrote about it in the first article. Everywhere plus minus 50%. I do not claim that it will be absolutely everywhere, but the sample is large. I checked both shares and currencies.
About trends and pullbacks.... There is indeed a peculiarity with trends and pullbacks in the market, but it is not simple and you can't tell it so quickly. Maybe I'll write about it sometime. The point is that the size of candles changes on trend and pullback movement, but the most interesting thing is the law and reasons why the change happens.
 
great series of articles, thank you author
 
Smith-John:
great series of articles, author please accept thanks
Thanks)
 

I did not find a stop loss in your code.

If I understand correctly, all research results are obtained with no real risk limit per position?

 
Evgeniy Ilin:
For some reason it seems to me that it is better to get away from Mql in general, because the testers are so slow and miserable, even MT5 is so slow and miserable, even MT5 rapes the hard drive and RAM, for the sake of interest, don't be lazy to take a look. And the speed of tests is extremely low...

You say counter-revolutionary things, Filipp Filippovich.

 
Sergey Voytsekhovsky:

Good evening.

I suspect that in line 256 sign 87 there is a typo, instead of "-" it looks very well "=".

Please confirm, I'll correct it myself, or maybe some hidden meaning I do not understand, error not burn, burns as Warning.


I checked this point, replaced the minus with =, nothing changed. So it doesn't affect much. The programmer who wrote this is a genius) no matter what sign is, the pause works the same way).

 
Aleksandr Masterskikh:

I didn't find a stop loss in your code.

If I understand correctly, all the research results are obtained with no real risk limit per position?

There are stop losses and take profits too. There are several modes of stoploss operation, it can be calculated from the current volatility for each position separately or for the whole series together. Stoploss and profit can be set up very flexibly. There is an additional virtual stoploss on the maximum drawdown per series, then it controls that the loss on open positions does not exceed the set value. There is a virtual stoploss on the minimum equity.

Here is a test with stoplosses from 2010 on GBPUSD

Here is a whole block of settings responsible for stoplosses and Take Profits

The algorithm is very flexible, you can customise a lot of things

 
Evgeniy Ilin:
For some reason it seems to me that it is better to get away from Mql in general, because the testers are so slow and miserable, even MT5 is so slow and miserable, even MT5 rapes the hard drive and RAM, for the sake of interest, don't be lazy to take a look. And the speed of tests is extremely low. I have a software that tests tens or even hundreds of times faster. I don't know what they put in there) they are probably mining on us). I have a friend he runs a test on the fifth 10 years so he eats 25 gigs of RAM ))) well I just laugh. Service Desk said go to the forums ask what's wrong DDD. More recently I am inclined to the fact that Mql code should be made as a layer between your software and the terminal if necessary. And the logic should be a separate independent software. And with this approach it is not difficult to integrate with other terminals, Ninja Trader for example or CTrader, the logic 1 time rolled out in your software, rolled out robots receivers of commands in each terminal and just accept commands from the software. Just this software is cool when you bought a robot and once tested and rejoiced for the backtest, but in this case, when there are one hundred and five hundred settings, I just know by hearsay how much time it takes, time that goes nowhere, for this reason I try to make simpler and less settings always.

I've been talking to some guys, they're doing an open-source trading platform. I was interested, because you can connect to many things, even to Tinkov. You can even order from them a connector to mt5. I asked about the tester and they said that their tester is slower and not as cool as mt5. The idea of their platform is cool, but apparently it is not so easy to make a good tester.

I also have a tester that consumes a lot of memory up to 12-20gb on 28 instruments. But I use minutes. And of course it takes a long time to load from the screw. I don't know how technologically it is possible to reduce memory consumption, probably everything can be done, but other budgets are needed.

In general solved the problem with memory by putting 32 gigs and nvme ssd with a speed of 3500 mb/s. So far the drive has never become a bottleneck. But I don't do any optimisation, just tests.

The trend in software is that hardware requirements are constantly growing. It is understandable, nobody will bother with optimisation as in the 80s when every kilobyte was saved. I think we should just accept it.

As for the logic in third-party software, this I thought, and to the terminals just make a connector. This is of course a universal solution. But then the question arises, isn't it better to connect via FIX api at once?
 
Maxim Romanov:

there are stoplosses and take profits too. There are several modes of operation stoploss, it can be calculated from the current volatility for each position separately or for the whole series together. Stoploss and profit can be set up very flexibly. There is an additional virtual stoploss on the maximum drawdown per series, then it controls that the loss on open positions does not exceed the set value. There is a virtual stoploss by minimum equity.

In the test with stoplosses from 2010 on GBPUSD

Here is a whole block of settings responsible for stoplosses and takeprofits

The algorithm is very flexible, you can adjust a lot of things

Now it's clear, thanks!

 
Denis Kirichenko:

Counter-revolutionary things you say, Filipp Filippovich.

Well, it's a cliché, it's not me who invented these things ) a lot of people say them. Filipp Filip Filipovich still chooses words ) . Outside the site, I would speak quite differently ). Of course, I realise that in my article I will never say such things, but here you can give a little truth). Not too much, of course, but who knows will understand. I'm afraid to say too much

1234567891011
New comment