Is the tester working correctly ? - page 6

New comment
 

With regard to the correctness of the tester, the algorithm of bar formation seems to be more important, especially it affects the breakdown of levels or news releases. Maybe it is possible to simulate the bars according to Fibo levels. For example, in the "all ticks" mode, the tester analyses the price for three minutes ahead and forms the previous bars considering the next ones. In reality, the price does not move up and down in the candle's body.

 
FION:

With regard to the correctness of the tester, the bar formation algorithm seems to be more important

Absolutely correct. The rest (IMHO) is not. If your system essentially depends on the trajectory of price moves within a minute bar - it is unlikely to be stable and workable. Well, you'll get a great result in your tester, then what? - Then what if you sell it? (sorry, to those who don't understand) .....
IMHO - to check the stability of the system, the trajectory should follow the worst way: if the bar closing is higher than the opening, then first to the low, then to the high. And mirrored in the opposite case.

Good luck.
 
VladislavVG писал (а):
FION wrote (a):

With regard to the correctness of the tester, the bar formation algorithm seems to be more important

Absolutely correct. The rest (IMHO) is not. If your system essentially depends on the trajectory of price movement within a minute bar - it is unlikely to be stable and workable. Well, you'll get a great result in your tester, then what? - Then you have to sell it... (sorry, to those who don't understand) .....
IMHO - to check the stability of the system, the trajectory should follow the worst way: if the bar closing is higher than the opening, then first to the low, then to the high. And mirrored in the opposite case.

Good luck.



If this one-minute bar is after the news release, its low-high may be 30-50 pips, if the tested EA has such features as trailing stops, they all get blown out. The real signal rarely rolls back further than 50% but not low-high. It seems to me correct to simulate the real signal rather than fight the EA with the tester.
 
FION писал (а):

With regard to the correctness of the tester, the algorithm of bar formation seems to be more important, especially it affects the breakdown of levels or news releases. Maybe it is possible to model the bars according to Fibo levels. For example, in the "all ticks" mode, the tester analyses the price for three minutes ahead and forms the previous bars considering the next ones. In reality, the price does not move up and down in the candle's body.


I agree, for the task is not to model the worst conditions for testing but to model the ones closest to reality. Therefore I have a question and suggestion, respectively:
1. Who prefers exactly what period of testing and whether the contents of bars are significant
2. For those who are ready to share the difference between results of testing of their own EAs on simulated and real data, I am ready to provide POTENTIAL real data (not DEMO!) for say... last month... or two months :)
 
max_cpr писал (а):
FION wrote (a):

With regard to the correctness of the tester, the algorithm of bar formation seems to be more important, especially it affects the breakdown of levels or news releases. Maybe it is possible to simulate the bars according to Fibo levels. For example, in the "all ticks" mode, the tester analyses the price for three minutes ahead and forms the previous bars considering the next ones. In reality the price does not go up and down in the candle's body.


I agree, for the task is not exactly to simulate the worst conditions for testing, but to simulate the closest to reality. I therefore have a question and suggestion - respectively:
1. Who prefers exactly what period of testing and whether the contents of bars are significant
2. For those who are ready to share the difference between results of testing of their own EAs on simulated and real data, I am ready to provide POTYCH real data (not DEMO!) for say... for last month... or two :)
What to do with tick data on MT-4 tester? Probably need another tester.
 
FION писал (а):
max_cpr wrote (a):
FION wrote (a):

With regard to the correctness of the tester, the algorithm of bar formation seems to be more important, especially it affects the breakdown of levels or news releases. Maybe it is possible to simulate the bars according to Fibo levels. For example, in the "all ticks" mode, the tester analyses the price for three minutes ahead and forms the previous bars considering the next ones. In reality the price does not go up and down in the candle's body.


I agree, for the task is not exactly to simulate the worst conditions for testing, but to simulate the closest to reality. I therefore have a question and suggestion - respectively:
1. Who prefers exactly what period of testing and whether the contents of bars are significant
2. For those who are ready to share the difference between results of testing of their own EAs on simulated and real data, I am ready to provide POTENTIAL real data (not DEMO!) for say... last month... or two :)
What to do with tick data on MT-4 tester? Probably need another tester.
What's the idea? :) What do you think the MT4 tester generates inside bars when testing on minutes?
By the way, take a closer look at what the tester generates inside bars (not necessarily on minutes)
 
max_cpr писал (а):
FION wrote (a):
max_cpr wrote (a):
FION wrote (a):

With regard to the correctness of the tester, the algorithm of bar formation seems to be more important, especially it affects the breakdown of levels or news releases. Maybe it is possible to simulate the bars according to Fibo levels. For example, in the "all ticks" mode, the tester analyses the price for three minutes ahead and forms the previous bars considering the next ones. In reality the price does not go up and down in the candle's body.


I agree, for the task is not exactly to simulate the worst conditions for testing, but to simulate the closest to reality. I therefore have a question and suggestion - respectively:
1. Who prefers exactly what period of testing and whether the contents of bars are significant
2. For those who are ready to share the difference between results of testing of their experts on simulated and real data, ready to provide real data (not DEMO!) say ... for the last month ... or two :)
What to do with the tick data on the MT-4 tester? Probably need another tester.
What's the idea? :) What do you think the MT4 tester generates inside bars when testing on minutes?
By the way, look closely at what is generated by the tester inside bars (not necessarily on minutes)
It is clearly not a tick sequence but a certain algorithm. As for the testing period - it depends on the strategy (flat or trend, intraday or long-term), when working on 4H timeframe the period is longer than when working on 5 min. I think that any EA requires periodic reoptimization, the more frequent the smaller timeframe is used for work.
 
It is not a tick sequence that is generated, but a certain algorithm. When working on 4H timeframe the testing period is longer than when working on 5 min. I think that any EA needs periodic reoptimization, the more frequent the smaller timeframe is used for work. <br / translate="no">
Look at the description of fxt file format. Something tells me they don't contain algorithms :) And clarify, please, how does the testing period depends on the flatness and trendiness of the tested strategy?
 
max_cpr писал (а):
It is not a tick sequence that is generated, but a certain algorithm. As for the testing period - it depends on the strategy used (flat or trend, intraday or long-term), for example, when working on 4H timeframe the period is longer than when working on 5 min. I think that any EA requires periodic reoptimization, the more frequent the smaller timeframe is used for work.

Look at the description of fxt files format. Something tells me that they do not contain algorithms :) And clarify, please, how the testing period depends on the flatness and trendiness of the strategy under test?
When we are talking about the tick history, it only makes sense to discuss the formation of minutes, the other timeframes can be generated correctly on this basis. Regarding the second part of the question - trend strategies assume medium-long term trading using larger timeframes, therefore testing should be done on longer periods. By flat - we mean intraday trading, when the price is in flat 70% of the time, respectively, depending on many factors (winter, summer, war, no war, etc.) intraday volatility is changing significantly, so the MTS working in this mode should be optimized more often (once a month, half a month).
 
FION:
VladislavVG:
FION:

With regard to the correctness of the tester, the bar formation algorithm seems to be more important

Absolutely correct. The rest (IMHO) is not. If your system essentially depends on the price movement trajectory within the minute bar, it is unlikely to be stable and efficient. Well, you'll get a great result in your tester, then what? - Then what if you sell it? (sorry, to those who don't understand) .....
IMHO - to check the stability of the system, the trajectory should follow the worst way: if the bar closing is higher than the opening, then first to the low, then to the high. And mirrored in the opposite case.

Good luck.



If this one-minute bar is after the news release, its low-high may be 30-50 pips, if the tested EA has such features as trailing stops, they all get blown out. A real signal seldom rolls back more than 50% and not low-high. I think it's correct to more accurately simulate the real signal rather than fight with the tester.
So you want to adapt the operation of the tester to the trading conditions of one or two cases ? :). I don't think this is right (or rather I'm sure it's wrong). IMHO - there is nothing worse in the market than trading with false confidence in the performance of a non-working system.
Also, look closely - in most cases during news releases, the price goes first to the side opposite to the main movement, then to the side of the main movement, and often, contrary to all expectations, just straight to the stops (wherever the trader puts them :) ). This is the first and the second: how many brokers will give a trader an opportunity to place or modify orders during the news?

In any case, good luck.
1234567
New comment