From theory to practice - page 178

 
Alexander_K2:

And in general, the summary is that accounting and chasing every tick is not justified either mathematically, physically or economically, right?

The giant disadvantage of this fact is that it is impossible to use archives to test a strategy. For me it is very important to know the number of ticks over a certain time interval. That's why I'm stuck for so long - I wait about 1-2 days for a single trade. How many times I need to test it to be sure it works?

If I switch to OPEN, I will know for sure that I have 60 quotes in 1 hour. And there are archives for testing. Will it really come to this? Can you imagine, Nikolay, how much time is just wasted?

Allow me to make a couple of comments. You can no longer be dissuaded from applying probabilistic methods to problems where probability theory does not work. So within your methods.

Where do the minutes come from, have you thought? Well, from them, from the ticks. Using ticks, you can transform them to any representation, even to 10-second data. And also in such a way, as required by the trading system. What makes you think that ticks are not suitable for the simulation of trading on the history? On the contrary, in the terminal, the methods of testing "by all ticks" are considered as the most accurate.

Does the fact that the best duration of ticks series in a row appeared to be 8 hours suggest you any thoughts? After all, it is the duration of the trading session. See Nikolai Demko's comments https://www.mql5.com/ru/forum/221552/page19#comment_6170686 to my pictures with the analysis of market activity during the day https://www.mql5.com/ru/forum/221552/page19#comment_6168925. This 8 hours is exactly what characterizes your methodology, a sliding window on ticks the size of a session, and in case of a significant deviation of the spread characteristic - a return to the average is traded. I think that even this methodology, which is extremely smoothing reality, could be improved by making the sliding window linked to the phases of the sessions for each pair.

 
And I'm serious.
 
Vladimir:

Let me make a couple of remarks. You cannot be dissuaded from applying probabilistic methods to problems where probability theory does not work. So within your methods.

Where do the minutes come from, have you thought? Well, from them, from the ticks. Using ticks, you can convert them to any representation, even to 10-second data. And also in such a way, as required by the trading system. What makes you think that ticks are not suitable for the simulation of trading on the history? On the contrary, the terminal considers that the "all ticks" testing methods are the most accurate.

Does the fact that the best duration of a series of ticks in a row, which suits you best, is equal to 8 hours inspire you? It is the duration of the trading session. See Nikolai Demko's comments https://www.mql5.com/ru/forum/221552/page19#comment_6170686 to my pictures with the analysis of market activity during the day https://www.mql5.com/ru/forum/221552/page19#comment_6168925. This 8 hours is exactly what characterizes your methodology, a sliding window on ticks the size of a session, and in case of a significant deviation of the spread characteristic - a return to the average is traded. I think that even this methodology, which is extremely smoothing reality, could be improved by making the sliding window linked to the phases of the sessions for each pair.

Look here, Vladimir. The work is close to completion one way or another. I have done many things by myself, and you and others like you have helped me a lot. The model will be posted as I promised. It seems to be working - I have just made a deal of 400 pips on EURJPY.

But, there are still moments that confuse me.

1. Non-locality in time. I.e. there must be a strict dependence of a number of ticks on time for the law of the root of t to function. It can be achieved by working at a frequency of guaranteed tick reception. In my case, it is 2.57 seconds. It took me a long time to calculate this value. But if I switch to another broker, I'm sure I will have to recalculate everything again.

2. If I calculate sample volume using Chebyshev method, then due to different average value of increments we obtain different observation sliding window for all pairs. For most - 8 hours, for some - 12 and even more! Frankly speaking - I'm tired of calculating.

3. The most important thing! Well, well, here I work on frequency 2,57 sec. Where can I get archives for testing? Is it really necessary to wait for year to speak confidently about system operability?

PS If you specifically need the full working version of my project - I can throw in a personal message.

 
Dmitriy Skub:
I'm serious.

Maybe you are. Shelepin literally worshipped the golden ratio... I don't have time for experiments anymore - I need a Man with theoretical proof of this or that way of taking data

 
Alexander_K2:

Maybe it does. Shelepin literally worshipped the golden ratio... I don't have time for experiments anymore - I need a Man with theoretical proof of one way or another of taking data

Key words: I don't have time for experiments anymore - that seems to explain everything.

 
Roman Shiredchenko:

Key words: I don't have time for experiments anymore - that seems to explain everything.

Nah, I'm already bogged down in theory. I'm already down to analytical market formulas :)))

And, for example, all my family and friends need money immediately. I have to speed up - I'm looking for ways to test on stories. I have to think of something with tick archives. Learn to convert them to the right format. For example 1 tick in 3 seconds, 10 seconds etc. Something to think about as well...

 
Alexander_K2:

No, I'm already bogged down in theory. I'm already up to analytical formulas of the market :)))

And, for example, all my family and friends need money immediately. I have to speed up - I'm looking for ways to test on stories. I have to think of something with tick archives. Learn to convert them to the right format. For example 1 tick in 3 seconds, 10 seconds etc. Something to think about as well...

Thanks. I see. You need a sane TS - the market won't go anywhere. There's a lot to lose...

 
Alexander_K2:

3. The most important thing! All right, I work at frequency 2.57 sec. Where can I get archives for testing? Is it really necessary to wait for a year to speak confidently about system operability?

Here's the problem... Are you kidding me? Even I offered you, you don't take it, and then there's nowhere to take it... You're making trouble. Here's a list! of sources where you can get tick archives for free:

http://ratedata.gaincapital.com/

http://www.histdata.com/

http://advancetools.net/index.php/instrumenty/tikovye-ob-emy/istoriya-tikov for 3 different DCs

http://truefx.com/?page=downloads

http://ticks.alpari.org/ for 10 different account types

Re.

"2. If you count the Chebyshev sample size, because of the different average incremental value, it turns out that there is a different sliding window of observation for all pairs. For most it's 8 hours, for some it's 12 or even more! Honestly - tired of calculating."

I repeat: read Nikolai Demko's comments https://www.mql5.com/ru/forum/221552/page19#comment_6170686 . And later, somewhere he explained that sometimes two sessions, not 8 hours, will be the determining factor: ", for some - 12 and even more! ". I remember about the Mexican peso.

Finally understand the physical meaning of what you are doing. After all, many are trying to clarify, best of all, in my opinion, https://www.mql5.com/ru/forum/221552/page24#comment_6175789. Or a completely open example, Gorban mentioned by me above discovered that directional dispersion is not reduced in echolocation, worked in this direction and came up with a way to dramatically increase directional accuracy in hydroacoustics by changing the echo reception point. In other words, by hopping from hill to hill in this picture by Yuri Asaulenko. I think it is not necessary to explain what is the accuracy of direction finding in hydroacoustics?

You also noticed that the laws of large numbers don't work, and.... that's it.

 
Alexander_K2:

No, I'm already bogged down in theory. I'm already up to analytical formulas of the market :)))

And, for example, all my family and friends need money immediately. I have to speed up - I'm looking for ways to test on stories. I have to think of something with tick archives. Learn to convert them to the right format. For example 1 tick in 3 seconds, 10 seconds etc. There is also something to think about...

Making plans to earn money on forex, you can. Only there is always a high probability of making God laugh). It would be better not to give information to relatives. Then you won't have to make excuses in case you fail. And if you do, it would be a pleasant surprise for them.

 
Vladimir:

Here's the problem... Are you kidding me? Even I offered you, you don't take it, and then there's nowhere to take it... You're making trouble. Here's a list! of sources where you can get tick archives for free:

http://ratedata.gaincapital.com/

http://www.histdata.com/

http://advancetools.net/index.php/instrumenty/tikovye-ob-emy/istoriya-tikov for 3 different brokerage companies

http://truefx.com/?page=downloads

http://ticks.alpari.org/ for 10 different account types


For the links - thanks, of course. But, you don't understand - these archives still need to be converted! As if the ticks come every 1 or 5 or 10 seconds! I stand by my opinion - it is in the different number of ticks in a given time interval, the main difficulty of this task.