An effective trading strategy based on multi-currency analysis of multiple DCs - page 3

 
solandr - I am far from accusing anyone, I am simply stating a fact.
Regarding the phrases you quoted, I admit that what I said was a mistake. In my subsequent posts I have answered all the questions that arose. And saying that people here are creative, I said it without any irony, and in vain you took offense at me for it.
I think there is no point in going back to this conversation.
 
Piligrimm:
solandr - I am far from accusing anyone, I am simply stating a fact.
Regarding the phrases you quoted, I admit that what I said was a mistake. In my subsequent posts I have answered all the questions that arose. And saying that people here are creative, I said it without any irony, and in vain you took offense at me for it.
I think there's no point in going back to that conversation.

I was partly interested in the discussion. It would be interesting to see your indicator with 12-bar prediction.
I have 2 questions in relation to this technology:
1. Apparently the prediction succeeds on every tick, or at least on every bar. In all probability, the next prediction differs from the previous one in one way or another. What does your technology do to get the final result? Is the last known forecast selected, are all 12 forecasts considered, and if so, in what way (simple summation, with weighting coefficients, with quadratic, by what criterion is the weight determined? -by time, corr. coefficient, etc.)?
2. Could you please show a picture of the indicator? (the code, as far as I understand, is not disclosed)

(I was working on similar idea last year, but had to put it aside for more important work; semi-immediate, unfinished working version can be seen here: 'Did you see this picture?' of 28.03.2007 22:34; hope to come back to it in May, to finish it off:)
 
SK. писал (а):
It would be interesting to see your indicator with a 12-bar prediction.

Well, well... especially after Kristi's re-rendering of it, it's easy to imagine what it looks like and what good it does.

2 Piligrimm: the MT5 tester will support multicurrency back-testing, so don't tell me, and if we are talking about the real one, it's nonsense to put the terminal online because of the forward testing of your idea. I think you should not have decided to stop the topic for nothing, because nobody needs this kind of altruism here ;)
 
If there is a difference at least in a hundred milliseconds when calculating the distance between two brokerage companies, it may be used. The interprocess communication is not difficult to implement on remote threads, so we will take it into account in the future. As for answers, especially concerning pipsing, someone may be really bad at pipsing. As for the idea itself, they do need support, not in arguments, but in thoughts that may occur based on a graphical output, of course when there is nothing to think about, thoughts do not appear, it all depends on the depth of thinking, probably the thinker.

Z.I.: What would you think about, if there were only quotes in front of your eyes:). Probably only about how to see them in a chart, and if you imagine that there is something more than a chart, because a person who has not studied at the same school and the chart will not think:) Try to think beyond your knowledge and everything will be, well, if not all then at least something for sure:). Try to imagine how you would look for answers without tips, probably only by groping and theoretical research.
 
xnsnet:
As for the idea itself, the support of the idea is still needed, not in the reasoning, but in the thoughts that will visit, based on the graphical output, of course when there is nothing to reflect on, thoughts do not appear, it all depends only probably on the depth of thinking of the thinker.
There was a time when I, working in DC, performed a cluster analysis of quotes from 12 banks. At that time it was supposed to find the connection between future fluctuations of quotes and time difference between quotes from these banks. So, even at that time it was determined that the pattern we've discovered can be used only for pipsing and only in strong market moves. And not only clustering was performed, but also data analysis using several other methods (or expert systems, as some of them are called by the author). And I cannot understand what one can see in the method proposed by the author for analysis and trading even with short periods (not speaking about medium- and long-term), if one cannot practically find anything useful even with clean, "unbrushed" (not averaged by the dealing centre) quotes?

I am also interested in this topic, but I have not come across any new ideas since then.
 

I find it difficult to evaluate the experience of other people besides myself, my experience is not that great, not to mention the usefulness and its possession is not always reflected as necessary, despite the fact that the experience in a variety of directions and Hobbies, I prefer to think that for something it was necessary and sooner or later will be useful, and even from what I say that in each experience I come to a maximum, my maximum may differ from the maximum of another, more maximally man.

About the ideas is right, because none of us have completed them, and the one who has completed them is unlikely among us :) I have spent about three months researching more suitable areas and it is only this one that really got me, I do not mean DC, but as rightly pointed out by Pilgrim, it may yet improve something, without the facts, I cannot say what exactly. Speaking of facts, there are too few to make one big fact out of it.

I thank my father with a long experience in applied programming, since the days of little computerization, for once being so adamant in condemning the poke method without bending to theoretical investigation, at least I respect one of those rules and I still break the other one to this day:)

 
DrawDown:
xnsnet:
As for the idea itself, support for the idea is still needed, not in the reasoning, but in the thoughts that will visit, based on the graphic output, of course when there is nothing to reflect on, thoughts do not appear, it all depends probably only on the depth of thinking of the thinker.
There was a time when I, working in DC, performed a cluster analysis of quotes from 12 banks. At that time it was supposed to find the connection between future fluctuations of quotes and time difference between quotes from these banks. So, even then it was determined that the pattern we've discovered can be used only for pipsing and only in strong market moves. And not only clustering was performed, but also data analysis using several other methods (or expert systems, as some of them are called by the author of the branch). And I cannot understand what one can see in the author's proposed method for analysis and trading even with short periods (not speaking about medium- and long-term), if one cannot practically find anything useful in clean, "unbrushed" (not averaged by the dealing centre) quotes?

I'm interested in this subject too, but since then I haven't come across any new ideas.
LOC ;)
 
SK. писал (а):
Piligrimm:
solandr - I am far from accusing anyone, I am simply stating a fact.
As for the phrases you quoted - I admit that what I said was a mistake. In my subsequent posts I have answered all the questions that arose. And saying that people here are creative, I said it without any irony, and in vain you took offense at me for it.
I think there is no point in going back to this conversation.

I was partly interested in the discussion. It would be interesting to see your indicator with 12-bar prediction.
I have 2 questions in relation to this technology:
1. Apparently the prediction succeeds on every tick, or at least on every bar. In all probability, the next prediction differs from the previous one in one way or another. What does your technology do to get the final result? Is the last known forecast selected, are all 12 forecasts considered, and if so, in what way (simple summation, with weighting coefficients, with quadratic, by what criterion is the weight determined? -by time, corr. coefficient, etc.)?
2. Could you please show a picture of the indicator? (the code, as far as I understand, is not disclosed)

(I was working on similar idea last year, but had to put it aside for more important work; semi-immediate, unfinished working version can be seen here: 'Did you see this picture?' of 28.03.2007 22:34; hope to come back to it in May, to finish it off:)

Firstly, it is not an indicator, but an expert system. I am not dealing with indicators, I have made one, as they say - the devil has lead me there and I am not going to do more.
The Expert Advisor system I made has no direct relation to the discussed theme, I just mentioned it as an example of using multicurrency analysis and it is based on another principle than the strategy suggested in this thread. I do not plan its commercial use and would not like to make a PR for it.
But if I briefly describe its work: it is built on the principle of analysis of 15 currency pairs, the forecast is made on each new bar, for 12 steps forward, as I wrote, by High, Low, Close, as well as the trend is selected from Close with the help of Vivelet transformation, in this case, the forecast of each of these signals is made independently on the first exit of the system - on the first bar, on the second exit - on the first and second, and third, etc.Total, the Expert Advisor system has 48 outputs that give independent decisions. Correspondingly, the shorter is the forecast interval, the higher is the accuracy. The resulting signal is obtained by adding 12 values for the first bar, 11 values for the second one, 10 for the third one, etc. I should do the same for all 4 signals. Such averaging allows to get rid of random noise and errors of the forecast unit which are compensated by summing up. And the accuracy in this case is much higher than in case of one-time forecast. Then the obtained results are summed up with values of the forecast made for the previous bars during the previous system operation cycles.
 
Piligrimm:
But to make a long story short: it is built on the principle of analysis of 15 currency pairs, the forecast is made for each new bar, for 12 steps forward, as I have already written, by High, Low, Close, as well as by the trend which is separated from Close using Vivelet transformation, besides, the forecast of each of these signals is made independently for the first output of the system - for the first bar, for the second output - for the first and second, for the third output - for the first and second and third, etc., until the 12th output, which makes the forecast for all 12 bars. Altogether, the expert system has 48 outputs, for which independent decisions are given. Correspondingly, the shorter is the forecast interval, the higher is the accuracy. The resulting signal is obtained by summing up 12 values for the first bar, 11 values for the second one, 10 for the third one, etc. I should do the same for all 4 signals. This kind of averaging allows to get rid of random noise and errors of the prognosticating unit which are compensated by summation. The accuracy in this case is much higher than in case of one-time prognosis. Then the obtained results are summed up with values of the forecast made for the previous bars in previous system operation cycles.
I don't understand what the "first system exit", "second exit ...", etc. means.
Also, if I may ask, why a simple averaging of the predicted 12-bar segments is adopted in the calculations?
After all, as far as I understand it, the closer a forecast is made to the present, the greater its credibility, and the highest is for the most recent forecast.
At the same time, the forecast made 11 bars ago (which last bit still affects the overall forecast, i.e. the nearest unformed bar) is the least reliable. The forecasts themselves may differ significantly. Don't you think it's necessary to average the forecasts to get the resulting forecast curve, using weights proportional to the reliability or correlation coefficient relative to the simple average of the forecasts?

And another question about the lag of signals between DCs. Can you estimate the lag approximately quantitatively? Is it concerned with "fluffiness", i.e. lagging alternates with advancing, or there is a shift of the general average of one DC in relation to the average of the other, and if so, by how much (in seconds)?
 
Yes by the way about the specific data would be interesting, how many milliseconds gap and in what intervals and throughout the graph, a tenth of a second is almost invisible speed, what to talk about ten milliseconds and whether it can be caused by the speed of your internet connection or routes of your provider?

Do a simple ping to the DC servers or better yet cite a tracert table, geez I didn't think of that at all :) The channel to the Internet is also no small factor:) Otherwise it turns out that this is a void and the game in this direction is only an illusion, self-deception, of course again, I'm not saying for sure:) Because then the orders would not pass:)

In addition I generally suspect that there is a timestop, relative to which the DCs are provided with data on an equal basis, although again, I do not claim anything:)