Discussion of article "Studying candlestick analysis techniques (Part II): Auto search for new patterns"

New comment
 

New article Studying candlestick analysis techniques (Part II): Auto search for new patterns has been published:

In the previous article, we analyzed 14 patterns selected from a large variety of existing candlestick formations. It is impossible to analyze all the patterns one by one, therefore another solution was found. The new system searches and tests new candlestick patterns based on known candlestick types.

In order to develop the new candlestick pattern generation algorithm, we need to define the key rules:

  • New patterns will consist of one, two or three simple candlestick types.
  • The simple candlestick types are: long candlestick, short candlestick, spinning top, doji, marubozu and hammer.
  • Candlestick types will be divided based on the direction: bullish and bearish. The exception is the doji candlestick.
  • Simple candlestick types can be repeated. Example: a pattern of two bearish long candlesticks.

The general scheme of new pattern creation is shown in Figure 1.


Fig.1 New pattern creation algorithms.

Thus we have a certain pool of candlesticks, of which new patterns will be formed. These new patterns will contain 1-3 candlesticks with or without repetitions. The pool will totally contain 11 basic candlesticks. Generated candlestick patterns will be analyzed according the same principle, which was discussed in the first article.

Author: Alexander Fedosov

 

It would be useful to divide the analysed interval into 2 segments - select promising patterns on one segment, and check their consistency on the second segment (analogue of in- and out-of-sample).

Or make the interval setting from both sides, so that you could limit the analysed interval manually.

 
Andrey Khatimlianskii:

It would be useful to divide the analysed interval into 2 segments - on one segment to select promising patterns, and on the second segment to check their consistency (analogue of in- and out-of-sample).

Or make the interval setting from both sides, so that you can limit the analysed interval manually.

I have a slightly different idea. To make a separate class and a set of methods to be used in indicators and Expert Advisors. Before there was theory, now there is practical application.
And the consistency of any pattern can be checked anyway. If a pattern is found, it is immediately checked where the price went, how quickly (1-3 candles) it will reach the set profit. In the table it is the coefficient K.
 
Alexander Fedosov:
The consistency of any pattern can be checked. Found, immediately check where the price went, how quickly (1-3 candles) will reach the set profit. In the table it is the coefficient K.

I meant to count K on one piece and check on another.

 

Candlestick classical models do not take into account important "internal" candlestick parameters, namely: the ratio of candlestick body to candlestick amplitude, the depth of correction inside the candlestick, the relative amplitude of candlestick shadows.

All this reflects the dynamics of the process very well. If we add these parameters, the forecasting will improve significantly (especially if we use such analysis simultaneously on several scales). Actually, this is done in the impulse equilibrium theory.

 

Is it possible to sort by any value? Or is it possible to upload the table to Excel?

 
s22aa:

Is it possible to sort by any value? Or is it possible to upload the table to Excel?

I will take into account in the future update.

 
Alexander Fedosov:

I will take it into account in the future update.

I wanted to start candlestick analysis myself, and now I have such a gift in the form of a wonderful EA from you, Alexander.

Well then, I have more suggestions.

The Expert Advisor does not see the instruments of the Russian stock exchange. It sees only the instrument on the chart of which the Expert Advisor is running. Not critical.

It is necessary to choose the testing period, not just the last candles.

It would be nice to be able to test patterns one by one. We find promising patterns on a small period of time, and then run them one by one on a larger period of time. That's how it is roughly.

Using candlestick analysis just head-on did not give positive results. Wrote a simple trading advisor, which trades on the results of testing .

This is the RTS index gluing contracts.

P.S. While writing, I thought that maybe I have something wrong with the RTS index. Since the testing of the Expert Advisor trading on the patterns of internal and external bars, showed positive results for the last two months and negative results if you test the period of a year. I will check what results will be obtained on the CME.

 
Alexander Fedosov:

I will take it into account in the future update.

And if it is not difficult, please tell me in a future article how to import perspective patterns. The code is not simple and it is difficult for non-advanced people to understand what is written where and from where and from which buffer the results can be taken.

It takes a long time to do it by hand, and there are quite a lot of patterns. Only for purchase I have the following selection of promising ones on M5

         if(paternDn(4,3,2,i,close[i]))  
         if(paternDn(8,2,9,i,close[i]))
         if(paternDn(2,2,5,i,close[i]))
         if(paternDn(9,5,3,i,close[i]))
         if(paternDn(2,2,9,i,close[i]))
         if(paternDn(9,8,1,i,close[i]))
         if(paternDn(2,9,8,i,close[i]))
         if(paternDn(9,8,2,i,close[i]))
         if(paternDn(6,9,8,i,close[i]))
         if(paternDn(8,1,3,i,close[i]))
         if(paternDn(9,8,5,i,close[i]))
         if(paternDn(2,8,8,i,close[i]))
         if(paternDn(5,1,2,i,close[i]))
         if(paternDn(9,2,2,i,close[i]))
         if(paternDn(5,2,2,i,close[i]))
         if(paternDn(4,2,2,i,close[i]))
         if(paternDn(3,4,8,i,close[i]))
         if(paternDn(8,8,6,i,close[i]))
         if(paternDn(1,1,2,i,close[i]))
         if(paternDn(4,8,5,i,close[i]))
         if(paternDn(9,2,1,i,close[i]))
         if(paternDn(8,4,9,i,close[i]))
         if(paternDn(9,8,9,i,close[i]))
         if(paternDn(9,2,2,i,close[i]))
         if(paternDn(8,2,8,i,close[i]))
         if(paternDn(4,9,1,i,close[i]))
         if(paternDn(3,2,9,i,close[i]))
         if(paternDn(1,6,9,i,close[i]))
         if(paternDn(8,5,8,i,close[i]))
         if(paternDn(8,9,9,i,close[i]))
         if(paternDn(9,2,9,i,close[i]))
         if(paternDn(3,3,8,i,close[i]))
         if(paternDn(1,8,1,i,close[i]))
         if(paternDn(8,8,1,i,close[i]))
         if(paternDn(8,10,11,i,close[i]))
         if(paternDn(8,8,10,i,close[i]))
         BuffDn[i]=EMPTY_VALUE;
 
s22aa:

And if it is not difficult to tell in the future article how to import perspective patterns. The code is quite simple and for the non-advanced it is difficult to understand what is written where and from where and from which buffer the results can be taken.

It takes a long time to do it by hand, and there are quite a lot of patterns. Only for purchase I have the following selection of promising ones on M5

A couple of days ago I sent the third part for testing. There is a library for working with patterns. You can use it to create indicators and Expert Advisors. There will be examples of how to do it.
At the moment, I receive requests for the Analyst's development. As soon as I gather enough requests, I will try to implement them.
 
very enlightening information.
12
New comment