Thanks At this time Iām doing the trailing stop with somewhat complicated rules:
If price touches the 23.6%CD then I move the stop to the -23.6%CD
If price touches the 38.2%CD then I move the stop to the 0%CD or break even
If price touches 41.4%CD then I move the stop to 0%CD plus 3 pips, then I lock 3 pip profit. Liteforex also gives multirebate of 1.5 pips when the trade is older than 2 minutes and the profit/loss is 3 or more pips, so this way I lock 3+1.5 pips
If price touches the 50%CD then I move the stop to 23.6%CD
If price touches the 61.8%CD then i move the stop to 38.2%CD
If price then touches the 78.6%CD, then I move the stop to 61.8%CD, and start trailing the stop about 1ATR
For each level touched I take a partial profit, that is 7 positions, I divide my max risk by 7, and assign that part to each level. Is a little complicate, but because this is a program, not by hand, I donāt care very much
I think Iām going to test placing only 3 parts instead of 7, one on 38.2%CD (TP1), other on 61.8%CD(TP2) and the last on 88.6%CD trailing with the same rule as you (50% of TP).
And yes, it launches alerts and can send emails and pictures of the pattern detected.
I only use 5M time frame, so there is no duplicated patterns.
And yes, if for example, the butterfly in XA-B falls on 78.6 +/-5% AND BC-D falls on 161.8%-5% to 261.8%+5% AND XA-D is on 127.2%-5% to 161.8%+5% then I consider it a valid butterfly pattern.
The most recent improvement I did on the analysis of historical data is to log if at D point there is a reversal candlestick pattern and an extreme value of stochastic and also the status of AB=CD (I donāt understand very well how can I say that if a pattern has a AB=CD or not, so I donāt know if my procedure is the right one or notā¦ yet!)
I just started following this thread recently, and I got a quick question when applying the pattern tradings. So if there are two harmonic patterns forming in the same pair but different time frame, what should I do? Should I trade the one in the larger TF or just skip the trade? In addition, what if there are two patterns formed in the same pair and TF? I have attached few charts from GPB/USD to explain my questions.
maybe they are the same pattern. The rule is the bigger, the most important one. Measure them, if they are the same pip length, then maybe they are the same pattern.
Well, I finished the tests. This results are considering the BEST profit at 78.6%CD, the results are this in percents:
Pattern Bat with AB=CD loss = 37.3976 win = 62.6024
Pattern Butterfly without AB=CD loss = 5.5556 win = 94.4444
Pattern Butterfly with AB=CD loss = 27.1560 win = 72.8440
Pattern Crab with AB=CD loss = 20.2256 win = 80.0000
Pattern Gartley without AB=CD loss = 34.3005 win = 66.1140
Pattern Gartley with AB=CD loss = 37.6439 win = 62.5480
With 61.8%CD this are the results:
Pattern Bat with AB=CD loss = 37.3976 win = 70.7461
Pattern Butterfly without AB=CD loss = 5.5556 win = 100.0000
Pattern Butterfly with AB=CD loss = 27.1560 win = 78.4404
Pattern Crab with AB=CD loss = 20.2256 win = 85.8647
Pattern Gartley without AB=CD loss = 34.3005 win = 71.6062
Pattern Gartley with AB=CD loss = 37.6439 win = 69.7621
I show the 61.8%CD as a method of verification of my algorithm, I found in the Internet that the accuracy of Gartleys are about 70%-80%, Bats about the same number, Butterfly and Crab 90%
Here the values at 61.8%CD shows more or less the same results, and a good percent if instead of 61.8%CD the 78.6%CD is used.
The tests are for 1000 samples on butterfly and crab, and 2000+ on bat and gartley.
Congratulaions, that is one difficult task. I have done lots of programming of EAs and know that backtests can be quite the project. Is this written in mql4? Is it something you wish to share.
I would have never considered trading these patters in M5 timeframes, it always seemed like waiting for a more substantial 1 hr setup was the way to go. Can you post a picture of example pattern on 5 min? Like what is a typical A-D swing in amount of pips?
What impresses me the most is the logic required to define a pattern.
I have collected every post of TMB and the images associated with each post, I also have simplified this into an outline of all he has said about these patterns. If anyone wants a copy pm me your contact (email or skype) and I will send you the word documents.
BTW: I have ran many EAās that will produce 500 trades in 3 months with 85% winrate with 1:1 Risk:Reward, when ran over 2 years they end up losers. If you could do your tests over longer period of time, it might give us a more clear picture.
I used various of them to start myself in Harmonic trading, specially the 54_pattern indicator found on the code.mql4.com site
That indicator, in addition of various more like ZUP indicator helped me to learn about how the patterns are formed, how they are detected. What I did is only a better way to detect them, using a new and adaptive zigzag method. Basically, instead of using a fixed ādepthā for the zigzag, I tested all the high/lows from the starting point to backwards until I found a combination that matches the pattern Iām looking for OR when I went backwards for to many bars (1 day of m5 bars)
Mm, Iām going to build a new function to get patterns based on C point instead of D point, that could take some time while I finish the current function
But if Iām right, a good C to D trade could be when C is almost the same as A, because that will be a double top/bottom.
Iām doing the EA (for MT4) on M5 because it is faster for testing, but I think I will use it in M15 or M30 because a spread of 3-5 pips on M5 is huge when the potential profit is 10-20 pips.
All my charts that I posted here are M5, and the average size of CD is about 30 pips, so the 61.8% is about 16 pips minus the spread.
The potential on Harmonic Trading is the fact that the patterns are very well defined, no guessing is necessary, so a computer program can do the task of looking for potential trades pretty good, at least thatās my theory.
What Iām having problems is with the fine tune, WHEN a valid pattern (in the numbers) is valid with the current market moves, when to trail the stops, when do not. I think all can be resumed in a set of rules, but I donāt know the rules very well hehehe, so right now while in testing the patterns are pretty good, on live charts they fail. The false breakouts are a headache for bots!
To my knowlege of AB=CD and of your description in your posts, I believe you have the AB=CD in your system. The meaning of AB=CD is that CD is at least 1x length of AB, e.g. CD can be 1xAB=CD (100% CD), 1.27xAB=CD (127% CD), 1.618xAB=CD (161.8%CD), 2.0xAB=CD (200%CD), or 2.618xAB=CD (261.8%CD).
Specially, it can be 0.786xAB=CD (78.6%CD) and 0.886xAB=CD (88.6%CD) but i forget which book used it. However this is not valid in TMB method as far as I understand.
I agree that use 5m time frame may be not the best choice. Even some patterns in 5m maybe valid and retrace 50% or more but you may only get a small amount of pips due to the narrow range of price movement. Could you test in hourly chart (1H), which may gives less patterns but the patterns maybe more valuable?
Hi Gt12Ry7Wl you have got some very nice patterns. Iāll try to answer your questions with my limited knowledge.
If a pattern is formed in a higher time frame, you normally can see it in consecutive lower time frames, for example, 1W pattern can be seen in 1D, 4H, ā¦ It is the same pattern so whichever time frame chart you use, you trade with the same pattern. Normally, using a lower time frame your fib is more accurate (if you cannot accurately define your fib due to system limitation, for example, in a 1W pattern, 1 minimeter may contain 50 pips, while in 1H chart 1mm may cover 5 pips).
If in a pair you can find patterns comes with patterns, it is a good sign that this pair respects the fibs and the price movement is highly tradable with the patterns. you just trade one pattern, then be prepared for the next one, like in your third chart, the CD leg of the first pattern, is the XA of the second pattern. They are not contradict with each other.
However, sometimes a pattern can be formed inside a pattern, which can be led to the same point and directions, adding extra confirmation to each other. If two patterns are led to different directions, they can normally separeted in different time frames. The lower time frame pattern may be the imediate dealing pattern and it may lead the price to complete the CD leg of a high time frame pattern if both patterns are tradable .
Hope I have expressed clearly and it will be understandable.
Yes, Iām starting now to test on higher time frames. I just added support for Alternate Bat and Deep Crab patterns, also checking the type of candlestick in the possible D point, and the actual percent of retracement. The idea is to know which factors influence more.
The success rate is improved with the AB=CD in Gartleys, for about 5%, but the number of tradable patterns are a lot lesser. Other patterns shows a variation of about 1-2% so I think the AB=CD is not a strict rule, at least with the current method of AB=CD calculation.
At this time Iām doing the trailing stop with somewhat complicated rules:
