|От 24 септември 2011|
Do not trade cycles below 4 h time frames based on ehler's type of cycle analysis on EUR/USD. On this sample it worked on daily because we had cycle activity. However when those cycles are present they are calculated very well.
I did that because I wanted to test the pure cycle analysis versus reality.
Not a grail.
Some 206 days later I made again some tests.
The daily frame was a mess, but as the market was in trend mode that is not surprising
The 4 h time frame had some very strong points but finishes with 292 losson a 10000 account with 0.1 lots
The 1 h time frame: A complete and surprising success:
635 profit and 58 % of winners with 1.16 profit/loss ratio
The 30 m time frame
350 profit 57 % winners with 1.06 profit/loss ratio
Do not forget that there is no optimization here. We identify cycles according to the cycle period indicator.
Just while I was talking about cycles I realized that there is a parameter within the cycle period indicator that is affecting the output greatly.
And making some quick simulations I see that under some parameters of alpha the expert turns out to be a beast.
This is really interesting. The averaging and smoothing inside the cycle period calculation in fact is critical.
If you change the smoothing a little bit and you will have completely different cycle period readings.
Question : Why a different period regarding both MAs ?
double FastMAPeriod= iCustom(Symbol(),Period(),"CyclePeriod",Alpha,0,0);
double SlowMAPeriod= iCustom(Symbol(),Period(),"CyclePeriod",Alpha,0,0)/2;
I would have got same values or the opposite , means
FastMAPeriod = iCustome(....) /2
and SlowMAPeriod = iCustom(...)
... there is something I don't understand ?.... ;)
Hi, this is explained here. This is a pure cycle system (supposing that there are cycles in the prices and exploiting them).
When doing tests I see regularly more than 55 % of winners. That means to me without going any longer that there is something.
Some market frequencies (time frames) do not offer trading cycles: the very long (daily) and the very short (15m and below).
The medium market frequencies of 30m and 1 h represent a certain interest for trading cycles
That of course are results of tests on EURUSD on the recent market data from the last 6 - 8 months.
So the next logic question was about the alpha.
First we see that modifying the values of the alpha affects greatly the cycle period evaluations and that affects irectly the results of the expert.
Second and that was the second idea why not to use a neural network, or a basic kernel in order to use the the maximum the genetic algorythm of mt4.
That is what I did, as the stock values is 0.07 I used the chaotic kernel (you can use others too) to calculate the best value of smoothing giving the most optimal results in back-tests.
For the moment I can't make any objective evaluations because the results are hard to believe and also because when I see the actual trades they are counter-intuitive and some of them very stupid to me. And despite that the equity curve is nice.
But that can be explained easily by the fact that the market according to Ehlers has two basic modes cycle mode and trend mode.
In this expert ea we use only the cycle mode, and that is logic that it would counter the trend. But despite that it is able to generate a positive equity curve. Hmm.
So it is possible to make two things.
1. We can implement the Ehlers filter. He uses the Instanetious trend line combined with cycle period evaluation in order to classify between trend mode and cycle mode.
2. We can see easily that the expert is performing greatly within choppy periods and high volatility swings but fails to perform into hard unidirectional movements.
2.1 So a manual intervention would be needed.
2.2 Or this expert may be used to hedge the risks of another expert that performs greatly in hard unilateral movements.
So that is all for the moment. I will release the expert pretty soon so you can test it.
It will be available for logged users for free (as usual).