ASCTrend system - page 120
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Statistical instruments versus DSP
There is a strong belief that the more advanced digital filters are better. OK but why and how?
Here I will try to make some hypothesis that will try to explain what is happening.
Here is the story. Recently on a rival forum forex factory, there has been a challange of the strong belief that the digital filters are better. The challange came from the very contrarian and intelligent fajst_k. He showed that the Ehler's filter do not give an edge to automated trading strategies. (Please note that I mean "automated", those digital intruments give an edge in the manual trading giving us better instruments for manual trading)
This challange had not an adequate answer. I tryed to repeat the experiment. I used a random strategy with a crossing over of Simple Moving Averages versus crossing over of a JMA. The SMA without optimization was able to outperform the JMA, after optimization the JMA had an edge, and this in not surpising (I used Neuroshell for the tests).
The ideas is as follows. The SMA is a statistical instrument. It is a better instrument to track for a global statistical solution. The JMA and the other digital filters are really almost unable to track a global solution using the same strategy of crossing-over.
On the contrary they are better at observing the local characteristics of the market. And they are better instruments for optimization.
Please repeat the results I may be wrong. I did a simple test for a crossing over of SMA versus JJMA. The most of the SMA strategies were large green islands on the optimization matrix in the Metatrader. The digital filter gave me some peaks, all the others places were loosing.
That leads to a conlusion that the digital filters cannot be used in the same way as the statistical instruments. They are different, they are not better, they are different. This time I used Metatrader with the JJMA, becaseue Neuroshell does not give us a clue what are the results of the optimization compared to other results. The results for the short sample were similar, when I expanded the sample the results were a disaster for the JJMA, and the SMA still holded its statistical edge.
On the other hand the digital can be and are used at a great benefit in much more complex strategies, where the SMA and the other similar instrumens are highly inefficient. The digital filtes require much more complicated approach to make them useful. and you cannot just transpose a strategy with a SMA to a JJMA and hope that it will be globally better. The digital filters require an innovative approach.
So let go back to the ASCtrend. We know that the stop line in our open source version is based on a simple SMA cross-over between Low SMA = 9 and High SMA= 18+Risk.
When we use a genetic optimizer you can find here for free you may tests for a global solution that will give you better results.
After that the signal is based on the oscilator WPR. Do you see this is the essence itself of a trend following strategy. This is the best our fathers and grandfathers did in the XX century.
We can apply the same approach but making it a little more optimized using the things that are at our hands.
Ok we saw that for the stop we have to look for a global solution with a genetic optimizer.
As for the signals based on the oscilator we have a choice.
We can also try to male a global solution. Or we can try to make a local solution. Here tests are required.
As for the local solution we can combine an optimized local solution of the Signal that will be in the same direction as the global solution. So we can add the probabilites of two independant calculations and methods. AND THAT IS OUR MATHEMATICAL EDGE.
So let repeat again:
The old version of ASCtrend stops is not an inferior to the digital version. The digital ASCTrend is better in trying to catch local particularities of the market.
(Just an hint when we use a daily frame if we optimize we always optimize a short term sample, bacause we do not have a good longterm sample of daily or weekly bars, this comes easy to me, but people confuse, the sample is a statistical notion and has nothing to do with the length of the period we are exploring).
So when you see nice daily results that does not means that the system is profitable daily. That simply means that we have used a little sample to draw conlcusion and that may means nothing really. So for a daily chart the digital version may be better. For an hourly chart we have now a sample and we can look for statistical meaningful resluts (here we have to be afraid of the Black Swan phenomenon and that iswhy we have stops)
The new digital version of stops Astrend is at best optimized for the local caracteristics of the market and will be used:
- as a complement to the short term optimization of Asctrend signal
- as a complement to the long term optimization of the signal ASCTrend
In that way we can expand the possibilites of the system by adding an independant instrument to the system.
The digital smoothing of ASCtrend may enhance our possibility to smooth the noise. We can explore more options that may be profitable or not.
A link for an expert that can be used for statistical meaningful results for the stop line ASCtrend:
https://www.mql5.com/en/forum/general
Low SMA = 9 and High SMA= 18+Risk.
We can vary the high SMA and we will get results very quickly. Remember the genetic algorithm is like a leaf, with a leaf you can know a lot for the tree.
Another important thing. The map is not the territory.Our concepts how powerful they may be are just maps they are not the territory. This is another topic in which I will not enter for now.
Remember in our version the Fast Moving is always 9. I challenge the validity of this concept.
We can use the genetic optimizer of Metarader to look How Not to Fall into Optimization Traps
I have modified this expert to use the jjma instead. You will see by yourself what I mean, sorry I will not attach it now.
So
- use the SMA version looking for a global solution.
- use the JJMA versio looking for a local solution. Please if you find a way to find a global solution with jjma please let us know.
And finally a must read article:
How Not to Fall into Optimization Traps? - MQL4 Articles
ASCTrend and trend following strategies today
Hey, We have a new version of the ASCTrend with digital and other smoothing. This feature may change drastically the characteristics of the system itself. That means that when you change even a small thing in a system that will change drastically the behavior of the whole system itself.
This is a butterfly effect. DO NOT EXPECT THE DIGITAL FILTER SOLUTION TO BE BETTER AUTOMATICALLY.
In fact it is fun The digital version of the ASCTtrend led to a speeding of the stop line and making more reactive.
The digital version of the ASCTrend signal led to a slowing down the signal smoothing a lot of noise.
The digital ASCTrend stops have lost its statistical robust qualities and gained in the possibility to track the local market conditions.
The digital ASCTrend signal may be more able to look for a statistical solution because we can eliminate a lot of noise.
The Fractal ASCTrend with FRASMA is a compromise between a statistically sound solution and a tracking of the local particularities of the market.
Please do not expect that the fact you have inserted a yurik filter inside that will transform a good system into a holy grail system. Even the inverse may be true.
So let go back to the basics.
Why this system ASCtrend works?
Basically originally and still is, this system is a trend following system. The Research and Development on this forum made a new ideas how to use it in a short term basis. This was possible by adding additional filters.
The ASCTrend system works because it is a trend following system, and the trend following system normally work, they are sound systems (they are characterized by how profit/risk ratio and a lot of false signals below 50 % ). If the Trend following concept ceases to work we may ask a lot of questions of the future of the markets.
On the contrary the trend following concept works better and better, the markets have become increasingly volatile.
Look at the Euro but at a weekly graph look at it. It has become really unpredictable where it will go, and how long it will take. The recent trend in September in the Euro was amazing and it moves really fast. That makes the politicians really uncomfortable because they are not able to adjust the macro policies and the economies as a response of the possible adverse effects caused by the quick changing rates of the currencies. Their task at a macro level has become really difficult. Their control is very expensive and uncertain if they try to interfere directly with the Forex markets, look what happened lately. I have a theory of it based on the rise of the machines in the market. Basically it is a Robot psychology, as continuity of most of Asimov's robot stories.
They fear the volatility, but not the intra-day volatility. They really fear the powerful short term trend (6 months or so). Those powerful streams force them to adapt to it and once they have adapted the stream has reversed and they have to adapt again their macro policies.
Optimized digital Asctrend versus normal
Here I post the results after the genetic optimization process.
You can see a clear difference between the simple SMA results and the JJMA results.
In fact all the SMA results are positive and green and there are a few results of the JJMA that are good.
This is a simple cross-over strategy. For a very short period from 01.11.2010 to 10.11.2010 for a 15 m. time frame.
I can challenge the hypothesis I made in the previous posts that the SMA a better at tracking the statistical sound results and the JJMA is good at finding locally optimized results.
The reverse hypothesis will be that the JJMA is exploring just a bigger sample of possibilities and it gives a better picture of what is going on.
Where is the truth?
ASCTrend with no low MA limitation
Here it is.
JJMA cross-over expert and another version of AsctrendStop
Here I attach a new version of ASCTrend sig. This time I free it from the original settings that the low period is 9. We can set up whatever period we want. Just a hint you need to calculate the slow period because
High period = 18 + Risk
So you have to manually calculate the risk as a result from the genetic optimizer.
Risk = High period - 18; LOL basic mathematics
I attach a version of Universal MA with JJMA.
You can see how it looks the digital ASCTrend stop with genetic filter optimization. Here we use a simple stop and reverse strategy. The genetic optimizer found out a set of solutions requiring longer term filters.
I really do not like the results. No obvious pattern appears, the solutions appear at first look randomly distributed without clear and large green islands are missing.
However at a closer look look at the bottom line. The set of lower average period of 9 gives a set of results with high period of 24 to 31. Still the 9 lower period is the best.
Disclaimer:
Guys I am not a programmer I can hardly understand a code I just replaced some basic functions in a code.
Hi I have added the following changes:
Digital ACTtrendsig with freedom of choice of the limits of the oscillator.
The overbought level is 66:
The oversold level is 33:
Why this let change it to what we want.
High_level 66+RISK;
Low_level 33 -RISK
We may want to use other settings, why limit us with the levels.
The other change is the way to change the risk.
Originally:
value10=3+RISK*2;
value 11=value10
I changed it to
value10=4+RISK;
And the WPR period is equal to value11=value10
In fact we are going to use a digital filter. We may want to smoothly change the WPR period to explore more options.
And when we change the levels the results may be drastically different. This is a butterfly effect. You change a small thing but the consequences are big.
I cannot say that this is better, this is different.
ASCTrend New Digital
Hi Folks,
I was back testing the digital mod and I was not really satisfied.
The reason for this is that the cross-over strategy is not really a very good strategy for a digital filter. I wanted something simple as I am not a coder.
So I remember that the color coded versions of jjma look very sexy. Why would we need a cross-over when the change of color work so well.
Then I thought of something easy. I changed the settings. I wanted just one filter not a cross over of two. So a temporary solution is to use the settings of the filter and to make a cross - over with itself but with shift of one bar. Visually it makes sense and I am sure we can find a statistical sound solution for that mod. As you remember I was not able to find any sound statistical solution of the jjma cross-over system.
OK I think that this is enough. Look at it yourself.
Another example
In fact the idea is that we do not need a cross-over of digital filter to generate signals. This idea does not help us. We have to optimize two parameters the length and the phase. I think that one digital filter is better and gives us more consistent results. Visually it is the color coding of the filters. It is better than a cross-over of filters. Everything has to be tested but in practice I have better and more consistent results with this mod, because it is more consistent with the digital filters character. We have more consistent stops. I do not mean that the normal mod is bad but this is different.
I give an example of the normal jjma based on the cross over and the new digital mod.
I use jma of 26. The normal mod has the same smoothing of the slower filter.
Great System
Thanks for all the R&D every one. I am following this system with great interest. I have been forward testing the EUR/USD with some very good results.
Thanks Much!