Hi all I started this thread to gather in one place the ideas for the next generation of ASCTrend.
This is necessary because this new instruments require appropriate testing as they are new. It is not possible to use new instruments in older strategies without verification.
So here it is the thread. I will summarize what is done until now.
1. The ASCtrend stops depend on moving averages and they can and were replaced with jurik filters.
One of the options was to use cross - over as the original.
Another option is to use the signal with a shift of itself. I think that with advanced digital filters the cross-over strategy is an obsolete idea(but still working, I want the signal line to be as close as possible to the filter line)
2. However the verification by expert shown that still the SMA beats the jurik in the long term when Jurik is used as cross-over. So what?
We could use tested SMA as global solution and optimized Jurik ASCtrend 2 as local solution.
Moreover the tests of the SMA cross - over showed that we have a stable profitable solution at the 1 h frame.
3. The ASCtrend 2 is considered as a module carrier. You can insert whatever filter you want into it. You can make your own mod.
Oh yes we had now a whole family.
I particularly like the AMA mod. and the SSA mod (you need three dots to minimize the recalculating)
-Take care when you apply a sufficient amount of lag you can have a non repainting solution, however there is always a catch, the solution you will get in real trading will differ from the solution when you apply on the graphic afterwards, if that was not the case we would have a Holy Grail and this is certainly not.
3. However I wanted to make something different with the Signal indicator. The idea is as follows. I used a digital filter, the jurik WPR from the Kositsin collection. And I applyed an adaptation according to the dominant cycle measured by Discrete Hilbert Transform.
I applied by error one full cycle and not a half of the cycle as it is traditionnal. However he results were good.
However I wanted something different. And I used an adaptive JStochastic instead of the WPR, and we had a system of a different kind. Much more reactive.
However I have working prototypes consider them as prototypes. Basically the R&D should not rely not on closed source algorithm as the Discrete. However the people need a choice as the Hilbert Transform is commercial (but hey it costs 10 Euro) (I used because it is the best I had available, and I use it for me).
And I say that without any intention of commercial promotion.
-we can use the Coronna system
-we can use the Mesa library from Richcap.
However this is a principle question because the knowledge is usefull when it is combined with another knowledge.
ASCTrend with cycle period adaptation
OK here it is a free version of the digital ASCTrend with adaptation by the cycle period of Ehlers.
You need the Kositsin jjma series to run the jma WPR.
The Len parameter is very important as important as the Risk.
The default is 8 but I would change it to 3. Feel free to experiment. Moreover the different time frames may require different settings of the Len settings (digital smoothing)
John, firstly I'd like to say you did a great job with ASCtrend. I've been watching this thread for a while and hopefully in time I'll take a closer look on it and support you in developing it. Neverthless, while spliting the main thread into two, some of the posts are gone. Could you once more post the indicator AscTrend on Stochastic.
Thanks in Advance
The TSD ASCTrend as said New Digital is a philosophy not just a trading system.
I want to present here my research in this domain.
OK the TSD ASCTrend originally relies on the Trend. The trend as a powerful force in the market. And the system is designed to read the market language.
So if you go to the daily frame you will see that the system is very profitable. Of course this is not the only system that can achieve similar results (e.g. Trend magic (Forex factory system) does the same, the Guppy moving averages (Dolly systems) etc.)
It should be emphasized that the traditional use of the TSD ASCTrend is working and it is theoretically sound.
Here on the TSD thread the system is used with kind of a TOP GUN approach. Really the system is used to its limits with a lot of stress tests.
However what is under the hood:
Basically the market language is read as two components:
1. Global solution (market voice): Cross - over of two SMA: fast = 9 Slow = 18+3*Risk.
2. Local solution (market whisper) : momentum, the WPR is used with thresholds.
Here the risk controls in the same time the thresholds x1 and x2:
So we want the WPR to excede those levels in order to enter into position.
x2=33 - RISK;
This is oversimplification but it is how it works.
The new Digital TSD ASCTrend relies on digital filtering of the components. So we can replace the WPR with jurik WPR the Moving averages with the family of jurik averages.
The results as reactivity are nice we are able to achieve a greater reactivity, nice looking indeed. However the tests showed that the cross - over of jurik averages is unable to produce as sound result as the SMA is. Really the SMA is really hard to beat.
With the WPR it is the same think it is difficult to beat the normal WPR because what matters are the thresholds.
However with the digital smoothing we are able to avoid some whipsaws.
The second idea is to use an adaptivity according to the dominant cycle.
The third idea is to replace the WPR with something else, like Adaptive Stochastic.
Unfortunately the best instrument I have is the Hilbert transform of Forex Digital Solutions. And it is a commercial, not expensive but commercial and the idea is that the system remains free ( I just posted the source of what I did, so you can do the same with FDS Hilbert transform if you wish).
That is why now we use the cycle period, because it is free and it is on the forum.
The second important point is that the digital TSD ASCTrend not nesseraly will have the same philosophy.
I have gathered and organized what is available here about the use of the Chaos theory . There is more to come.
TSD Digital ASCTrend shots
Oh here I will show some shots from this morning. The adaptive gave the following results, the risk is set to 0. The normal gave the following results, the risk is set to 10.
The adaptive is basically a non parametric tool (this mod) you can change the risk (this is the first prototype, maybe I should change the threshold levels).
In fact the change of the smoothing affects more the indicator than the Risk. This is the "Len" parameter. So you should know that as it is adaptive when the current cycle is lower it will reacts as a fast ASCtrend (so than you can add manually digital smoothing if you want more noise filtered).
And when trending we should not have counter trend whipsaws. That is the basic idea to get as much of the trend as possible and that logic is programmed into the indicator and we should rely as less as possible to arbitrary decisions.
The default is to 8, on this shot I add just 1 degree of jurik smoothing. Now I consider it too slow (maybe 3 could be the default).
I see that it reacts sometimes like a faster ASCTrend and sometimes like a slower ASCTrend all depend on its reading of the market conditions.
However there is a delay in the adaptation but this is common to all the cycle adaptation periods (at average 8 bars). The adaptive sometimes reacted as fast as the risk 3, and sometimes slower than the risk 10.
Digital ASCTrend on adaptive JStochastic
Here it is another working prototype. The heart of the indicator is not WPR anymore. This is specially designed for the fully devoted Stochastics fans.
It depends on Cycle Period and the JJMAseries (in the include folder) from the Kositsin library to work.
The default settings are set for 1 h frame. Feel free to experiment.
Do not forget to install also the PriceSeries.mqh file in the include folder.
I am thinking about using a Simple Neural Net which will tell me what is the probability of any given signal based on the current market conditions.
Some tests have shown that a Neural Net auto-trained according to 1.5 of the current dominant cycle has a greater probability hit rate than the same Net trained on more bars in the past.
This would be a leading filter. The best I can propose for trend filter is in the Elite section:
MAMA will take care of you
OK here is the Mama mod of ASCTrend. I particularly like this mod.
You need also the PriceSeries (but I am not sure that all the dependencies have been solved).
The theory is complex. The application is easy.
Look the iVAR, when it is below 0.5 the price movements are persistent, that means that the market condition is well suited for this type of system on on this time frame.
When the iVAR is above 0.5 the price movements are antipersitent. Do not expect too much. Go in a lower frame and search for persistent opportunities. Or use another system, or settings (the JStochastic allows much more reactivity).
This is non directionnal filter which would be the basis for the digital ASCTrend system.
what do you mean by autotrained?Do you mean trained by backpropagation?Do you have any sample code?Also is the Hilbert transform (commercial one) applicable all the oscillators?How do you compare it to Igorad's Corona cycle indicator in elite section or is it an advantage to use both?Here is a PNN code you may like it to use but I am not a coder.
Next price predictor using Neural Network - MQL4 Code Base
This is the link. I wanted to train it to exactly 1.5 of the current dominant cycle. And surprise, surprise the results were better.
So instead to use for example a fixed period of 300 bars I use the number of bars according to the current cycle period. I implemented cycle period. It works but there is something wrong and it crashes from time to time. However the 1h hour time frame is not so dynamic and you can do it manually without crashes of any kind. I did it because I wanted to see how it looks like in a simulator. I made some manual tests and here are the results not very impressive but still this can be of some use.
The idea is very basic if the Oscillators are able to be adapted to the dominant cycle why not a Neural Net. This was also another old idea when we used the polynomials and adapted them to 1.5 of the dominant cycle instead of using Oscillators. (I made an Adaptive polynomial it is funny).
I am not a programmer. But the programmers are busy people with a lot of projects. And I am forced to do thing by myself otherwise I would wait for years to do so many mods in so little time. And I have to do many experiments to select what would work.
However if I am able to make a prototype someone can see and make it better.
The idea is a simple one. All the systems are related. There is hardly an independent system. If someone is able to make a progress with one system this can be useful for the improvement of other systems.
Yes the Hilbert transform can be used to whatever you want. It takes two lines of code. It is nice to apply on Jurik. The same is for the cycle period.
Unfortunately I cannot make it work on Corona system. Maybe because the change in the cycle period is abrupt it does not work (I do not know in fact).
Basically those are my ideas. And I have to make a lot of combinations to see what may work at best:
Strategies for improvement:
Hybrid SSA srategies with a lot of lag: e.g. Trend magic on SSA
Hybrid digital strategies on jurik: You take anything and you add jurik filter: e.g. Trend magic on jurik, Trader's Dynamic Index etc.
Cyber on jurik to make it adaptive: Cybernetic Trend magic
Fractal adaptation on digital filters: this I will keep for myself for now
Hybrids of the hybrids: Yes it is cool. How it looks like :Fractal adaptation of Double smoothed Brain Trend with adaptation on the cycle period.
Adaptive Neural net on the cyber cycle:
-Input of all this into existing composite systems with volatility adaptivity. Trend magic, ASCTrend, Brain Trend etc.
-Whenever possible making multiframe indicators
e.g. Multi-frame Trend magic with digital filters, with SSA, with cycle adaptation of the digital filter.