The ASCTrend is giving down trend. I believe that the use of this wonderful Trend following system trying to time choppy market for scalping is very difficult without the assistance of other instruments.

We have a a high fractal dimension the iVAR is above 0.5. Even if there is established trend the iVAR shows the market conditions at the current time frame of reference.

The automatic trend channel is giving us the support and the resistence and the neural net is trying to make some kind of predictions there. It got it right, the last time, compare with the previous shots but definitely it is not a signal generator. I consider it as an independent opinion.

Here in this case we may think that the price is in the up part of the channel and we can go short. However the Neural net is telling us that the price can stay there for a while. Of course a break - out cannot be predicted by this instrument.

Here I show a screen shot if the ideas of the intended use.

As New Digital mentioned on many occasions the ASCtrend does not give by its design a signal at the beginning of the Trend (as e.g. the Brain Trend does).

Here our main concern is:

The Trend is it going to continue after we have a signal?

We have several possible answers.

1. We need a good filter which will give us the general movement of the trend. Here I use the FullSSA normalize. if it is above zero go only long, if it is below zero go only short.

This is a very advanced indicator. Use open prices if you do not want to make calculations for every tick (this may be very demanding for the PC).

Use enough amount of lag (30 is a critical minimum). It may recalculate, so it is not really for use in timing tops or bottoms, but it is a great trend filter.

2. We can use the iVAR, which calculates the fractal dimension.

When it is below 0.5 (persistence: the next movement is more probable to be in the same direction that the previous) it gives us a Trend probability .

When it is above 0.5 (anti - persistence, the next movement is more probable to be in the opposite direction) we have range probability.

A good use is to look simultaneously several time frames, e.g. 15, 30 are critical for the day-trading.

However we have an estimation of the persistence, or anti-persistence of the price series. And it should be given a strong emphasis here that there is not a direct relation between those two mathematical notions and the Trend as it is defined by the Technical analysis (higher highs and lower lows.)

3. Additional indicators.

-We have a linear regression channel tuned to 1.5 of dominant cycle. This channel is giving us immediate support and resistance lines.

-Neural net tuned again to 1.5 of dominant cycle. Here we have a prediction.

This is an interesting concept of different use of the Nyquist interval of 1,5 of dominant cycle for linear regression and Neural Net. This is a new idea I hope we can explore together its potential.

How we can use this together?

Her we have a 15 m. time frame. For this frame I used the stochastic with adaptation to 1 dominant cycle (when the time frame is really low I prefer those settings, because the 0.5 of dominant cycle is much more reactive and gives us too many signals however, we can slow it down by adding risk levels or by adding more digital smoothing).

We have a long signal. But is it a valid signal?

1. We have a linear regression channel, which is horizontal. Well it looks we are in a range mode.

Inside the channel the Neural Net (BPNN adaptive) is giving us predictions.

2. The SSANormalize is giving us permission to go long.

3. However the fractal dimension is giving us a clearly antipersistent market conditions. Well we need a iVAR below 0.5 to take a trade.

Conslusion we have a range mode. That means we expect a break - out. Of course we expact that the break - out would be in a direction of the major trend. However you need a considerable stop to use that strategy.

When we go to the 30 m. time frame.

Again we have a contradiction.

1. We have a clearly down channel trend, with a Neural prediction to go short.

2. However that is not confirmed by the SSANormalyze which indicates an up momentum.

3. The fractal dimension is high again.

We have a probability of High Phase Space singularity

This is a particular market state: High fractal dimension on both 15m and 30 m frame.

The phase space is very complex and the price is going up and down. The directional indicators usually do not work in those environments. We have a lot of pink noise who eats the close stops like piranhas.

Please look at the 7 of january of 14:45. That is a typical example of High Phase Space Singularity combined with a lot of volatility.

Again we have a high fractal dimension in two key levels 15m. and 30m. The phase space becomes very complex. Nobody can have a correct solutions. The human traders can't, the automatic algorithms cannot do so either. I call it a singularity because the price action is really strange.

This is the opposite of the Low Phase Space Singularity. Then all the automatic algorithms are able to get the correct solutions and they start driving the price like crazy. This singularity begins by a fractal break - out, but not every break -out leads to singularity. Here on the 30 m shot you can see the fractal - break out just after the blue rectangle. You will see how the fractal break - out confirms the technical break - out. This is a very powerful pattern.

Here We see how the neural net made a correct prediction of the price action inside the adaptive linear regression channel. Keep in mind that because it is constantly adapting and recalculating this cannot be used as a signal, we use only as additional direction estimate for the ASCTrend and for better timing in direction of the signal.

The iVAR gave us a correct estimation of the price action activity conforming to the mathematical estimation of the Hurst exponent. When we have a high fractal dimension we should apply statistical methods like support and resistance, linear regression, Bollinger bands etc.

Really it is like a light in the tunnel, when you start to understand it you cannot trade anymore without it. It is ADDICTIVE.

It helps us to achieve a market reading as if we had many years of market experience (the experienced traders read the fractal characteristics of the price even without recognizing it, they can make differences impossible for a newbie who has learned for a month all the patterns of the Technical analysis). The experienced traders feel the fractal characteristics of the price action and this is a major part of their intuition. We can replace that by a mathematical estimation.

And every market has its own characteristic fractal dimension, for example the Euro/Usd has a different Hurst exponent than the GBP/USD.

PS There are red arrows that are the product of the SSA_Normalize I have deleted some of them that is why it looks like they are repainting.

Their idea is to record a change of the momentum of the SSA_Normalize. Usually I do not consider them but sometimes they can be astonishingly accurate. I have longtime hesitated to delete them from the code but I let them as they were.

Just for fun look at the 5 m time frame. It is interesting how when we have low iVAR we tend to have bigger movements and vice versa.

It is really funny to see haw the BPNN gives me a prediction how the price would hit the support level and jump back.

You see clearly that even if the signal is good when the price action is anti persistent there is not a lot of room left for profit to the other end of the channel. Usually in those kind of market conditions not only ASCTrend but every directional indicator performs badly even if it is with no lag.

And that is the reason when a fellow trader sees a system that gives awesome signals. He starts trading it. Awesome. After that we have anti - persistent market state. The guy starts to use his awesome system and gets stopped out many times. He says Shit, I will find another system.

He finds a ranging system that performs great. After that the market change again and he start to take falling daggers when the market get persistent again. After that he founds another trending system (he does not believe the previous).

And this is a never ending story, we have all been experiencing.

Here I add my alma mod of the BB Caterpillar Squeeze on Alma.

I particularly like this mod. This time the Histogram is based on Alma and the dots are the Caterpillar squeeze. I used the open price for the Caterpillar, as I do not want it to recalculate for every tick, that can hurt some weak PC.

I also add the normal Squeeze with Alma. In the normal mod you do not have any Caterpillar (SSA). You have the normal squeeze with green bars. Look at the video to see how it has to be used.

How you should use it. Well the green dots of the normal squeeze are the main way to distinguish the trend from the range.

However if you use this instrument only on time frames wherein you actually have low fractal dimension you will have better results. This is highly opportunistic approach, for example today only on 1 m time frame we had low fractal dimension and the best signals were on 1 m time frame.

This can be used as it is. The Alma slope is giving the general direction of the trend. And the Caterpillar dots are giving the change in the momentum.

Other way is to use as a slow trend filter in other strategies. The choice is yours.

However I do not like the idea of the squeeze even if it works. It does not make sense to me. The idea is to compare the Keltner Channel with the Bollinger Band. Inside the code we compare the ATR with the standard deviation. When the standard deviation exceeds the ATR (multiplied by a coefficient) we have trend mode and vice - versa. The code uses some kind of Keltner multiplier to make this works. However this to me looks like comparing the fruit Apples with the I pod. Those are different things to me.

However the idea is really nice. Take a look at this post.

The idea of Jean - Philippe is on the screen shot. We compare the normal Bollinger Bands with the Fractional bands. When the 2 standard deviations of the fractional bands exceed the 2 standard deviations of the Bollinger bands we have a transition of the probabilistic distribution of the market.

Unfortunately I cannot make it works so a little help by a coder will be welcome.

The market then cannot be approximated by the Gaussian distribution, we may have an infinite variance.

Hi John Last, great thread. Let me ask you a question, you said that each pair have different Hurst Exponent EUR/USD have 1.5, how you calculate it for each pair? Are you use it to set to lets say to 1.5 of dominant cycle ?

There is a nice tutotial. The Tisean software is much more complicated.

I did an experiment I calculated the Lypapunov exponent of the Eur/Usd. As expected it was slightly positive. However I found out recently that I did a wrong use of the software and all my estimations were corrupted.

It was fun I calculated the Pearson correlation between EUR/USD and GBP/USD.

After that I wanted to know what is its Lyapunov exponent of its variation in the time.

The results showed that the variation sometimes is really deterministic and sometimes more chaotic than the time series themselves. That is why I am a little bit skeptical for inter-market Neural net models. Sometimes we have a strong correlation (you really have a linear dependance) and sometimes it is a complete chaos that the neural net could not solve.

I am following your thread at moment. Taste some indicators from you. But, May I ask a question? Is real cycle length from output the indicator of CyclePeriod? Is it good for NN when length of training data set is changed too fast? Thanks.

Use

Here I add how we can use this.

The ASCTrend is giving down trend. I believe that the use of this wonderful Trend following system trying to time choppy market for scalping is very difficult without the assistance of other instruments.

We have a a high fractal dimension the iVAR is above 0.5. Even if there is established trend the iVAR shows the market conditions at the current time frame of reference.

The automatic trend channel is giving us the support and the resistence and the neural net is trying to make some kind of predictions there. It got it right, the last time, compare with the previous shots but definitely it is not a signal generator. I consider it as an independent opinion.

Here in this case we may think that the price is in the up part of the channel and we can go short. However the Neural net is telling us that the price can stay there for a while. Of course a break - out cannot be predicted by this instrument.

Files:Respected JohnLast, I can't open BPNN, maybe U can help

Bpnn

Hi, There is a RAR file.

Extract the file into a normal dll file.

This file need to be installed in the experts/library folder.

This is a mql base indicator

Next price predictor using Neural Network - MQL4 Code Base

Please read the thread there. This is a working neural net implementation for Metatrader.

However it keeps crashing all the time. So a third party fix was prepared and I have posted the fixed version.

BPNN dll temp fix - MQL4 Code Base

My contribution is to constantly adapt it to the cycle period. I am still testing it. Sometimes it gives really interesting things.

The next idea is to use the "output" into other indicators.

In fact you will have an indicator which will show you what is predicted to the close of the opened bar.

Until now many uses: FullSSA with this looks promissing (I call it FullNSSA), however I am still testing it.

earned, I Founded working dll

Interpretation

Hi,

Here I show a screen shot if the ideas of the intended use.

As New Digital mentioned on many occasions the ASCtrend does not give by its design a signal at the beginning of the Trend (as e.g. the Brain Trend does).

Here our main concern is:

The Trend is it going to continue after we have a signal?

We have several possible answers.

1. We need a good filter which will give us the general movement of the trend. Here I use the FullSSA normalize. if it is above zero go only long, if it is below zero go only short.

This is a very advanced indicator. Use open prices if you do not want to make calculations for every tick (this may be very demanding for the PC).

Use enough amount of lag (30 is a critical minimum). It may recalculate, so it is not really for use in timing tops or bottoms, but it is a great trend filter.

2. We can use the iVAR, which calculates the fractal dimension.

When it is below 0.5 (persistence: the next movement is more probable to be in the same direction that the previous) it gives us a Trend probability .

When it is above 0.5 (anti - persistence, the next movement is more probable to be in the opposite direction) we have range probability.

A good use is to look simultaneously several time frames, e.g. 15, 30 are critical for the day-trading.

However we have an estimation of the persistence, or anti-persistence of the price series. And it should be given a strong emphasis here that there is not a direct relation between those two mathematical notions and the Trend as it is defined by the Technical analysis (higher highs and lower lows.)

3. Additional indicators.

-We have a linear regression channel tuned to 1.5 of dominant cycle. This channel is giving us immediate support and resistance lines.

-Neural net tuned again to 1.5 of dominant cycle. Here we have a prediction.

This is an interesting concept of different use of the Nyquist interval of 1,5 of dominant cycle for linear regression and Neural Net. This is a new idea I hope we can explore together its potential.

How we can use this together?Her we have a 15 m. time frame. For this frame I used the stochastic with adaptation to 1 dominant cycle (when the time frame is really low I prefer those settings, because the 0.5 of dominant cycle is much more reactive and gives us too many signals however, we can slow it down by adding risk levels or by adding more digital smoothing).

We have a long signal. But is it a valid signal?1. We have a linear regression channel, which is horizontal. Well it looks we are in a range mode.

Inside the channel the Neural Net (BPNN adaptive) is giving us predictions.

2. The SSANormalize is giving us permission to go long.

3. However the fractal dimension is giving us a clearly antipersistent market conditions. Well we need a iVAR below 0.5 to take a trade.

Conslusion we have a range mode. That means we expect a break - out. Of course we expact that the break - out would be in a direction of the major trend. However you need a considerable stop to use that strategy.

When we go to the 30 m. time frame.

Again we have a contradiction.

1. We have a clearly down channel trend, with a Neural prediction to go short.

2. However that is not confirmed by the SSANormalyze which indicates an up momentum.

3. The fractal dimension is high again.

We have a probability of High Phase Space singularityThis is a particular market state: High fractal dimension on both 15m and 30 m frame.

The phase space is very complex and the price is going up and down. The directional indicators usually do not work in those environments. We have a lot of pink noise who eats the close stops like piranhas.

Please look at the 7 of january of 14:45. That is a typical example of High Phase Space Singularity combined with a lot of volatility.

Again we have a high fractal dimension in two key levels 15m. and 30m. The phase space becomes very complex. Nobody can have a correct solutions. The human traders can't, the automatic algorithms cannot do so either. I call it a singularity because the price action is really strange.

This is the opposite of the Low Phase Space Singularity. Then all the automatic algorithms are able to get the correct solutions and they start driving the price like crazy. This singularity begins by a fractal break - out, but not every break -out leads to singularity. Here on the 30 m shot you can see the fractal - break out just after the blue rectangle. You will see how the fractal break - out confirms the technical break - out.

This is a very powerful pattern.Files:Some minutes later

Here We see how the neural net made a correct prediction of the price action inside the adaptive linear regression channel. Keep in mind that because it is constantly adapting and recalculating this cannot be used as a signal, we use only as additional direction estimate for the ASCTrend and for better timing in direction of the signal.

The iVAR gave us a correct estimation of the price action activity conforming to the mathematical estimation of the Hurst exponent. When we have a high fractal dimension we should apply statistical methods like support and resistance, linear regression, Bollinger bands etc.

Really it is like a light in the tunnel, when you start to understand it you cannot trade anymore without it. It is ADDICTIVE.

It helps us to achieve a market reading as if we had many years of market experience (the experienced traders read the fractal characteristics of the price even without recognizing it, they can make differences impossible for a newbie who has learned for a month all the patterns of the Technical analysis). The experienced traders feel the fractal characteristics of the price action and this is a major part of their intuition. We can replace that by a mathematical estimation.

And every market has its own characteristic fractal dimension, for example the Euro/Usd has a different Hurst exponent than the GBP/USD.

PS There are red arrows that are the product of the SSA_Normalize I have deleted some of them that is why it looks like they are repainting.

Their idea is to record a change of the momentum of the SSA_Normalize. Usually I do not consider them but sometimes they can be astonishingly accurate. I have longtime hesitated to delete them from the code but I let them as they were.

Just for fun look at the 5 m time frame. It is interesting how when we have low iVAR we tend to have bigger movements and vice versa.

It is really funny to see haw the BPNN gives me a prediction how the price would hit the support level and jump back.

You see clearly that even if the signal is good when the price action is anti persistent there is not a lot of room left for profit to the other end of the channel. Usually in those kind of market conditions not only ASCTrend but every directional indicator performs badly even if it is with no lag.

And that is the reason when a fellow trader sees a system that gives awesome signals. He starts trading it. Awesome. After that we have anti - persistent market state. The guy starts to use his awesome system and gets stopped out many times. He says Shit, I will find another system.He finds a ranging system that performs great. After that the market change again and he start to take falling daggers when the market get persistent again. After that he founds another trending system (he does not believe the previous).

And this is a never ending story, we have all been experiencing.

Files:ALMA your friend beyond others

Here I add my alma mod of the BB Caterpillar Squeeze on Alma.

I particularly like this mod. This time the Histogram is based on Alma and the dots are the Caterpillar squeeze. I used the open price for the Caterpillar, as I do not want it to recalculate for every tick, that can hurt some weak PC.

I also add the normal Squeeze with Alma. In the normal mod you do not have any Caterpillar (SSA). You have the normal squeeze with green bars. Look at the video to see how it has to be used.

How you should use it. Well the green dots of the normal squeeze are the main way to distinguish the trend from the range.

However if you use this instrument only on time frames wherein you actually have low fractal dimension you will have better results. This is highly opportunistic approach, for example today only on 1 m time frame we had low fractal dimension and the best signals were on 1 m time frame.

This can be used as it is. The Alma slope is giving the general direction of the trend. And the Caterpillar dots are giving the change in the momentum.

Other way is to use as a slow trend filter in other strategies. The choice is yours.

However I do not like the idea of the squeeze even if it works. It does not make sense to me. The idea is to compare the Keltner Channel with the Bollinger Band. Inside the code we compare the ATR with the standard deviation. When the standard deviation exceeds the ATR (multiplied by a coefficient) we have trend mode and vice - versa. The code uses some kind of Keltner multiplier to make this works. However this to me looks like comparing the fruit Apples with the I pod. Those are different things to me.

However the idea is really nice. Take a look at this post.

https://www.mql5.com/en/code/8900

The idea of Jean - Philippe is on the screen shot. We compare the normal Bollinger Bands with the Fractional bands. When the 2 standard deviations of the fractional bands exceed the 2 standard deviations of the Bollinger bands we have a transition of the probabilistic distribution of the market.

Unfortunately I cannot make it works so a little help by a coder will be welcome.

The market then cannot be approximated by the Gaussian distribution, we may have an infinite variance.

Files:Hi John Last, great thread. Let me ask you a question, you said that each pair have different Hurst Exponent EUR/USD have 1.5, how you calculate it for each pair? Are you use it to set to lets say to 1.5 of dominant cycle ?

Selfis Software

Please check this post

https://www.mql5.com/en/forum/178285/page6

And the tread.

This is the Software

Thomas Karagiannis Home Page

There is a nice tutotial. The Tisean software is much more complicated.

I did an experiment I calculated the Lypapunov exponent of the Eur/Usd. As expected it was slightly positive. However I found out recently that I did a wrong use of the software and all my estimations were corrupted.

It was fun I calculated the Pearson correlation between EUR/USD and GBP/USD.

After that I wanted to know what is its Lyapunov exponent of its variation in the time.

The results showed that the variation sometimes is really deterministic and sometimes more chaotic than the time series themselves. That is why I am a little bit skeptical for inter-market Neural net models. Sometimes we have a strong correlation (you really have a linear dependance) and sometimes it is a complete chaos that the neural net could not solve.

Is real cycle length from output of cyclePeriod?

Hello, John,

I am following your thread at moment. Taste some indicators from you. But, May I ask a question? Is real cycle length from output the indicator of CyclePeriod? Is it good for NN when length of training data set is changed too fast? Thanks.