Zigzag indicator and neural nets - page 4

 
eugenk:
SK, I agree with that 100%. And in general I see that zigzags are usually used by fans of Elliott Waves. And to this theory I treat, to put it very lightly, quite cautiously. However, the zigzag has no equal in data compression. I think you have to admit that too.


I think you have to separate the wheat from the chaff.

Elliot waves are one thing. Whether they're good or bad... He says they are, others say they aren't. I, for one, am inclined to think that one should only speak of a trend, but not demand that it necessarily exists. It's more a matter of faith or more (more strictly) a matter of statistics.

A zigzag, on the other hand, is a way of graphically representing some characteristic of the market. Speaking of recognition, the Zigzag can show the required characteristic with 100% accuracy, namely, the presence of a sequence of extrema of a certain value in the past history (which are wrongly called fractals). The point is that the data set obtained as a result of Zigzag does not improve the initial set of quotes at all, but imparts two disadvantages - coarsening and lagging.

Moreover!
This "compression" - I don't know what it is. If we're talking about averaging, there have been different MAs for a long time. And Zizag is so specific... On one hand it leaves a rectilinear angular trace as if smoothing "noise", but on the other hand it builds its extrema at hya and lows of single candles without taking the neighboring candles into account.

In my opinion, there is only one characteristic of the Zigzag - the clearness of extrema. And it is this visualization that leads astray new traders that are not aware of the fact that this picture is a reference to the past history and is not useful for forecasting. Simply to understand this, it is necessary to churn out Zigzag on dozens of forums for a long period of time, gaining experience in a very bestial way - by application and observation. Yet, nowadays it is possible to act more effectively - to think for an hour or two and understand what is what, not to spend months or years on this process.

 
SK. писал (а):

Elliott waves are one thing. Whether they're good or bad... He says they are, others say they are not. I, for one, am inclined to think that one should only speak of a trend, but not demand that it necessarily exists. It's more a matter of faith or more (more strictly) a matter of statistics.

A zigzag, on the other hand, is a way of graphically representing some characteristic of the market. Speaking about recognition, the Zigzag can show the required characteristic with 100% accuracy, namely the presence of a sequence of extremums of a certain value in the past history (which are wrongly called fractals). The point is that the data set obtained as a result of Zigzag does not improve the initial set of quotes at all, but gives it two shortcomings - coarsening and lagging.

What's more!
This "compression" thing - I don't know what it is. If we're talking about averaging, there have been different MAs for a long time. And Zizag is so specific. On the one hand it leaves a rectilinear angular trace as if smoothing "noise", but on the other hand it builds its extrema on the hi-jacks and lows of single candles without considering the neighboring candles in the calculation.

In my opinion, there is only one distinct characteristic of the Zigzag itself - the visibility of extrema. And it is this visualization that leads astray new terriers who are not aware of the fact that the picture represents the past history and is not useful for forecasting. Simply to understand this, it is necessary to churn Zigzag in dozens of forums for a long period of time, gaining experience in a very bestial way - by application and observation. But nowadays it is possible to act more effectively - to think for an hour or two and understand what is what, not to spend months or years in the process.


As disadvantageous as the ZigZag is, it has one undeniable advantage: the ZigZag can be constructed for any sequence of numbers. All you have to do is specify just one number - its sensitivity. The result is an interesting graph that graphically represents two poorly formalized notions of Forex - a trend and a flat. Based on this chart we can give a very specific meaning to the concept of "wave". Therefore, you do not need to be Eliott to be sure that waves exist. And it is not Eliott's understanding of the wave nature of price movement that remains on his conscience, but only the determination of specific patterns of the 5-3 type. Using modern MT4 tools everyone who is proficient in MQL4 can write a simple script (like the one written by eugenk) to see the statistical picture and understand how Elliott structures dominate the market. For some reason EWT supporters don't do that. Maybe out of fear of questioning their faith ? :-))

I have a question concerning the suitability for prediction. What do you think is suitable for making forecasts in Forex? If source series - the price - by definition is not predictable, what can be said about all tools that are derived from it?

 
Yurixx:

What do you think is suitable for forex forecasting? If the original series, the price, is by definition not predictable, what can you say about all the instruments that are derived from it ?

I do not agree that the original series is not amenable to prediction.

Answering formally, perhaps one could say that no single instrument is amenable to prediction. I think it is not quite correct to put the question in this way at all. It is not about tools. They merely reflect some mathematical characteristic.

For forecasting, you need to find the parameters that characterise the market and exploit them. For example, repetitiveness, cyclicity, periodicity, fractals. I.e. it's necessary to identify dependencies to which the market is subject to some extent. And when this is successful, then, at the stage of strategy implementation into a program code, one or another tool or a combination of them can be used, and not necessarily from the list of standard ones. Not all market laws are described by simple tools.

 
Yurixx, that there are waves, large waves add up to small waves, etc., is simply the theory of decomposition of functions into series. Elliot hasn't said anything new here. This is obvious to anyone who knows about Fourier series. However, Elliott was rather referring to a wavelet decomposition, for his wave patterns are localized in time. Confusion with all this begins when the universality of these very patterns is postulated and when the 5-3 law is introduced that leads to the Fibonacci sequence. The first is plausible. In fact all the charts except for the minute one look so similar that they are nearly indistinguishable by eye. Though it too must be justified in some way. The 5-3 law, on the other hand, is something completely incomprehensible. Why not 4-2 and 7-5? The impression is that Elliot knew about golden ratio and Fibonacci numbers, and chose such wave numbers in the law of combination quite deliberately. So I don't believe in it, for where it comes from is not explained in any way at all. At the same time it is very possible that some patterns do exist. Most likely they are not too obvious and change with time. It would be very interesting and useful for the family budget to look for them :)
 

I see the correct use of ZigZag, only to split the input stream, into peaks and troughs. I.e. getting classes to which to try to tune the NS (train the network to recognise these classes). Not to feed it to input, because zigzag at time t=0, can not help in recognition (trough or peak). Here is a picture, see it in another thread

Fig. 1

You can make it more complicated and recognize 4 classes, peaks and troughs at 5 min and 15 min, etc.

Fig.2

You can also try different currencies. Only with increasing classes, it will be more and more difficult to draw red lines as in Fig.2. In this case, perhaps Voronov's diagrams can help.

To Alex-Bugalter

... I'm afraid you have not quite understood me. When one speaks about zigzags one automatically means the wave theory...

It's amazing how you do it. I say one thing and you automatically imply another :-).

If by wave theory you mean the monograph "The Wave Principle" by Ralph Nelson Elliott, published in 1938. That's not what I meant at all. Moreover, I would not call it a theory at all, but rather a set of postulates. A theory is a bit different.https://ru.wikipedia.org/wiki/%D0%A2%D0%B5%D0%BE%D1%80%D0%B8%D1%8F It was his followers who raised a set of postulates (see the title of the monograph "...principle") to the rank of a theory. And strictly speaking, it is not a wave at all, but an oscillation, so look at how they differ.

https://ru.wikipedia.org/wiki/%D0%92%D0%BE%D0%BB%D0%BD%D0%B0

 
Prival, I don't know, I have little interest in peaks and troughs per se. I don't know, Prival. I'm not interested in the peaks and bottoms themselves. I think that the more interesting question for the network training is the time of closing a position with a profit of N pips. For the GMT on EURUSD I think N should be 25-50. My experiment with zigzag on Matlab shows about 40. If I could get such an estimate from the network, I'd be glad. And where the price will go after closing the position, I think it's already unnecessary. If you see that the next forecast is also favourable, do not close. And if all goes well - take a hit. It's a question of tactics, not strategy. By the way, not to reproach the bandits, many people confuse tactics and strategy here.
 
"It can't be, because it can never be". And what do respected SK. and Yurixx say to the existence of respected nen, who, judging by his published balance charts, trades extremely successfully on the real, using some ZZ patterns (Gartley, Pessavento) and Fibo tools to predict suppo/resistance levels - without any NS?
 
eugenk:
In my opinion, the more interesting question for network training is the time of closing a position with a profit of N pips.

I wrote here 'Random Flow Theory and FOREX' that you can't set NS to do that. Sorry, but from my point of view it's at least a waste of time. Quotes don't have the property of taking a profit of N pips. It is a property of the trading system you created.
 
Mathemat:
"It can't be, because it can never be". And what will respected SK. and Yurixx respond to the existence of respected nen, who, judging by the published balance charts, trades extremely successfully on the real, using some ZZ patterns (Gartley, Pessavento) and Fibo tools to predict levels of support/resistance - without any NS?


I don't know who trades how or on what basis the technology is built.

Fibo is quite different. Fibo has a pre-recognition that if there were 4 waves, then wait for the fifth (with high probability). Fibo is useful, because these tools have a predictive component (reflecting a pre-recognition of a discovered market property).

But the zigzag is not. Only it has an illusory excitement that here they are, the tops. But in fact, it is just a graphical representation of yesterday's actual events and has nothing to do with tomorrow.

I also think that we should not contrast these tools with neural networks. In the general case, of course, NS has a better chance, because NS can detect a market property that cannot be computed in a regular way with other modelling methods, but only by chance, if you are lucky.

 
Prival:
eugenk:
I think the far more interesting question for network training is the time to close a position with a profit of N pips.

I wrote here 'Theory of Random Flows and FOREX' that you can't set NS to do that. Sorry, but from my point of view it's at least a waste of time. Quotes don't have the property of taking a profit of N pips. This is a property of the trading system you created.
I agree. The market rather has the property to correct depending on the size of the previous trend. That's why the profit should be calculated as a floating value depending on previous events. But not a predetermined N. The desired value should not be N, but a function of the dependence of N on the nature of market development in the last history.