Testing real-time forecasting systems - page 63
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
And why is that? After all, there is nothing but them.
I just don't know any valid methodology to analyse them. Not only is time non-stationary, but ticks come in whenever they want. From fractal analysis (not indicators) there is nothing to do there, ticks are very complicated system, plus there is a huge influence of DC on these ticks (remember winwin from the Championship). You cannot forecast them, you cannot analyse them correctly.
In general, your research is encouraging and admirable!
I am sincerely glad that I give you hope, but there is nothing special in my research. I am not a mathematician, rather an amateur in this field.
I just don't know any effective methods of analysing them.
That's right, I've always said you have to know your mushroom spots.
Not only is time unsteady, but the tics come in whenever they want.
It's an outrage! What are the Powers That Be looking at?
From the point of view of fractal analysis (not as an indicator) there is nothing to do there, ticks are a very complicated system.
And here, colleague, let me disagree with you. Not only the fractal analysis is created especially for ticks, but ticks are the simplest system. Very simple assumptions are enough for its modelling - there is practically nothing to simplify. A conventional OHLC representation is not 4, but 10 times more complicated than ticks, and you, dear, are dealing with it. I can imagine how hard it will be for the ticks when you do pounce on them. :-)
to Yurixx
Во-во, я всегда говорил - грибные места знать надо.
So share the locations, if indeed there are any. I really hope you're not out there picking fly agarics and toadstools.
This is an outrage! What do the Powers That Be do?
That's what makes our activities largely pointless :o) In all seriousness, there are simply no methods of dealing with such rows. Objectively, you can't even calculate the average, not to mention moments and other useful things (for example, there's no concept of time lag for such series). But if you have a knack for the fine art - let me shake your hand!!! I'm thrilled!!! Then you don't need to sulk at the Higher Powers - you are the Power yourself!!! :о)
Well, colleague, let me disagree with you. Not only the fractal analysis is created especially for tics, but tiki is the simplest of all systems. For its modelling very simple assumptions are enough - there is practically nothing to simplify. A conventional OHLC representation is not 4, but 10 times more complicated than ticks, and you, dear, are dealing with it. I can imagine how hard it will be for the ticks when you do pounce on them. :-)
Why are we using the "you" word, but okay. Actually, fractal analysis was historically created for quite a different purpose. Yury, there are not so many methods that can give an objective estimation of Chaos. One of them is to calculate correlation integral for each nesting and total Chaos estimation (element of fractal analysis). It's what I used, with some assumptions. The dimensionality of the system for ticks is huge, there's complete Chaos with a capital letter, there's nothing to do there.
I think that the method of calculation does not play any fundamental role. One should only remember that we are talking about informational entropy, not thermodynamic or anything else.
it has nothing to do with the dimensionality of space. Besides, dimensionality is just an input parameter to the model.
I don't think it's right or I'm missing something.
I am willing to restate. I would like to get to the bottom of it. The gist of the question is, does it always make sense to choose the "winning" forecast by the maximum value of its entropy.
Then the idea that entropy values should have cyclicity, because there are certain periods in the market (session boundaries, news), when entropy really rises, was just voiced here. This is something I agree with. Entropy = volatility, right? However, the direction of predicted movement (which is what determines a correct prediction) is unlikely to be accounted for in any way by entropy. In general, I would suggest selecting an implementation with an average entropy value of all.
Why would you get it wrong? This formula will give the lowest possible prediction horizon for a very complex system, of large dimensionality.
Correction. Not the minimum, but the maximum possible horizon. Still, I don't understand how the time in the resulting prediction interval is measured. We have bar samples in the data, and it would be logical to assume that the interval is also in bars, but I can't find an explanation for that directly in the formula.
to marketeer
Думаю, что если посчитать энтропию для данных, преобразованных в пространства разных размерностей, мы получим разные значения, так что размерность пространства имеет отношение к энтропии. Более того, мы специально делаем предобработку входных данных с целью повышения их энтропии. А уж то, что размерность задается параметром модели - это обычное дело, у меня тоже так.
I don't really know what you mean anymore? Before you were talking about K-entropy, which is used in the formula. It has to do with the dimensionality of the system, that's what I wrote about. I.e. I'm getting a bit lost in the main idea. :о)
Then there was the idea that entropy values should be cyclical, because there are certain periods in the market (session boundaries, news) when entropy really rises. This is something I agree with. Entropy = volatility, right?
I was writing about choosing different levels of entropy as a criterion, depending on time, not about the cyclicality of entropy itself. But maybe you're right, maybe there is a correlation with volatility (just need to clarify what it is, which is not easy :o) It's not that simple, I wrote earlier that I don't work directly with price, I use a transformed series that has the following properties:
It's a whole study, you can't just answer that.
Correction. Not the minimum, but the maximum possible horizon.
I don't get it...
So share the places, if there really are any. I really hope you're not out there picking fly agaric mushrooms and toadstools.
Yes, I have always expressed my fondness for ticks and have made no secret of working with them. What I collect there, we'll see after I've eaten it. The autopsy will show!
It is this circumstance that makes our activities largely meaningless :o) In all seriousness, there are simply no methods of dealing with such rows. Objectively, you can't even calculate the average, let alone the moments and other useful stuff (for example, there is no concept of time lag for such series).
Jesus, it's a good thing I didn't know that before. Otherwise I definitely would not have been able to work.
And why did we switch to "you", but okay. Actually, historically, fractal analysis was created for something else. Yuri, there aren't many methods that can give an objective assessment of Chaos. One of them is to calculate correlation integral for each nesting and total Chaos estimation (element of fractal analysis). It's what I used, with some assumptions. The dimensionality of the system for ticks is huge, there's complete Chaos with a capital letter, there's nothing to do there.
You switched to us, so we switched to you. :-)
Come on Sergei, you can't even joke about it.
I have a pretty good idea of what you're talking about. These things are too high for me. I built my own methods myself, there's nothing complicated about them. I try not to touch Chaos with my hands, otherwise it will drag me down irreversibly. I try to evaluate Order. I have not encountered the problem of dimensions. That is, perhaps, all.
God, it's a good thing I didn't know that before. I wouldn't have been able to work.
to marketeer
I don't really know what you mean anymore? Before you were talking about K-entropy, which is used in the formula. It has to do with the dimensionality of the system, that's what I wrote about. I.e. I'm getting a bit lost in the main idea. :о)
I wrote about choosing different levels of entropy as a criterion, depending on time, not about the cyclicality of entropy itself. But perhaps you're right, perhaps there is a correlation with volatility (I just need to clarify what it is and it's not that simple :o) It's not that simple, I wrote earlier that I don't work directly with the price but I use a transformed series that has the following properties
It's a whole study, you can't just answer that.
I don't get it...
Yeah, it's getting a bit confusing. ;-) I then also don't understand what exactly was said about it having "nothing to do with dimensionality"(>>), if one now agrees that entropy is related to dimensionality.
Time-dependent entropy levels are not cyclic? It doesn't matter what you call it, though, imho. The point is that if the model is adequate, then during the periods of high volatility it should give the variants of realization, for which the average entropy value is higher, than in the periods of calm. Therefore a correction for levels should already be contained in the prediction. And the upward and downward deviations are just a dance around some average value, the most probable one.
On the subject of "didn't get it". There was this conversation: I gave a formula, you wrote, "This formula will give the lowest possible forecast horizon...". I pointed out that this is incorrect. The formula gives an estimate of the maximum horizon. What exactly is unclear?
Yeah, it's all getting a bit confusing somehow. ;-) I then also don't understand what exactly was said about it having "nothing to do with dimensionality"(>>), if one now agrees that entropy is related to dimensionality.
I was referring to your "i.e. a measure of the scattering of a random variable over space". I don't know what that measure is, where it is scattered and what it has to do with entropy.
Isn't time-dependent entropy levels cyclic? It doesn't matter what you call it, though, imho. The point is that if the model is adequate, then during the periods of high volatility it should give the variants of realization, for which the average entropy value is higher, than in the periods of calm. Therefore correction for levels should already be contained in the prediction. And the upward and downward deviations are just dances around some average value, the most probable one.
The levels for the criterion would in some sense be cyclic (assuming those levels are triggered correctly as process identification). But I haven't written anything about the cyclicality of all entropy (or the entropy field), that has to be looked at. Again, you may be right.
On the subject of "didn't get it". There was this conversation: I gave a formula, you wrote, "This formula will give the lowest possible prediction horizon...". I pointed out that this is incorrect. The formula gives an estimate of the maximum horizon. What exactly is unclear?
OK, I was writing in the context of the last conversation about the value itself (how far you can predict)
Today the picture for FDAXZ9 is as follows:
Sell at market opening, take at 5616, stop at 5673 area.