Testing real-time forecasting systems - page 40

 
marketeer >> :
Great that time, but why the upside down density? ;-)

Well it happens that way, a little bit wrong. It's a concept! :о)

 
Then there is just a question about the concept ;-) - It somehow escapes me, why do we need to get the wave termination zone and centre at all, if we end up looking at distribution densities anyway? The only assumption is that in cases where the estimate goes beyond the boundaries of the zone, then the prediction is considered unreliable at all and we go into hibernation mode.
 
marketeer >> :
Then there's just a question about the concept ;-) - It somehow escapes me, why do we need to get the wave termination zone and centre at all, if we end up looking at distribution densities anyway? The only assumption is that in cases where the estimate goes beyond the zone boundaries, then the prediction is considered unreliable at all and we go into hibernation mode.

I don't really understand the question, but I suspect you haven't looked at the statistics for the completion of the ZZ waves. Roughly speaking, if you take all the statistics - a wave can complete anywhere, the "range of existence" is quite large. I think I wrote that we are talking about refining the zone, which is predetermined by some prediction system. I.e. quite simply - finding the most probable value in the found zone. It's very simple.

 
It may well be that I missed something (if by "watching" you mean studying your work - there's too much interesting stuff in it ;-) plus the discussions get quite extensive, I haven't seen such statistics in the main material on https://forum.mql4.com/ru/17052). So the densities are counted by projections within each zone. I wanted to clarify this, as in principle you could count densities on, say, the actual history of the same depth as used for the forecast, i.e. on a broader statistic compared to a single zone.
 
marketeer >>:
Вполне может быть, что что-то пропустил (если под "смотрением" понимается изучение ваших работ - слишком много в них всего интересного ;-) плюс обсуждения получаются довольно объемные, в осн. материале по https://forum.mql4.com/ru/17052 не видел такой статистики). Значит плотности считаются по прогнозам внутри каждой зоны. Я хотел это уточнить, т.к. в принципе можно было бы считать плотности, скажем, на фактической истории той же глубины, что использовалась для прогноза, т.е. на более широкой статистике по сравнению с отдельно взятой зоной.

It should not be there. This is an "asset management" topic - about applying linear programming as the basis for lot and risk optimisation to strategies which essentially predict ZZ segments. And the consideration of zones and densities is a very brief introduction, just to get you up to speed. :о)

 

Continuing real-time testing. Forecast for next Monday-Thursday (adjustments possible):


Quote: ........................................................................EURUSD

period:.............................. ..............................................M15 (on the chart respectively one countdown - 15 minutes).

forecast horizon: ...............................................400 counts (about 4 days)


System testing. Not for trading

Black crosses - probable realization of the process (do not pay attention)

Day - about 100 samples

System testing. Not for trading

Do you have something similar?

 
grasn писал(а) >>

Nice hedgehog!

And I love astrology, but I'm experimenting with non-traditional astrologer tools. For example, I was in a cafe, drinking coffee. Decided to use the hard sediment in a cup for its intended purpose, a very malleable material for creativity :o). As a result - a dirty table and very doubtful forecast on 15 minutes of EURUSD coffee grounds for the whole Monday, which makes the forecast absolutely useless, because it's sure to be wrong.

This is what the initial state of the system probability field looks like:

x - time samples

y - price levels

z - probability

I apologize at once that there are integers by levels. These are peculiarities of model calculation and peculiarities of MathCAD for these types of charts. At least I haven't figured out how to fix it yet (to feed normal numbers for visualization). Nearby for clarity I have placed correspondence of levels to their numbers:

The most probable movement is sideways, there is a probability that it will go down and a very small probability that after the 40th count (each count is 15 minutes) there will be a strong move up.

Next, the same, but only a contour representation (processed, the main thing left out):

A few statistically possible process realisations need not be looked at at all (it's too early to look at them closely yet), except to estimate the main statistical parameters:

There might be a strong move down in the 7-12-sample zone to the 1.3232 level. But maybe not, statistics is an exact science in the aspect of probability, it may or may not.

PS:

I hope nobody will trade at those levels, the overall assessment of possibility is very modest (for various reasons).

What does the probability give? Explain, maybe I am missing something.

For example, we have higher probability of growth. We open a buy position. The chart goes down. I thus add to our probability that it is more likely to go down. The next time we have a down move and open a sell the price goes upwards thus increasing the probability of an up move according to the history and statistics. And so on. After all, the market does not have the laws of probability, it is the algorithm of mass behaviour. And it has its own mind, but not the probabilities. IMHO. I.e. it would be closer to the analysis of all possible behaviours that influence the price, but the probability is somewhere in the tail.

 

Hi, I have a similar picture.

it's not pro, it's NNTP v1.3 (previous version)

 
grasn >> :

Well, it happens, it was a bit of a mistake. It's a concept! :о)

It's the phrase of the day, though it's a bit late for me to notice it...

 
Helex >> :

What is the probability? Explain, maybe I am missing something.

For example, we have a higher probability of growth. We open a buy. The chart goes down. I thus add to our probability that it is more likely to go down. The next time we have a down move and open a sell the price goes upwards thus increasing the probability of an up move according to the history and statistics. And so on. After all, the market does not have the laws of probability, it is the algorithm of mass behaviour. And it has its own mind, but not the probability. IMHO. That is, it would be closer to the analysis of all possible behaviors that affect the price, but the probability is somewhere in the tail end.

Sorry for my technical illiteracy, but I didn't understand much of what you said. I will only point out that there is always a probability, no matter what you construct or assume. That's the way the world works around you and me, it's woven of probabilities.

And what does probability give? Maybe I do not understand something.

That's not surprising, I haven't told you anywhere about the model I use.

And it already has a mind of its own, but in no way probability. IMHO. I.e. it would be closer to the analysis of all possible behaviours that influence the price, but the probability is somewhere in the tail.

If it clears things up for you, I get a statistically possible realization of a process (trajectory) and I write about it all the time. Such a trajectory in particular is shown in the forecast above. As to the estimation of possible behaviours, I do it to some extent and probability plays the main role, the model is conceptually based on the theory of self-organising stochastic systems (though in a slightly simplified form).


As for the mind, though, it's a philosophy. Sometimes you remember old mistakes and think, "f...., where were my brains?" :o)