From theory to practice - page 802

 
Evgeniy Chumakov:


this is essentially 90% of the grail.

That's the way it is and nothing else.

 
Alexander_K:

Don't get so upset, Che... You've made revolutions, why should you be sad?!

I'll tell you this just to be clear. This is a forum for programmers, remember that. High-end coders, but that's all...

There are almost no mathematicians, physicists, etc. There are no people capable of voicing an idea, testing, reporting. Professional work is a gimmick here. Here there is the usual rabble of bidders and slackers.

More or less work in the MoD branch, but things are slow there, and it's a pity...

As for me, I periodically show reports. These, for example:

This is the result for one day.

But as you can see, it's not very good. Tired of saying the same thing - without the persistence/anti parameter there is nothing to do in the market. And no one will even budge to look in this direction! Even under my promise to bestow the grail on such people...

I have to do everything myself. I'm experiencing an excess - we'll see.

On the subject of persistence and anti, I have a ready-made implementation.
What will it do?
I myself am a professional coder in 4 languages, but a lot depends on coders too. For example reliability of executable functions or classes.
Those who know .NET may use it and extract different information as I have in one indicator.
And so on...
Personally, my problem now comes down to using neural network to "analyze" with statistical coefficients the parts of the market where my robot doesn't trade well. I don't even know how to do it any other way. Otherwise I do not even know how to do it. The result of trading is similar to a grid with martin but with one nuance - only one order is opened and closed and due to minuscule risks the robot has to go through fully antipersistent (Rmax-Rmin)*3 (otherwise 3*the full range where the price was for all time) points to fail.
 
Martin Cheguevara:
As for persistence and anti, I have a ready-made implementation.
What will she give?

Everything. This so-called "indicator of discord" is the Grail. Nothing else is needed.

 
Alexander_K:

That's it. This so called "breakdown indicator" is the Grail. Nothing else is needed.

Indicator of what?) Can you explain your idea in more detail. Why it has to be the Grail...
 

Once again, I will repeat what this indicator should look like.

I take the kurtosis of the distribution of tick increments and I see that at an excess of >20 a trend movement almost always begins.

Who does not have such or similar indicator in his pocket, he will soon slide to the daily consumption of vodka near the station toilet. Without it, no averages, semi-averages, channels or other nonsense will help.

 
Alexander_K:

Once again, I will repeat what this indicator should look like.



How do you build the price channel now? Are the formulas too complicated or not?

 
Evgeniy Chumakov:


How do you build the price channel? Are the formulas too complicated or not?

I use two channels:

1. as Koldun recommended - sum of increments in sliding window + variance with quantile.

2. Dispersion without quantile relative to moving expectation (Variance Gamma Process).

Both variants show profit with the kurtosis, and both are at 0 without it.

Conclusion: the Kvariance Indicator is much more important than the strategy itself.

 
Alexander_K:

Once again, I will repeat what this indicator should look like.

I take the kurtosis of the distribution of tick increments and I see that at an excess of >20 a trend movement almost always begins.

Who does not have such or similar indicator in his pocket, he will soon slide to the daily consumption of vodka near the station toilet. Without it all averages, half-means, channels, etc. will not help.

You are right, but there is one little nuance, firstly, the sharp increase of kurtosis sharply increases the oscillations, it means that the spreading bell "gets bigger" in both directions and it means that the ability to make profit from this movement decreases sharply, as it is not stable (stable trend movements are very, very rare).
Secondly, the price can "jump" to a standstill. In this case it will simply take a loss, and the bell will not change.
Of course, I don't know... but even on ticks the minute and tick spikes are almost identical. I was looking at Dycoskopy. It can be 1 tick or 3... even 10...
In third, with this approach until you wait for a steady trend (because you cannot win with a sharp jump in it, you cannot define it before it is possible to get lost), you will lose your deposit.
The only option is to move by sigma in search of the optimal entry at the surge and find the most optimal result.
In the fourth, are there figures that show that you can really gain profit during the last five years with an ekcess >20? For example a test with an order on a signal where TP=SL?
 
Martin Cheguevara:
As for persistence and anti, I have a ready implementation.
What will it do?


You seriously don't know how to apply it? It' s the Grail, if you really know on-line. You may take any trend-following strategy, even on the wave/parabolic and you may trade it only during trends. They may be of different types, i.e. flat, oscillators, or both, so as not to sit idle.

 
Martin Cheguevara:
You're right, but there's one little nuance. First, if the kurtosis increases sharply, the vibrations go up, so the bell of distribution gets "fat" in both directions.

You got that right. I respect that. It's obvious you've been working on this subject.

You don't need it both ways. You only need one! You have to look at the asymmetry for that. You're not?