Machine learning in trading: theory, models, practice and algo-trading - page 3470

 
Renat Akhtyamov #:

ok, don't forget to tell me ;)

I am of the same opinion that it is very useful to know the background of maths and not to use ready-made bits and pieces

I've been following the attacks on you for a long time, but I'm on your side.

bravo!

Thanks, it's nice to hear that there are those who are close to my approach.

 
mytarmailS #:
And if I reverse the trade?

If I reverse it, I will get a 90% classification error, and theoretically, instead of a profit factor of up to 1.5, I will have a corresponding loss.

 
СанСаныч Фоменко #:

If I reverse it, I get a classification error of 90%, and theoretically, instead of a profit factor of up to 1.5, I get a corresponding loss

but practically it could turn out to be a profit

 
Maxim Dmitrievsky #:

Well, it should be.

There's a lot of them. This is an interesting article, by the way.

as I don't understand what it's for, never done anything with it.

The article is interesting, but the usefulness of all this is not measured.

It is better to search for one channel and build ZZ on it.

The question of calculating the probable range of trend reversal points by ZZ has been interesting for a long time.

Standard ZZ is actually built by Donchian channel, this knowledge can be used in TS, including for entry with minimum stops...

 
Aleksey Vyazmikin #:

The article is interesting, but the benefit of it all is not measured.

It is better to look for one channel and use it to build ZZ.

The question of calculating the probable range of trend reversal points by ZZ has been interesting for a long time, the screen shows my thoughts on it.

The standard ZZ is actually built on the Donchian channel, this knowledge can be used in TS, including for entry with minimum stops.....

I don't get it... maybe there is some infographics of what it gives?) in terms of statistics, if possible.

It just shows the high/low for a certain period.
 
Maxim Dmitrievsky #:
I don't get it at all... maybe there is some infographic of what it gives?) in terms of statistics, if possible.

It just shows the highs/lows for a certain period.

ZZ needs to be worked with to see what it can provide. For incentive to think about using it, it is enough for me that traders often use it in their TS.

I have an idea to try to generate new data according to the ZZ structure, which corresponds to the opinion that price has a goal, but no clear path - the antipode of the samurai. Accordingly, randomise bars from point to point of ZZ in order to assess the quality of the model.

 
Aleksey Vyazmikin #:

ZZ needs to be worked with to understand what it can give. The fact that traders often use it in their TS is enough to motivate me to think about using it.

I have an idea to try to generate new data according to the ZZ structure, which corresponds to the opinion that price has a goal, but no clear path - the antipode of a samurai. Accordingly, randomise bars from point to point of ZZ in order to assess the quality of the model.

ZZ standard - some dummy.

It is better to write rules of construction yourself, which will not be redrawn.

For example:

1. By fractals

2. By swings (trajectories of price movement according to the colour of candles)

3. By swing zigzag (3 swing zigzags in one direction, the middle one is the highest - as a fractal)

4. According to the minimum price movement in one direction, until the correction does not exceed N points.


None of these possible variants is not redrawn and does not miss liquidity withdrawals (surges), in return you get an adequate and transparent zigzag.

 

Surprisingly, I have not been able to find a correlation between the results of the models on the test sample (on which the training is stopped) and the independent sample (which did not participate in the training).

Usually their relationship looks like this

X is the test sample andY is theindependent sample.

And this is how it turned out today, using the new quantisation method.

Yes, it is still not very good, but even such a change is pleasing as a positive dynamics towards building stable models.

 
Ivan Butko #:

ZZ standard - some dummy.

It is better to write rules of construction yourself, which will not be redrawn.

For example:

1. By fractals

2. By swings (trajectories of price movement according to the colour of candles)

3. By zigzag of swings (3 swings, the middle one - the highest - as a fractal)

4. According to the minimum price movement in one direction until the correction does not exceed N points.


None of these possible variants is not redrawn and does not miss liquidity withdrawals (surges), in return you get an adequate and transparent zigzag.

The standard ZZ is enough to build on the Donchian channel that it does not redraw. I have a universal ZZ, which takes as input buffers of any indicators, on which it builds ZZ.

Maybe there is something better, I have not yet developed a definitive methodology for selecting ZZ. One of the criteria is to evaluate the effectiveness of the forecast about the beginning of the trend segment at the moment when the ZZ vector changes.

 
Aleksey Vyazmikin #:

ZZ needs to be worked with to understand what it can give. The fact that traders often use it in their TS is enough to motivate me to think about using it.

I have an idea to try to generate new data according to the ZZ structure, which corresponds to the opinion that price has a goal, but no clear path - the antipode of a samurai. Accordingly, randomise bars from point to point of ZZ in order to assess the quality of the model.

Sounds delusional