Machine learning in trading: theory, models, practice and algo-trading - page 1453

 
govich:

The faces seem to be the same, but the questions and answers... the same... It's like Groundhog Day or something)))

ZZ is 80% again...

An unsophisticated person would suspect some kind of conspiracy or a problem with his or her tower.

It seems that this is the 3rd or even 4th "wave" about the same thing, I can not say exactly because more than 2 / 3 posts have not read.

WHAT'S GOING ON, GUYS?

So it just says that the opinions of regulars have already formed and everyone goes in their own direction, and each direction is correct. Except ZZ which also can be correct if it is used without mistakes, but I am sure, those who use it do something wrong. IMHO of course. And when you start to use it correctly, then the result is suddenly wrong and immediately so grounded and the world is kind of in the wrong color, because I do not see a special advantage of using ZZ. Neither on the output, nor on the input.
And about the same, I do not agree and above I wrote that I have done since then, for the last 100 posts is the only ideas voiced and that can be taken on board. Right away, without transformation. Now give me some examples of ideas voiced in this hundred that you could try other than mine?
So who's helpful here? That's just it .... In the meantime, what I have voiced is fundamentally grounded. In development, I invented a new method of optimization, or rather its organization.....
 
Maxim Dmitrievsky:

If you're referring to English, I don't know what you nerds are doing here, but scientists are still studying the Brownian Fractional Motion to model volatility. There are no more accurate methods of describing market movements in the world yet. That is, starting with Black and Scholes and on to newer research.

https://tpq.io/p/rough_volatility_with_python.html

https://www.quantstart.com/articles/derivatives-pricing-ii-volatility-is-rough#ref-gatheral

So far all I see from you is a discussion of candle color predictions, zigzags and other kindergarten nonsense

There's a yuima package for R that has all this stuff - fractional Brownian, Levy flights, etc etc. There is a book about it, which may be useful at least because of the bibliography.

The YUIMA Project
The YUIMA Project
  • yuimaproject.com
The YUIMA Software performs various central statistical analyses such as quasi maximum likelihood estimation, adaptive Bayes estimation, structural change point analysis, hypotheses testing, asynchronous covariance estimation, lead-lag estimation, LASSO model selection, and so...
 
govich:

haha

I mean it'snot fantastic, like for ZZ, I don't argue, it's easy to get 95%, but it's useless, I mean it's fantastic 65% quality to predict purely future price changes, without past ones, on which ASR depends directly.

But if the price is obviously over 55% then I guess I screwed up, because I cannot predict much more than 50%, but I have ZZ, so the price is equally "cool", what does it mean? That it is possible to trade on SB?

1. If it's easy, give a concrete example with numbers, if you know how.

2. No need to advise ("take", "look") do it yourself and prove by concrete examples your statement. And the reference to the "big brothers" ... You could have written more simply: "A man told me so".

There are too many clever chatterboxes.

 
mytarmailS:

Hello Vladimir!

How are you doing with the script that I sent you, have you tried to experiment with it? Maybe you've developed the idea and this approach to regression

In private.

 
Andrey Dik:

Try 2 layers and reduce the number of neurons in the layers, down to 1 in each layer.

before the white vertical line - sample, after - oos

The more neurons - the more probability of fitting (more degrees of freedom), try to decrease the number of neurons as long as the neuron can produce at least somewhat sane results.

That is, the clearer the information in inputs and the rougher the mesh, the better.

Man, you're right.
 
Ivan Butko:
Man, are you right.


The results have improved?
 
Aleksey Nikolayev:

There is a yuima package for R that has all this stuff - fractional Brownian, Levy flights, etc. etc. There is a book about it, which may be useful at least because of the bibliography.

thanks, there are some unheard of models at the end, I'll read it

 
Maxim Dmitrievsky:

Thanks, there are some never-before-heard models at the end, I'll read

There are a lot of things there. For example - compound Poisson processes, which Alexander from TP branch invents and never invents)

 
Andrey Dik:


have the results improved?

The scatter (sawtooth amplitude) of the balance has slightly increased, the frequency of transactions has decreased, but on forward it repeats its stability for quite a long time. I've tried 20, 50 and 1000 neurons in 2 layers - immediately goes to the bottom, or some kind of chaos, although the training period is even line up. I also tried 30 layers of 10 neurons - same thing. I put 3 neurons in 2 layers - stable)))).

Put it on the real, I'll check it out.

 
Andrey Dik:

Try 2 layers and reduce the number of neurons in the layers, down to 1 in each layer.

before the white vertical line - sample, after - oos

The more neurons - the more probability of fitting (more degrees of freedom), try to decrease the number of neurons as long as the neuron can produce at least somewhat sane results.

That is, the clearer the information in the inputs and the rougher the mesh, the better.

I'm tired of drooling, it's too beautiful.

This is in time for what period? What is the secret technology?