Machine learning in trading: theory, models, practice and algo-trading - page 1013

 
Maxim Dmitrievsky:

I think you need to reduce the number of trades, it seems, on every bar.

Another algorithm with fewer trades:

depth three

[[ 4838  4123]
 [ 9926 11966]]
трайн наоборот
[[ 4975  3986]
 [ 9925 11967]]
трайн 
[[3867 2501]
 [8014 6702]]
тест

15

[[ 3652  5309]
 [ 2298 19594]]
обратный трайн
[[ 8607   354]
 [  587 21305]]
трайн
[[ 1370  4998]
 [ 2577 12139]]
тест

At 15 everything crashed)

Also not a very good model.

All together:

3

15


 
Goodbye, everyone. If you find any good predictors, let me know.)
 
forexman77:

Another algorithm, where there are fewer trades:

depth of three

15

at 15 everything is broken)

Also not a very good model.

All together:

3

15


I would not have enough to start taking out more trades, maybe I should decrease the number of trades, because there is too much, play with the trading logic.

 
forexman77:
Goodbye, everyone. If you find any good predictors, let me know.)

I will not tire of quoting excerpts from Kolmogorov:

In other words, the following are considered:

1. Returns

2. the ACF for returns.

If the ACF satisfies the following condition:

then such a discrete returnee series is predictable.

That is all.

There are no other predictors, have not been and will not be.

 
Alexander_K2:

I will not tire of quoting excerpts from Kolmogorov:

In other words, the following are considered:

1. Returns

2. the ACF for returns.

If the ACF satisfies the following condition:

then such a discrete returnee series is predictable.

That is all.

There are no other predictors, have not been and will not be.

You like to throw in all sorts of formulas like that. How to do it in practice, in words you can explain. I'm not familiar with formulas very well.

If you can explain them, I don't think it would be too hard to put them into code. In its turn, I will use the ACF indicator by price series.

 
forexman77:

You like to throw all sorts of formulas. How to do it in practice, you can explain in words. I'm not very familiar with formulas.

If you explain it, I think it won't be too difficult to implement in code. Moreover, I have done the ACF indicator by price series.

Strictly speaking, in a sliding sample of returnees I should calculate the ACF estimation for this discrete series. If it is periodic, the next returnee is predicted to be 100% according to Kolmogorov. But I do not know the criterion for assessing the periodicity of ACF. I cannot just look at it by eye...

 
Ivan Negreshniy:

And where is such a group, if not a secret, and whether it is possible to look there?

There only the divine guru and a couple of his padawans add people, throw me in a private message your skype and data about yourself, I will ask, but I do not promise anything because I am not an authority there, only spirit disembodied, dirt on slippers. It is a gray cardinals, puppet and company, who will be noticed for near-market activity, branded with shame for life, it is possible to wash away the shame only for tens of billions of greenbacks.

 
Olga Shelemey:

Strictly speaking, in a sliding sample of returnees you should calculate the ACF estimator for this discrete series. If it is periodic, then the next returnee is predicted 100% by Kolmogorov. But I do not know the criterion for assessing the periodicity of ACF. It is not "by eye" to look at it, really...

I know that in ARIMA if ACF goes down smoothly and crosses zero once, then the segment is trendy.

If it crosses zero several times, then it's a retracement.

How do you calculate the returnees? There in formula 2 multiply by the sum of some cosines with something, I don't understand?

 
forexman77:

I know that in ARIMA, if the ACF goes down smoothly and crosses zero once, then the section is trendy.

If it crosses zero several times, then it is a rerun.

How do you calculate the returnees? There in formula 2 multiply by the sum of some cosines with something, I don't understand?

Yes, returns are simple = current - previous price.

ACF is also simple = the sum of the products of the current and previous returnees in the sliding sample.

One should only be sure that at a given moment ACF is periodic; if not - do not trade.

So how can we be sure that ACF is periodic at the moment? Do I know?

 
Alexander_K2:

Yes, returns are simple = current - previous price.

ACF is also simple = the sum of products of the current and previous returnees in the moving sample.

One should only be sure that at a given moment ACF is periodic; if not - do not trade.

So how can we be sure that ACF is periodic at the moment? Do I know?

What is meant by periodicity?

And as far as I know ACF is not just a sum of products. There is a much more complicated algorithm.

ACF of the trend plot:

A flat area: