Machine learning in trading: theory, models, practice and algo-trading - page 3566

New comment
 
What about the last TC? The one on the regression.
 
Maxim Dmitrievsky #:
What about the last one? The regression one.

It's not a TS, just potential entry points, so if you take all the signals in a row, it's a 100% loss.


If you know it's going to fall but don't know where to sell, you switch on the "sell only" mode and it trades .


I sort of divide the entry conditions into two phases.

1. Global buy/sell direction.
It is not rigid, poorly understandable and hard to programme.

2. Local precise entry.
It is easy to programme, but it has little forecast and is not very useful.
 
mytarmailS #:

This is not a TS, just potential entry points, if you take all the signals in a row, then 100% failure.


If you know that it will fall but don't know where to sell, you switch on the device in "sell only" mode and it trades .


I sort of divide the entry conditions into two phases.

1. Global buy/sell direction.
It is not rigid, poorly understandable and hard to programme.

2. Local precise input.
Easy to programme, but it has little forecast and is not very useful.

And if you combine signals from different TFs. Without backtests, it's hard to evaluate.

I got into markup of trades. How much the results on forward trades depend on the markup rather than on the signs.
 
Maxim Dmitrievsky #:

and if you combine signals from different TFs. In general, without backtests, of course, it is difficult to evaluate.

It can't be described by a usual backtest.

I think that this model describes some kind of arbitrageurs sliding, they see this outlier/anomaly/sliding and enter thus making a small impulse, and I enter with them for a while.

So different TFs, I don't think it will work...

 
Maxim Dmitrievsky #:

and if you combine signals from different TFs. Without backtests, of course, it's hard to evaluate.

I got into markup of trades. How much the results on forward trades depend on the markup and not on the signs.

Exactly!

And also on the point of separation into classes.

And there should be a lot of signs (100 or more) and you should be able to select from them (no more than 10) the most related to the marks.

 
mytarmailS #:

I created a prediction algorithm on the hourly and even the minutes.

It is actually possible to forecast 24 points a day ahead.


There you go, that's right.

a million options

You're not gonna believe this, but it's the most accurate prediction.

I mean, nobody knows where the price is going to go.
 
Renat Akhtyamov #:

There you go, that's right.

a million options

You're not gonna believe this, but it's the most accurate prediction.

I mean, nobody knows where the price is going to go.
Are you for real. or are you pretending to be?
 
mytarmailS #:
Are you for real. or are you pretending to be?

The multicoloured ones are right.

The rest is human heuristics.

No one knows who's going to think of what and where and when they're going to hit.

The price will react differently than you think it's gonna average out.

It's not?

Or do you think it's going to follow the trajectory that you've made fat.

Just come down to reality and don't ask stupid questions.

For example, I've traded 100 lots in real money.

the price will go down, no matter how you draw ;)))))).

Learn to earn with your head, don't torture your computer processor, it's useless.

I have a robot, of course, but it executes what my brain came up with, the whole algorithm, from start to finish.

 
Maxim Dmitrievsky #:
I'm getting into marking deals.

Any new ideas in this direction?

 
Aleksey Vyazmikin #:

Any new ideas in that direction?

A recombination of old ones. You won't know until you test it all.

There is still material for 5 more articles, probably, I'm too lazy to write.

1...3559356035613562356335643565356635673568356935703571357235733574
New comment