TSR - resuscitating trading systems - page 8

New comment
 
MetaDriver:

You could argue the second point, too. There is no linearity there. But I am more willing to go over the third point.

I agree 100%, I got carried away. There's no linearity there.
 
MetaDriver:

The councillor is rubbish, the idea works.

You wrote a summary of the idea almost immediately. It's not the idea that works, it's the method of fitting it that's not classic.
 
hrenfx:
So wrote almost immediately a generalisation of the idea. It's not the idea that works, it's the fitting method that's not classic.
Nope. The fit there is no more formulaic than that. A method of filtering out the "fitting component" is suggested. By mutual filtering of trade signals of two TS adjusted on different parts of the time series of quotes.
 
Perhaps I did not understand Reshetov. Then please explain to me in another clear language, what is interfiltration and how does it work?
 
hrenfx:
...... what is interfiltration and how does it work?
Due to the fact that the synthetic (of 2 TCs) system only makes transactions by mutual agreement of the two parties. If there is a contradiction, the transactions are cancelled.
int Supervisor() {

// Подгонка первой системы
   if (pass == 1) {
      return (perceptron1());
   }

// Подгонка второй системы
   if (pass == 2) {
      return (perceptron2());
   }

// Совместный танец двух систем :
   if ((pass == 3) || (! IsTesting())) {
      if (perceptron1() == perceptron2()) {
         return (perceptron1());
      }
   }
   return (0);
}
 

In short, the following is proposed:

  1. The "optimum" conditions for closing trades on two intervals at the same time are found. These parameters are fixed.
  2. TS (opening conditions) on the first interval are optimized. We obtained TS1.
  3. Optimized TS (opening conditions) in the second interval. We obtained TS2.

Then TC1 is crossed with TC2 through filtering: if signals at TC1 and TC2 coincide - the signal is passed, otherwise it is filtered.

Did you get it right?

 
hrenfx:

In short, the following is proposed:

  1. The "optimum" conditions for closing trades on two intervals at the same time are found. These parameters are fixed.
  2. TC (opening conditions) on the first interval is optimized. We obtained TS1.
  3. Optimized TS (opening conditions) in the second interval. We obtained TS2.

Then TC1 is crossed with TC2 through filtering: if signals at TC1 and TC2 coincide - the signal is passed, otherwise it is filtered.

Did you get it right?

Yes.

// That's it, Ivan, I'm pecking my nose right now. I'm going to bed. We'll continue tomorrow.

 
Figar0:
The advisor itself may not be worth anything, it is the idea it demonstrates that is worthwhile and it does a great job with that. What else is required?
The Expert Advisor is not worth anything at all, because if you look closely at R. Pardo's work, such systems should immediately be thrown away, because the TS is losing on OOS both times after two attempts
 
voltair:
Very interesting! If the problem of establishing patterns is solved, then... goodbye BP non-stationarity!? :) (I mean great!) But what allows you to assert uniqueness, i.e. the result of detection?

Why say goodbye to non-stationarity? Unsteadiness will never go anywhere. That's why the grail will never appear.

Figar0:

It's a pity that thread was so cluttered, there were scraps of useful ideas there and now it's hard to find them... But I remember your posts there, even though I didn't entirely agree with them.

And once again it is a pity that you cannot give a word or two about your method "to statedefinitely the existence of the regularities found by TC . Is it so secret and unambiguous? Maybe you can give us a hint about what it is?

Indeed, it would be very nice if the moderators cleaned up that thread. There is enough information in it to draw far-reaching conclusions.

The EA presented in the first post refers to TCs of the first type (as far as I understood from the code), while I consider and work with TCs of the second type exclusively.

 
Reshetov:
The Expert Advisor is not worth anything at all, because if you carefully read the work of R. Pardo, then such systems should be immediately thrown out to the trash, because the TS fails at OOS both times


if advisor does not cost anything, then any combination of loos systems will be loos. It's like filtering one random wander with another - you'll still get SB. Although you will get a third SB, which may well be directed upwards purely by chance.

If at least one of the crossed systems is robust, then it might make sense. It should be compared with a simple portfolio of these two systems - which is more profitable and/or under what conditions

12345678910111213
New comment
Reason: