Pair trading and multicurrency arbitrage. The showdown. - page 197

New comment
 
Aleksey Nikolayev #:

Well, you can't do it without bruteforcing anyway. The problem, imho, is that the global extremum obtained during optimisation does not necessarily correspond to any real regularity, but rather it corresponds to some overfitting. And even if you apply a more complex optimisation scheme with metaparameters and crossvalidation, it will only add opportunities for misses (overfitting of metaparameters, for example).

Intuition suggests that the reason is the weakness of real-world patterns versus the seemingly random patterns sporadically occurring in any sample. Something like pyrite, the "fool's gold" that prevents you from looking for real gold.

There was a thought in the MO thread that MO!=optimisation and it should be taken to its logical conclusion.

I should also add that MO is about prediction, not about finding patterns. Optimisation optimises, MO is trained to predict. Then people want to look for patterns with MO, it's like a separate approach that includes both MO and statistics and causal inference.

And this approach has a priori disadvantages, from which there is no escape. Because according to the Evidence ladder such causal inference is on the bottom rung in our case. Accordingly, the reliability of the results is always in question.


 
Maxim Dmitrievsky #:

I should also add that MO is about prediction, not about finding patterns. Optimisation optimises, MO is trained to predict. Then people want to look for regularities with MO, it is like a separate approach, which includes both MO and statistics and causal inference.

And this approach has a priori disadvantages, from which there is no escape. Because according to the Evidence ladder such causal inference is on the bottom step in our case.

and the standard

lag.

so any kind of prediction is just a headache.

You have to cleanse yourself of this stuff.

 
Renat Akhtyamov #:

in what thread?

link or title

I'll read it.

Equity balance indicator. Section below
 
Renat Akhtyamov #:

standard

lag

so any kind of prediction is just a headache.

You need to cleanse yourself of that kind of thing, completely.

You're not making any sense.
 
Maxim Dmitrievsky #:
You're not making any sense.

your answer is justified by the fact that it's hard to accept useless experiments.

 
Renat Akhtyamov #:

your answer is justified by the fact that it's hard to accept useless experiments.

Bullshit.
 
Maxim Kuznetsov #:

Renat, what are you talking about?

Anyway, here it is

back to the thread I linked you to yesterday.

Here it is, the arbitration situation, which the author of the thread is silent about, on which he is allegedly making money:

this indica is made on 4rka, because there you only have to enter your formula, the rest is done.

codebase indication

meaning of this situation: BID > ASK !

I remembered it because you asked a question about the spread.

I personally don't want to bother with it, as it is technically difficult to realise it in real trading.

in a tester - you can do it there, as there will be no problems with speed and reliability of trade orders execution.

// though I'll try to find a fast demo....
 
mvf358 #:
Equity balance indicator. The section below

I didn't see anything grailish there.

especially from a man who can't even test his system and who has a faulty Caps Lock key.

Besides, a person with grail has enough money to order what he wants from a local freelancer.

That's the minimum proof of grail.

But there's no such thing.

 
Maxim Dmitrievsky #:
That's bullshit

I'm a practitioner, not a theorist.

so I know what I'm talking about.

 
Renat Akhtyamov #:

I'm a practitioner, not a theorist.

so I know what I'm talking about.

You're collecting rubbish
1...190191192193194195196197198199200201202203204...261
New comment