Pair trading and multicurrency arbitrage. The showdown. - page 63

New comment
 
Dmytryi Nazarchuk #:

Bajan. No, it's a cliché.


all coefficients (correlations, cointegration, etc.) are purely speaking about the past.

"window function showed that in the past window win could be traded" :-)

PS/ and it's pretty weird to count them..skipping the long sex with data - remove known trends, common seasonality, individual outliers, bring in comparable values. Justify the method of "cleaning" and say in advance how it will affect the result.

 
It is necessary to search for cycles with the help of Görzel's algorithm. Or do some other amateur radio experiments. It will be trillions of money in the end, as it always happens with practical practitioners.
 
Andrey Dik #:
I tried chain on koins about five years ago on a few exchanges, even then it didn't work well.
No one is responsible for anything on koin, so it's pointless to demand speed.
Yes, for various reasons it will be difficult to realise. And if it is possible, the output will still be pennies. But in itself it is an interesting programming and mathematical problem.
 
mytarmailS #:
I'm sorry, but you're talking utter rubbish.
I don't agree with a single word and it's easy to prove it
IMHO, cointegration is of course the basis, but after its detection you need to study the spread to build the final TS. Co-integration itself does not give data for entry-exit levels, volume of trades, etc.
 

In crypto, levelling arbitrage (safe trade - buy A<-B and sell A->C->B at the same time, pocket the difference) does not work just like everywhere else. Everything is equalised right away and all the time.

But on binance, for example, a deal through the third leverage can be slightly more profitable than a direct one. Instead of A<-B take A<-BNB<-B . Nothing personal, just BNB backing/boosting the BNB process owner

 
Aleksey Nikolayev #:
IMHO, cointegration is of course the basis, but after its detection you need to study the spread to build the final TS. Co-integration itself does not give data for entry-exit levels, volume of trades, etc .
Of course it doesn't, it solves its task...
Namely, it answers the question - whether it is possible to trade one pair against another (in the past), while correlation does not answer this question, although there is a persistent illusion among many that it does.
 
Dmytryi Nazarchuk #:

put any currency instead of XXX.

There's no such thing. Not that I know of.

 
In general, the situation is like this, optimise for 10 years owl that got. On Monday I'll put it on demo. We'll see. The results are very interesting.
 
Aleksey Nikolayev #:
It is necessary to search for cycles with the help of Görzel's algorithm. Or do some other amateur radio experiments. There will be trillions of money in the end, as it always happens with practical practitioners.

But you don't need to look for anything. All cycles are known in advance. The global question is how to account for them :-)

Figuratively: 2 significant events with a difference of 5 hours between them occur every day. Cycle search algorithms will detect the cycle of 24 hours, 5 hours and 24-5=19 hours, and 5^a*19^b ; and multiples between the somnomials. Spectrum.

Cycle detection algorithms are needed only to confirm the significance of events (their cycles are traced and traces are confirmed by history). What is the point of applying them if everything is known in advance ?

 
Aleksey Nikolayev #:
or some other amateur radio stuff.
Or MOSH, for that matter.
1...565758596061626364656667686970...254
New comment