Not the Grail, just a regular one - Bablokos!!! - page 58

 
Mathemat:
And who says that pairs must necessarily be combined so that there is something quasi-stationary? What if it's better the other way round - as unsteady as possible?

A few times I've been involved in a discussion about paired trading - not once has there been any constructive discussion. There are a few misunderstandings for me, but always on top of that.

On non-stationarity. Stationarity of the residual means that if you entered on the variance, it will necessarily return to mo. In non-stationary, the variance is arbitrary and there is no such guarantee.

 
khorosh:
Aleksandr recommends the correlation of traded pairs between 35-75%.

There is no correlation for non-stationary series.

Just for the record. All correlation (regression) analysis is based on the assumption of stationarity of the series.

 
faa1947:

Read what you are replying to. It says 6,736 bars (year) by which the 236 bar window shifts.

Bloody hell! If you don't read the posts, then don't respond to them.

Show me where the word "year" was said in that post? How am I supposed to know what timeframe you are writing about.

 
Meat:

Show me where the word "year" was said in that post? How am I supposed to know what timeframe you are talking about?

And again, we are not interested in the result on a particular sample, but in the long term result on a moving sample. I.e. if it is 1 year, then we need a result for at least 5-6 years. Otherwise it all looks like pure fitting.

Yeah, it's hopeless.
 

OK, I apologise, I really didn't read your posts carefully. I get it now.

 
faa1947:

There is no correlation for non-stationary series.

Just for the record. All correlation (regression) analysis is based on the assumption of stationarity of the series.

With such a wide range of acceptable correlation values, maybe non-stationarity can be neglected? After all, this does not prevent Aleksander from making money.
 
khorosh:
With such a wide range of admissible correlation values maybe one can neglect non-stationarity? After all, it doesn't prevent Aleksander from making money.

For non-stationary series there is no correlation as a concept. That is why all forum threads discussing the correlation of two quotes are nothing more than a numbers game. Nothing sustainable can be built on this figure.

Quotors are not stationary and we proceed from that and never question it.

I write all the time about the stationarity of the sum of two non-stationary quotes. Granger substantiated this amazing phenomenon and got a Nobel for it. I wrote on this forum many times. There is even a branch.

 
faa1947:

For non-stationary series there is no correlation as a concept. Therefore, all forum threads discussing the correlation of two quotes is nothing more than flooding.

Quotes are non-stationary and that is what we assume and never question.

I, on the other hand, write all the time about the stationarity of the sum of two non-stationary quotes. Granger substantiated this amazing phenomenon and got a Nobel for it. I wrote on this forum many times. There is even a branch.

I don't want to reject your opinion ("the correlation of two quotes is nothing more than a flood") from the point of view of strict mathematics. But from the point of view of practical use of correlation we can see the following picture: let's take for example eurusd and usdchf quotes and measure the correlation between them using the script. We obtain the result close to -1 (inverse correlation is very high). Let's look at it visually and see if it is really true - almost a mirror image. We can also compare it with two other quotes where the correlation is very low. We visually look at these pairs and indeed see that there is no in-phase movement. These experiments confirm that correlation can be used for practical purposes to estimate the degree of in-phase movement of two symbols when choosing appropriate currencies for paired trading.

 
faa1947:

For non-stationary series, correlation is absent as a concept.

Where does it come from? What does it follow from?

I have heard about the requirement of normality of the distribution of the quantities studied for correlation, but the requirement of stationarity - where is it written and who requires it?

 
khorosh:

I don't want to deny your opinion ("the correlation of two quotes is nothing more than a flood") from the point of view of strict mathematics. But from the point of view of practical use of correlation the following picture can be observed: Take for example eurusd and usdchf quotes, measure the correlation between them using the script. We obtain the result close to -1 (inverse correlation is very high). Let's look at it visually and see if it is really true - almost a mirror image. We can also compare it with two other quotes where the correlation is very low. We visually look at these pairs and indeed see that there is no in-phase movement. These experiments confirm that correlation can be used for practical purposes to estimate the degree of in-phase movement of two symbols when choosing currencies for paired trading.


The basis of paired trading is cointegration, and we cannot use correlation. Co-integration can be estimated even visually - it's flatness. I.e. the tendency of the kotyr to return to the average for example. Right now eurusd and usdchf are cointegrated. It can be seen in the eurchf cross. But the flatness is in a very narrow range.

The basis of cointegration is the property that the greater the deviation, the more likely the return. The economic sense is that some participants have reasons to trade on the convergence. Trying to jump in front of them. Therefore, we need to understand the reasons why the instruments are now cointegrated, rather than fitting everything into a flat.