Correlation, allocation in a portfolio. Calculation methods - page 7

New comment
 
CHINGIZ MUSTAFAEV:

We are looking for the same signals at the same moment on different correlations, because of the different correlation we can take more assets to deversify/increase the profit.

We are approaching the same target by different ways))) If you take into portfolio instruments that do not depend on each other and behave differently depending on external factors, when they grow, you can hope with a reasonable probability that they all together will not decrease in price.

It makes sense. Estimating the correlation on a plot (any, complete, incomplete) by mathematical methods means nothing by itself. A true negative or false positive error cannot be detected by mathematical methods. But the goal is to hope/forecast that the portfolio will behave differently to external factors. And at some point it may not work.

I agree with Buffett, without an understanding of the process it is better to stay at home and not go to the market))))

 
Renat Akhtyamov:
I'm trying to imagine two segments under this condition. Will perpendicularly placed give a correlation equal to 0.0?

If so, this is roughly how a cross and a major move, for example, but not always.

Physically, the process looks like this (I wrote about it more than once)

- if the majors are trending, the cross is flat

- If the cross is trending, the major is flat.

Correspondingly, in these cases the correlation will be close to zero.

Apparently not, it searches instruments that are independent and different in fact. But it finds what it finds.)

 
Renat Akhtyamov:

It's not that it's big, it's that for each individual algorithm it will only be numbers that are peculiar to it. This means that applying any formula to a series of random numbers can only cloud the skull, nothing more.

And in general, it is difficult to imagine any random series.

In simple words - there is an algorithm of RNG, there is a logical sequence of operations for generating random numbers, which in fact will not be random and will necessarily obtain a logical distribution of generation.

For example, generate numbers from 1 to 10. For the RNG algorithm number 1 will fall out more often 5-rings, and for the algorithm number 2, 10-rings. And the repeatability of this distribution will be 100%.

And before you blurt out the theorem, I suggest checking it out in practice, making a million generations on each algorithm and make sure of what I have said.

Don't confuse PRNG and RNG. Only PRNGs are based on algorithms, while RNGs are hardware-based (e.g. quantum).

Try reading and understanding about the well-known "paradox of birthdays" or standard probability tests for PRNG and don't talk nonsense. It is almost impossible to see the difference between a PRNG and a good PRNG.

 

Portfolio construction is primarily an optimisation problem. It comes down to finding correlations when set in the form of Markowitz theory. There is an interesting transcendreamer's thread on the *** forum about portfolios.

And in my opinion, portfolios are better composed of trading systems rather than instruments themselves.

 
Aleksey Nikolayev:

There is no need to confuse PRNG and LNG. Only PRNGs are based on algorithms and RNGs are hardware based (e.g. quantum).

Try reading and understanding the well-known "paradox of birthdays" or standard probability tests for PRNG and don't talk nonsense. It's almost impossible to see the difference between a PRNG and a good PRNG.

Quantum can't be argued. sort of like low frequency sampling from uncontrolled high frequency sampling should work too, but also the paradoxes of starting a high frequency process))) Gaming machines have proved that you can't do that either)

Funny paradox, I read it in 5th or 7th grade in Quantum)))

 
Aleksey Nikolayev:

Portfolio construction is primarily an optimisation problem. It comes down to finding correlations when set in the form of Markowitz theory. There is an interesting transcendreamer's thread on the *** forum about portfolios.

And in my opinion, portfolios are better composed of trading systems and not of instruments.

This is what a view is all about

and theory is far from practice

The optimal forex portfolio is, oddly enough, composed of 13 pairs.

 
Aleksey Nikolayev:

Portfolio construction is primarily an optimisation problem. It comes down to finding correlations when set in the form of Markowitz theory. There is an interesting transcendreamer's thread on the *** forum about portfolios.

And in my opinion, portfolios are better composed of trading systems rather than instruments themselves.

This is a quantitative method of investing. My signal is strictly one, I can only take out at the expense of a large number of assets
 
Renat Akhtyamov:
I try to use two bars for this condition. Will the correlation be 0.0 if they are perpendicular?

If so, this is roughly how a cross and a major move, but not always.

Physically, the process looks like this (I wrote about it more than once)

- if the majors are trending, the cross is flat

- If the cross is trending, the major is flat.

Correspondingly, in these cases the correlation will be close to zero.

Yes, that's right. Different conjunctures, different events at the current moment in time.

Simply, when you make 1 quality signal, you cannot afford the correlation in your portfolio, well, no way. It is easier to put all the risks in 1 currency pair.

Again, I was asking about the method, how to find and automate it for a robot or indicator.
 
My observations have shown that you should use as limited a period as possible in order to catch rascorrelation

Approximately the period in which the signal operates.
 
CHINGIZ MUSTAFAEV:
You are right. Different conditions, different events in the current moment.

It's just that when you make 1 quality signal you can't afford correlation in your portfolio, no way. It is easier to put all risks in 1 currency.

Again, I was asking about a method to find and automate this for a robot or indicator.
Any indicator or calculation should consist of at least two lines, and the difference between them is the volume. But testing is necessary, not without it. Statistics allows for parallel buy and sell lines. But there is no guarantee that they will not move apart.
For one pair it is possible to try two wizards but they slow down badly.
There was also a system for correlation. It analyzes whether they converge or diverge. When they diverge - the upper line goes to buy, the lower one - to sell. But again, the correlation formula is terribly lagging behind.
123456789101112
New comment