How to minimise index correlation

 

The occasion was a post from here

https://www.mql5.com/ru/forum/114579/page19#576343 https://www.mql5.com/ru/forum/111317/page3

Let's assume that indexes really exist, but they must exist if we have currencies :)
And we can calculate the missing dollar index. Isn't it great?
But what is the dollar index, we can see only by calculating the before and after correlations. We calculate correlations in, say, 10 samples by moving window between all instruments and then display the average correlation before and after the conversion (to avoid the influence of negative correlation we will sum the modules when calculating the average). Suppose the transformation has decreased the symbol correlation (I say "suppose", because I have long ago removed all calculations, but anyone can repeat them). And if the correlation has decreased, then the dollar index is nothing more than a common basis. But since the correlation is not disappeared, we can continue the calculation even more general basis (introducing in the calculation have injections including the dollar index), and can continue this way for a long time, but the moment comes when the next index basis does not reduce the correlation, and increases it. That is, we have reached the limit. Thus, we have a huge amount of components, including almost non-correlated symbols that we habitually call currencies. The question arises: what should we do with them now? You cannot trade on these figures, though of course the conversion is reversible and you can always recalculate everything. Indexes are like a drunkard on a ship, it would be logical to extrapolate for some counts (because the result of any smoothing is lag, and indexes are useless figures without it) and shift a number of counts back to get the market status. But here's the problem - there is still no method that precisely extrapolates market data. Bam, faint. The reason is that the market is not stationary, and all methods of extrapolation require the constancy of the obtained transformation coefficients. Thus, using the known extrapolation methods we will obtain (albeit similar) forecasts with huge errors. Calculating backward from all these bases and indexes we'll get error accumulation and total nonsense forecast. The reason why it is needed when the extrapolation of currency pairs themselves can be done with fewer errors.
If someone says "I do not need extrapolation, I trade on indices", but it is impossible to trade on noise, and if you smooth out the noise, you get a delay, so what do you trade on? In short no matter how you look at it indexes don't give you an advantage over random entry. Amen.

----------------------------------------------

The author says that anyone can repeat the calculations.

I did not see how. but wanted to compare .

1 - so how to achieve the above.

 

You don't need to reduce anything to anything. It is a crooked method. Don't try to replicate it - don't learn a bad thing.

Go to Code Base, search for "currency index". You can find acceptable methods.

 
Freud:

.. So what the fuck is the point ...

That is the main question in the construction of NAHRINA indexes. And then, we can already talk about what to build, how to build it and whether it is worth building anything at all.

 
BoraBo:

This is the main question in the construction of NAHRINA indices. And then, we can talk about what to build, how to build it and whether it is worth building anything at all.


in the end only two things are important. the method of smoothing without lagging. and the indices.

To quote:

To quote: "First you need to find a lag-free smoothing method, and then solve the problem of decomposing quotes into indices. Suppose there is such a method.
Then the logical conclusion would be: if we have the method of non-delayed smoothing, then why do we need indexes?

i.e. if you have such a method, then indexes are essentially needed to select the best instruments.

So, there are instruments with a fan of smoothed, non-delayed lines from these instruments.

2- But if we have a method of smoothing without lagging, then why do we need indexes as well (as the author says).

You can smooth the already calculated indices in the same way. But will these 2 points be identical?

So the interesting point is that if such a smoothing method exists, it does not mean that this method is primary, and only then it may be applied to multicurrency.

 
Mislaid:

You don't need to reduce anything to anything. It is a crooked method. Don't try to replicate it - don't learn a bad thing.

Go to Code Base, search for "currency index". You can find acceptable methods.


Why the curve? The point is not to calculate the index, but to ask how the currency indices can be orthogonal to each successive reference. I don't remember it being in the Code Base.
 
Freud:

why a curve? the point is not in the index calculation, but in the question of how to whistle the currency indices to orthogonality (zero correlation) in each subsequent countdown. i don't recall this happening in the database. there are the usual stationary geometric average calculations.

For example, here's a central market)) Food products are traded - the rates are in rubles per kilo. There is a product beef and a product pork. Why should their price be independent? The source of dependency could be inflation, taxes, substitutability, etc. Should we invent a synthetic meat instrument (like 300g of pork+400g of beef+300g of lamb) that is independent? And independent of what? Another synthetic like fruit? Their prices would still end up being dependent. Even if it is possible to get synthetic food baskets that are poorly correlated in the isoria, why do we need them and where is the guarantee that they are not a fit with history?
 
Avals:

For example, take the central market)) There is a food product traded - the exchange rates are in roubles per kilo. There is a product called beef and a product called pork. Why should their price be independent? The source of dependency could be inflation, taxes, substitutability, etc. Is it necessary to invent a synthetic meat instrument (like 300g of pork+400g of beef+300g of lamb) that is independent? And independent of what? Another synthetic like fruit? Their prices would still end up being dependent. Even if the result can be synthetic food baskets that are poorly correlated in isoria, why do we need them and where is the guarantee that it is not a fit with history?


I'm not talking about complete absence of correlation. I'm talking about finding the moment when correlation decreased, decreased, reached a certain minimum, and began to increase, both for positive and negative correlation. that's where the minimum is interesting. only there are not 2 series.

We need to find a point (I suspect this is the geometric mean position) relative to which currency indices will have a minimum correlation between each other.

 
Freud:

I am not talking about the complete absence of correlation. i am talking about finding the moment when the correlation decreased, decreased, reached a certain minimum, and began to increase, both for positive and negative correlation. so this is the point where the minimum is interesting.

Well, each currency has a certain weight in the index, so that the sum of weights is 1. Then try to find the weights with minimal correlation. To accelerate, you can first search for weights in large increments, and once you have found the values at which the minimum correlation - reduce the step, selecting more accurately
 
Freud:


In the end only two things are important. the lag-free smoothing method and the indices.

to quote:

"First we need to find a lag-free smoothing method, and then we will solve the problem of decomposing quotes into indices. Suppose there is such a method.
Then the logical conclusion is: if we have the method of non-delayed smoothing, then why do we need indexes?

If there is such a method, then indexes are essentially needed to select better tools.

So, there are instruments with a fan of smoothed, non-delayed lines from these instruments.

2- But if we have a smoothing method without lagging, then why do we need indexes as well (as the author says).

You can smooth the already calculated indices in the same way. But will these 2 points be identical?

So the interesting point is that if such a smoothing method exists, it does not mean that this method is primary and only then it may be applied to multicurrency analysis.

We must start rehearsing the question: why do we need it, what do we want to see in the index? And then, about methods of interpreting index readings. And if you have methods that already work fine, maybe you shouldn't make such a big fuss.

 
Freud:



So the interesting point is that if a method of such smoothing exists, it does not mean that this method is primary, and only then can it be applied to multicurrency analysis.

If you have a lag-free smoothing method, it means that you know the future price values, it cannot be otherwise. This means that you already have the grail in your hands and you don't need any indices or indicators.
 
in general, smoothing cannot lag - it reflects what is in the formula. The method or parameters of smoothing may not correspond to the traded market inefficiency. That is, we must rely on the process to which we want to correspond with the smoothing