Correct calculation of currency indices. - page 3

 
Urain писал(а) >>

I've already written:
"Personally I need it for extrapolation as each pair has a mixture of 2 currency movements, it's hard to extrapolate correctly".

That's why you need an index calculation that, when checked, would give the values
of the currency pairs (at least an approximation)

That's the thing: roughly. I recommend that you do the following. Take the exchange rates of these currencies at three points. Let's say at 16:00 over the last three days. And take the value of the index at these points. And calculate the network, but not with the help of the computer, but with the help of the pencil. With a pencil. And you will see everything. I think the questions will disappear.

 

Check on 5-digit quotes or by the hawks.

If the error is due to the non-synchronisation of the quotes, it should reduce.

 
Mischek >> :

Making up your own indices can play a cruel trick . Much more important is what all real market participants see . As with non-standard ETFs. The good thing about the standard one-hour bar

the fact that it is the same for all. What is good about a S&P breakout on the USDX chart - it will affect the Straight pairs.

Maybe that's why 99.999% of them are losing. They see the same thing, and they shed the same amount of money fast. There is a clear correlation.

 
Zhunko >> :

Maybe that's why 99.999% of them are leaking. They see the same thing, and they shed just as fast. There's an unambiguous correlation.

Those who try to be smarter than the majority (in money) and write against the wind are the losers

 
Urain >> :

I have already written:
"Personally I need it for extrapolation as each pair has a mixed movement of 2 currencies, it's hard to extrapolate correctly".

That's why you need an index calculation that, when checked, would give the values
of the currency pairs (at least approximately)

To get simple components, you have to use spectral analysis. There you can get as many of them as you like.

But, if you apply spectral analysis to indices.... This is very cool! One has to know why and how to use it.

 
Zhunko >> :

To get simple components, you have to use spectral analysis. You can get as many of them as you like.

But, if you apply spectral analysis to indexes.... This is very cool! One has to know why and how to use it.

We are doing it right now, but not enough resources to get an accurate prediction (although it's the 4th stump, not the oldest trough),

and if the setting is too tight, it's "throwing the baby out with the soapy water".

My personal experience shows that the extrapolator starts to lie when USDX starts to move more than usual

i.e. USDx was relatively calm and its influence on the background of other currency was perceived
extrapolator sees it as noise (which can be ignored) and then USDx wakes up and "tails it".
And if you extrapolate the USDx, the lull is deciphered as a frequency beat,
and a wake-up is predicted at once...

 
Prival >> :

That's the thing: roughly. I recommend that you do the following. Take the exchange rates of these currencies at three points. Let's say at 16:00 over the last three days. And take the value of the index at these points. And calculate the network, but not with the help of the computer, but with the help of the pencil. With a pencil. And you will see everything. I think the questions will disappear.

I agree there are discrepancies not only (not even so much) in the indices, but even in EURUSD/GBPUSD!=EURGBP quotes themselves, but they are insignificant,
and the problem is not solved globally. After all if we do not calculate the indexes correctly, they will not only fail to take their frequencies with them, they will add other values.
As a result extrapolation will not be simplified but will not be possible at all.

 

I tried to calculate exchange rates from a system of equations:

EUR/USD = EURUSD

USD/JPY = USDJPY

and so on, plus a normalising equation

EUR*USD*JPY*CHF*GBP*CAD=1

After recalculation of obtained rates back to currency pairs I got deviation of 6-12%. However, it turned out that the pound price is 100 times more than the yen price, which means that the weights of currency pairs are different.

Now I converted to the quotation correlation with its moving average, the deviation obtained during reverse recalculation is about 0.1 per cent. This is the only reason why the price has appeared in relative units instead of absolute ones. To convert it into absolute units I am now trying to take the product of rates with different moving average lengths.

 
Urain писал(а) >>

I agree that there are differences not only (not even so much) in indices, but even in EURUSD/GBPUSD!=EURGBP quotes themselves, but they are insignificant,
and the problem is not solved globally. After all if we do not calculate the indexes correctly, they will not only fail to take their frequencies, but also will add other values.
As a result extrapolation will not simplify but will be impossible.

Let's do it this way, flies separate, cutlets the other way.

Let's build an index. It must be correct. If so, what is the criterion for the correctness of the index?

Let me explain. Suppose we calculate EUR/USD. For this purpose we use the index EUR and the index USD. We calculate it, divide it, it does not coincide with the current EUR/USD quote. And now think if the EUR/USD quote is true (because it is chosen as a criterion), then why do we need these calculations? The "truth we already know" is the EUR/USD quote, we just take it and that's it.

Now let's talk about the extrapolation. For extrapolation, the most important thing is the model that underlies the extrapolation. There are several types of errors in extrapolation. There are model errors and errors of the current measurements. Which in the end lead to extrapolation errors. Let me explain. If we know for sure that the cotier moves sinusoidally (this is the model), then having the current measurements we determine the amplitude, frequency, and phase of the oscillation. We substitute them into the sine wave and extrapolate, if we didn't measure accurately (amplitude and/or frequency and/or phase), there will be extrapolation errors.

You are trying to reduce the error of current measurements for extrapolation, this is good. But it is not the accuracy (ignorance) of the model that introduces the main error. And another simple example, let's say we are trying to predict the speed of a car driving along the Moscow Ring Road = kotir. And for this we use speeds of all other cars (measure their speeds and extrapolate). You can do that too (some analogue of group speed). Or you can just measure the speed of the car we're interested in and extrapolate exactly that speed. We can do it both ways, but models will differ, for a group speed has its own model, for a separate car (quote) has its own model, and each method will have its own measurement errors.

 

For individual currencies, you can also do tehanalysis, sometimes it helps...

Here is an example from a neighbouring thread http://www.umis.ru/study/trading_school/fc_methods?start=0