Finding the Natural Relationship of Currencies - page 4
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Well, if the synthetic, according to your hypothesis, is near the upper limit of the channel and should go down, then you sell x1 and x2 and buy x3.
So buying "+" and selling "/" )) Why don't you let the author try to answer that himself?
For example, I prefer to use multiplicative operators to describe synthetics, plus a degree operator to denote a fraction. I met people who use "+" and "-" (native, but acceptable approach). Now I've met people who operate with both +- and */ in their synthetic notation. Either I do not understand, or the author verbalizes a horse in a vacuum.
So buying "+" and selling "/" )) Why don't you let the author try to answer that himself?
For example, I prefer to use multiplicative operators to describe synthetics, plus a degree operator to denote a fraction. I met people who use "+" and "-" (native, but acceptable approach). Now I've met people who operate with both +- and */ in their synthetic notation. Either I do not understand something or the author is verbalizing a horse in a vacuum.
So "describe it to me" or "let the author do it himself"?
The author is not verbally abusing anything - he is asking the point
Andy Sanders 2017.01.02 19:46 EN
Perhaps someone knows and is willing to share the method for calculating the ideal synthetic?there is no such thing as perfect
A perfect synthetic is one that can select currency pairs to minimise variance, rather than adjusting the odds to those already selected.
it's not an ideal synthetic, but a dynamic flat portfolio which is rebuilt to fit the situation
Sometimes this rebuilding does miracles, but if the trend reverses, the flat model will break anyway.
to minimize variance there's nothing better than the ICG algorithm (Carl Pearson guarantees it)
Two already tested methods - PCA(https://www.mql5.com/ru/code/16997) and Linear Regression(https://www.mql5.com/ru/code/11859) have a major and inexhaustible disadvantage - they try to create correlations where there are none. In fact, here there is curve fitting through fitting the share of each instrument in the portfolio, as a result, at OOS it all quite predictably collapses.
no one promised that the forward will be the same as the calculated interval
Moreover, the description specifically states that the model will inevitably break down over time
(well, let it break, after all we can trade the very fact of model breakage)
Dmitry2017.01.02 20:32
There is no combination of currency pairs in forex that gives a stationary residual.
it is true! (except for castles-triangles, which are known to be a complete fade)
the sum of non-stationary series is non-stationary (there is no cointegration between them)
those who talk about stationarity in the Forex market are either wicked reckless robots or lost souls
Stationarity is possible only on a local segment
trading in the spirit of paired trading is possible only on linked indices or stocks (futures and options are also suitable)
The only thing is that the volatility of a portfolio is always less than the sum of volatilities of its components.
and it can be essential and change the distribution in a sector at least a bit
Andy Sanders2017.01.03 10:05
Also, yes, the correlation does change, but ... it does change on the principle of somewhere losing, somewhere gaining, i.e. if you choose the right set, the rise in one will be roughly offset by the other
Unfortunately, it is not always so
Trading according to the scheme of entering a correction in the direction of the trend is only possible in a stable trend market
it means that you should somehow have reasons to expect that the trend will continue
recently it was possible to do this with portfolios with a leading yen
but of course that would have to come to an end
it would be strange if some currencies would go up or down indefinitely
it's already buckling now (see p1.jpg)
better to trade trend breaks imho or going out of flat
The price distribution may have tails, but the main thing is that it should be a bell-shaped and relatively narrow, i.e. it should have the area of average values somewhere
Such models can easily be built on intervals without any strong news, but with the understanding that you need to be able to change your position in the air (close or re-build) at the news
Theideal solution is to have the possibility to choose the slope angle of the regression, along which the model runs, but this is of secondary importance.
in fact, the regression angle is a contingency caused by the scale (total value of the portfolio)
In brief, I am interested in how a portfolio is selected for strategies like Buy & Hold, buy and hold, what characteristics are optimised and how to speed up enumeration of possible portfolios
buy and hold is more like for stocks and there you need to find undervalued assets
Chorex is more of a return than a trend-following one but it depends on the timeframe.
portfolio rebalancing can be organised in 2 ways:
1 - recombine all possible combinations and build all models and evaluate them
this variant is good when the number of combinations is small
2 - Pre-select the most trendy instruments and use them to build a model.
this variant is more economical
for the horex, as a rule, many portfolios will be similar since the input currencies are the same
the 7 major combinations are traditionally considered (they have a lower spread)
but then there will either be a higher exposure to the dollar
or the dollar volatility will overlap within the portfolio
in any case, it is not good
imho, it is better to choose a small portfolio of no more than 3 or 4 instruments (to save on the spread).
but include crosses in the portfolio so that no currency is repeated
in this case there will be 5000 combinations of threes and so many fours
if someone needs, then in the attachment is an algorithmic search of all such combinations
Perhaps someone knows and is willing to share the method of calculating the perfect synthetic?
By ideal synthetician I mean one that can select currency pairs to minimize variance, rather than adjusting ratios to those already selected.
Two already tested methods - PCA(https://www.mql5.com/ru/code/16997) and Linear Regression(https://www.mql5.com/ru/code/11859) have the main and inexhaustible problem - they try to create correlations where there are none. In fact, here there is a fitting through fitting the share of each instrument in the portfolio, as a result it all quite predictably collapses on OOS.
I just want to find combinations of instruments that have minimum spread on history WITHOUT any coefficients and the search of possible combinations should be non-linear, i.e. include linear (X1 + X2 + X3) and non-linear (X1 / X2 + X3 ^ 2) combinations.
Explain to me for example X1 + X2/X3 in terms of bought/sold? In the synthetic description, are there both '+' and '/' (additive and multiplicative operators) at the same time? What does the author mean by that?
2. Yes, X1 ... Xn are pairs, but perhaps among pairs it's hard to find synthetics, so using * and / you can get some non-standard combination ... e.g. buy more euros as currencies rather than as pairs, although ... maybe this is also a kind of coefficient and is a fitting ... it should be a pure pair combination (individual currencies will still make pairs), without any coefficients
EURCHF / USDCHF * EURJPY + GBPUSD - USDJPY or EURUSD * EURJPY + GBPUSD - USDJPY // here the increased share of EUR, everything with "+" is bought, everything with "-" is sold
mmm, yeah, looks like there's only + and - left, the rest is a fit
* and / will just give another pair with increased overheads in the form of spread, possibly with a changed coefficient, so it makes no sense either
yes, the task is simplified, go through all pairs via + or - and find either the smallest variance or at least the most normal distribution of prices in a synthetic move
Perhaps, someone knows and is willing to share the method of calculating the ideal synthetic?
By an ideal synthetician I mean one that is able to select currency pairs to minimize dispersion without adapting ratios to already selected ones.
Two already tested methods - PCA(https://www.mql5.com/ru/code/16997) and Linear Regression(https://www.mql5.com/ru/code/11859) have the main and inexhaustible problem - they try to create correlations where there are none. In fact, here there is a fitting through fitting the share of each instrument in the portfolio, as a result it all quite predictably collapses on OOS.
I just want to find combinations of instruments that have minimal dispersion on the history with no coefficients at all, at that it is desirable to have non-linear search of possible combinations, or rather to include linear (X1 + X2 + X3) and non-linear (X1 / X2 + X3 ^ 2) combinations.
You are discussing a particular method of building synthetics which I do not share. But I can suggest one thing from my methodology, which can be used by you.
Its essence is that any synthetic can be associated with a vector (a vector is a variable). The vector's components express the price of the currencies (in thousands of USD in my case) that are bought or sold. This vector has the following properties: the sum of its components is zero. The sum of absolute values of its components divided by 2 is the value of the synthetic in USD. I.e., margin at 500 leverage = the value of the synthetic * 2. If the vectors are pronormalised, then the scalar product of the vectors, gives the closeness of direction of the synthetic. The vector representation gives vector clustering.
I have the reverse task: to obtain the list of currency pairs being bought and sold and their volumes from a vector. That's why I don't need those tricky entries.
You are discussing a particular method of building synthetics which I do not share. But I can suggest one thing from my methodology, which can be used by you.
Its essence is that any synthetic can be associated with a vector (a vector is a variable). The vector's components express the price of the currencies (in thousands of USD in my case) that are bought or sold. This vector has the following properties: the sum of its components is zero. The sum of absolute values of its components divided by 2 is the value of the synthetic in USD. I.e., margin at 500 leverage = the value of the synthetic * 2. If the vectors are pronormalised, then the scalar product of the vectors, gives the closeness of direction of the synthetic. The vector representation gives vector clustering.
I have the opposite task: to obtain the list of currency pairs being bought and sold and their volumes from a vector. That's why I don't need those cunning entries.
the topic is nonsense and can be covered
here's the dollar index calculation
we take equal shares
USDCAD share
+
EURUSD share
+
USDJPY share
+
USDRUB share
+
GBPUSD share
+
share in NZDUSD
+
share of AUDUSD
We divide by the number of components, on the basis of this we draw a chart, preferably from 10 components and trade. If you are not satisfied with the accuracy, you will not trade this way in MT.
Here's the dollar index calculation.
what does this have to do with shares?
how are you going to control the volatility of the dollar if you're going to trade it?
the commentary is nonsense, you can cover it up
reread only carefully what was written, then you can litter
P.S. this "index" formula adds up tugriks, money and potatoes ... Good luck trading such an index