Pair trading and multicurrency arbitrage. The showdown. - page 93
![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
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
Linear regression and other statistical methods, give bias estimates, curves start to float over time, not corresponding to reality.
We've been there. You should only take what's here and now. No zero points.
nostril to nostril. :-)
from slightly different angles
the screenshot shows the same time.
Is this a new modification of the indicator?
Linear regression and other statistical methods, give bias estimates, curves begin to float over time, not corresponding to reality.
Even those methods that position themselves as suitable for real time, still either lag or give bias estimates.
What do you suggest?
What is the starting point?
In the screenshot, the window is 24 hours wide, from there it's cyclical.
aligned with the RMS and actually shows that all currencies have the same rate.
and trading moments are not a break from the limits, but the narrowing/expanding of the "bundle" and the distribution of currencies in it.
Depending on this,"We sell at the top and buy at the bottom" (quoting Roman). When expanding one thing, when narrowing strictly the opposite. Parallel movements (of limits and individual currencies) are prohibited
We've been there. You only take what's here and now. No zero points.
What have you been through?
regression can be retrained on every new candle ! that will be your "here and now" on the same regression that you kind of passed.
in the screenshot, the window is 24 hours wide, that's where the cyclicality comes from.
aligned by RMS and actually shows that all currencies have the same rate.
and trading moments are not a break from the limits, but a narrowing/expanding of the "bundle" and the distribution of currencies in it.
Depending on this,"We sell at the top and buy at the bottom" (quoting Roman). When expanding, it is one thing, when narrowing, it is strictly the other way round. Parallel movements (of limits and individual currencies) are prohibited
So first you bring all currencies to the same rate, and then how do you get the result that all currencies go around zero? Mashka?
What you've been going through.
regression can be retrained on every new candle ! that's your "here and now" on the same regression you've been going through.
I'm not gonna argue. I don't have time for this. Better show me how to bundle currencies.
What you've been going through.
regression can be retrained on every new candle ! that's your "here and now" on the same regression you've been going through.
I have tried recursive regression which has no window and many other regressions which are not even heard of, and which are supposedly suitable for changing in time.
On the example of recursive regression which has one parameter forgetting factor, and only new incoming data are calculated.
Alas, there is also a bias of estimates, to check just take the difference of two instruments as a benchmark for comparison.
Suppose the benchmark is at the same price level, well, just hanging out in a small flat, which is very often
and it begins to add new points, and it turns out that entered a position on the signal calculated by regression, in fact, the benchmark is still in place,
and the curvuline calculated by regression inexorably begins to go to zero, that is, the estimate shifts and does not correspond to reality.
That is, time ruins the statistical approach. I have been saying for a long time that perhaps this problem can be fixed by switching to timeless charts,
where the chart change does not depend on time, like renko and similar.
But mt5 doesn't have such charts, so I've given up on it. And building my own timeless charts is such a waste, I'm tired of coding everything that should be in the terminal as it is.
So first you bring all currencies to the same rate, and then how do you get the result that all currencies go around zero? Mashka?
currencies (pairs) are brought to a common base, in USD; scaled by investment yield (log chart),
aligned on the screen by the general "geom.centre" of the movement.
daily window is taken, not MA - just de facto difference between then and now.
The screenshot demonstrates: stability and constancy of the overall rate, that the choice of the "yield" indicator for the OY axis is correct. It also shows how much % you can make in a day with one trade:-)
I have tried recursive regression, which has no window and many others that have not even heard of.
On the example of recursive regression, which has one parameter of the forgetting factor, and only new data are calculated.
Alas, there is also a bias of estimates, to check just take the difference of two instruments as a benchmark for comparison.
Suppose the benchmark stands at the same price level, well, just hanging out in a small flat, which is very often
and it begins to add new points, and it turns out that entered by the calculated regression signal, in fact, the benchmark stands in place,
and the curvature calculated by regression inexorably begins to go to zero, that is, the estimate shifts and does not correspond to reality.
That is, time ruins the statistical approach. I have been saying for a long time that perhaps this problem can be corrected by switching to time-free charts,
where the chart change does not depend on time, like renko and similar.
But mt5 doesn't have such charts, so I've given up on it. And building my own timeless charts is such a waste, I'm tired of coding everything that should be in the terminal as it is.
Great idea, but then how to compare many pairs with each other if each pair has its own time.