Regression equation - page 13
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
The main thing is to understand that, it turns out, ".... it's not about modelling complex natural processes (!!!), it's just about modelling the way decision-making by Forex market subjects..."
=)
Yep, it's all very useful.
As promised, picture. Performed analysis of pure price series (no preprocessing, trend removal etc) - AR(3) model on 11 counts. On the charts - prediction error: upper chart - for ANC, lower - quantile regression. Lines: blue - Close, green - High, red - Low (median and quantiles 0.9 and 0.1 were taken for QR, respectively). Blue lines are daily APR, for scale.
What I see here is the following:
(a) Absolute values of the error for the ANC in a calm market are almost the same as for QR, but(!) when a spike appears the ANC error changes more chaotically and generally reacts to it more weakly, while the second chart error looks more regular. This, in general, was the goal: to show the possibility of detecting "stationarity disruptions" at the expense of QR's ability not to react to these very spikes. And if they can be detected, it means that it is not outliers but additive random process, and not less stationary than the AP(3) from which we separated it.
b) if we consider outlier detection to be a useful signal, the second graph has many times more OSR, therefore a hypothetical :) trading system based on this effect will give as many times less false signals.
Of course, one can argue here, but this is what we get on M5 (AP(3), on 21 counts):
Here, already much more clearly.
In general, it seems to me that what I was saying is gradually confirmed. I will dig further in this direction.
I am attaching the library for QR calculation (compiled Gallant library, see link two pages before) and the header file with description. I don't attach indicators themselves, I can't separate them from the rest:))) but there is nothing difficult, the formula is already written
Gorgeous, alsu. On forex, it's time to scrap the MNCs :) Or is it not, Prival?
Great, alsu. In forex, it's time to throw the MNCs in the trash :) Or is it not, Prival?
I wouldn't throw it away. The Kalman filter, in its mathematics (essence) is an iterative ANC. https://ru.wikipedia.org/wiki/Фильтр_Калмана The wikipedia is not accurate. You can build it for more than just BGS. There's a caveat at the end that it's for coloured. I build for a uniform distribution law. There is a fundamental question in particular Stratonovich. We had a discussion about it with the students of Mechanics long ago. They solve it in the form of ITO, which I believe is wrong. There are some problems with time there. Well, such as this https://www.mql5.com/ru/articles/174.It must be solved exactly as Stratonovich gave, https://ru.wikipedia.org/wiki/Стратонович,_Руслан_Леонтьевич
If it works and lets you bring home a piece of bread, you should not throw it away. i feel a beating coming on... ))
It's great, alsu. It's time to throw out MNCs on Forex :). Or is it not time, Prival?
I can't say for forex, but I tried quintile regression in my stock-market strategy, where I have regression on regression and regression on regression. Quantile regression did not give me any advantage over ISC, it only takes longer to be calculated. Most likely due to banal process symmetry, because if it's symmetric, there's no difference between arithmetic mean and median... And that's where ISC rules. In my system, everything is symmetric - it may go up or down, with the same probability, although it's not normally distributed.
I fired Kalman, by the way. It took me a long time to bother with it, but, again, it gave me no advantages as compared to LOC, while it was consuming resources at the same time.
Well, a median is a median, and the methods for estimating it are well developed. What if you quantified it away from the median - say, 0.1 and 0.9?
2 Prival: I was joking about the dump, I had a smiley face there...
I can't say for forex, but I tried quintile regression in my stock-market strategy, where I have regression on regression and regression on regression. Quantile regression did not give me any advantage over ISC, it only takes longer to be calculated. Most likely due to banal process symmetry, because if it's symmetric, there's no difference between arithmetic mean and median... And that's where ISC rules. In my system, everything is symmetric - it can go up or down with the same probability, even though it's not normally distributed.
What do you mean there's no advantage over MNC? How did you measure it? Quantiles are a completely different operation.
Whereas ANC reacts to any change in the BP sample, quantiles in most cases don't care - they don't change. The most volatile quantile is the median. Any other quantile is less volatile.
Well, a median is a median, and the methods for estimating it are well developed. What if you quantify away from the median - say 0.1 and 0.9?
What do you mean there is no advantage over the ISC? How did you measure it?