Discussion of article "Exploring Seasonal Patterns of Financial Time Series with Boxplot" - page 21
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
I wanted to write the next one, with a double charge, but now I doubt it ) The consequences are too unpredictable.
You're welcome to contribute.
I offered you to demonstrate the methodology of searching for a pattern on other CRs, take crosses, it is also a CR, but it does not contain information about USD, there you will not be able to fit the methodology to what is already visible on charts ;)
EURJPY.
I don't analyse boxplots by hours in details, it is necessary to detail them because of the median emissions, I show the general picture, I have detailed them for myself.
The chart shows that the hours are not the same as for EURUSD.
On the basis of purely exploratory analysis (0 optimisation) I have selected the watch, I left only buying, although there are options to sell:
I repeat, 0 optimisation, i.e. you can improve the parameters in the vicinity.
I don't want to enter into another polemic.
I wanted to write the next one, with a double charge, but I doubt it anymore ) The consequences are too unpredictable
You're welcome to contribute.
The SB is a tricky thing. Its trajectories are very likely to form something similar to cycles (I have read about it and faced it myself when modelling). Therefore, I try not only to draw, but also, if possible, to calculate. But I am not ready to demand the same from others - testing and optimisation solve approximately the same problems and are much more understandable to most people.
The SB is a rather insidious thing. Its trajectories are very likely to form something similar to cycles (I have read about it and encountered it myself in modelling). Therefore, I try not only to draw, but also, if possible, to calculate. But I am not ready to demand the same from others - testing and optimisation solve approximately the same problems and are much more understandable to most people.
I wonder how tests (calculations?) would be carried out on the example with found cycles, maybe it is possible to automate the enumeration/search by some criteria?
I wonder how the tests would be done on the example with the cycles found, maybe we can automate the enumeration/search by some criteria?
I would say it's two tasks
1) for a group of samples (sample = box) I would apply at least a median test
2) you can try to fit cycles by optimising the statistics from the first point. But I think it is more appropriate to use a matstat like the one used in speech recognition.
I'd say it's two tasks
1) for a group of samples (sample = box) I would apply at least the median test
2) you can try to select cycles by optimising the statistics from the first point. But I think it is more appropriate to use a matstat like the one used in speech recognition.
yes, I'm reading about the test, thanks
I'm interested in any other tests for... what should I call it, for interchanging between 2 distributions, i.e. dependence of points of the second or several subsequent distributions on the first one. Maybe Mood's test is about that too, I haven't got into it yet.
or some kind of lag test for interchanges.I wonder how the tests (calculations?) would be done on the example with the cycles found, maybe it is possible to automate the enumeration/search by some criteria?
there is such a cool "torch method" :-)
A series of XY with its scatterplots is projected onto the X axis at different angles, convolutions of different moduli are made by the density of "shadows".
The method of dividing in half reveals the approximate periodicity and the rate of rise/fall corresponding to it.
A bit like part of the fourier transform or principal component method. The computational complexity is N^2(almost overshoot) but that's the least of my worries.
there's this cool "torch method" :-)
a series of XY with its spreads is projected onto the X axis at different angles, convolutions of different moduli are made according to the density of "shadows".
The method of dividing in half reveals approximate periodicity and the rate of growth/decline corresponding to it.
A bit like part of the fourier transform or principal component method. The computational complexity is N^2(almost overshoot) but that's the least of my worries.
Do you have the names of the modules in python or R? Python is better.
Yeah, I'm reading about the test, thanks.
I'm interested in any other tests for... what to call it, for interchanging between 2 distributions, i.e. dependence of points of the second or several subsequent ones on the first one. Maybe Mood's test is about that too, haven't got into it yet
If I understood correctly, these are goodness-of-fit tests, there are two main types
1) one sample - it checks conformity to a known distribution
2) 2 or more samples - tests whether the distribution of the samples is the same. Kolmogorov-Smirnov two-sample test, Mood also, runs tests
If I understand correctly, these are goodness-of-fit tests, there are two main types
1) one sample - the conformity to a known distribution is tested
2) 2 or more samples - check if the distribution of the samples is the same. Kolmogorov-Smirnov two-sample test, Mood also, runs tests
Well, something like that, yeah. The idea in my head was just to find a regression of one set of points on another and see the correlations. That's the easiest thing I could think of. Then you could get additional signals. I don't know how statistically correct that would be.
I.e. dependencies caused by changes in the previous hour will be traded, not only in the current hour. From the behaviour in the previous hour, the signals in the current hour will vary, if a correlation can be found. This is closer to fitting or optimisation.