Machine learning in trading: theory, models, practice and algo-trading - page 1592
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 do not know, this is a complicated interpretation. Market returnees are a mix of several distributions, each of which may be stationary. If you separate the flies from the cutlets, you can get a good TS.
I do not argue, most likely you can, the question is which way is shorter and more effective. I read a couple of dozen first pages of AK's topic, saw several times how he makes there such assumptions, essentially conjectures, which in my opinion cross out the whole point of statistical research as to fit the model to the "desired" result. I could be wrong, as I don't know much other than higher technical and google) But I've become uninterested so far.
Market returns are a mixture of several distributions, each of which may be stationary this is closer to the body, i.e. is the normal one part of it?
But anyway it seems to me that any distribution may be stationary only till the moment when something changes in the market - mood, news, trend etc. If you take it into account and learn how to cook it, then I think it will be worth it.
Market returnees are a mixture of several distributions, each of which may be stationary this is closer to the body, i.e. is the normal one a part of it?
But all the same it seems to me that any distribution can be stationary only till the moment when something changes in the market - mood, news releases, trend, etc. If you take this into account and learn how to prepare, then I think it's worth it.
The clustering algorithm itself makes them more normal, for example if it is a Gaussian mixture, there will be selected points described by a Gaussian for each cluster, with some outliers
the main question is in checking with new data
Gentlemen, could you please suggest the most grail neuron for flyleaf?
I broke the history into sections, and now I need to pull the neuron up here:
As you can see, there's really no correlation between the pairs.
Binary trading is lame.
You shouldn't draw conclusions from a piece of history. I do not know what is the logic of the indicator, but it is easy to check the relationship - create a predictor for each curve and record the position of this curve relatively to all other curves, in order on each bar, you will get 6 predictors with values from 1-6. You can try to set a target for each curve (currency pair) separately - whether price went up or down, or how its position changed relative to a specific curve or relative to all curves, and so on - different targets here, and see what will be classified better, what will be better, and how to use it in trade.
You should not draw conclusions from a piece of history. I do not know what the logic is in the indicator, but to check the relationship here is simple - create a predictor for each curve and record the position of this curve relative to all other curves, right in order on each bar, you will get 6 predictors with values from 1-6. The target is done as a shift of N bars in advance, you can try to separately target for each curve (currency pair) - whether the price went up or down, or how the position of the curve against a particular curve or against all curves, and so on - different targets here, and see what will be the best classified, what will be better, then think how to use in trade.
It has been suggested here that IO can help a lot when there is a stationary process
in what way?
Because the bad thing is not that a stationary process can become non-stationary, but that it can suddenly become non-stationary
It has been suggested here that IO can help a lot when there is a stationary process
in what way?
because the bad thing is not that a stationary process can become non-stationary, but that it can suddenly become non-stationary
Once again, the statistical (non)stationarity(time distribution variability) of a time series(s), has very little to do with the predictability of future increments using MO. Gaussian noise is stationary but not predictable, sort it, add heterocessativity, it will be predictable but non-stationary.
The real problem is in the variability of the market due to the constant efforts of most participants to outplay each other, because of this the patterns found on the history are usually already found by others and you can expect them to reverse, and "forces of nature", that is the fundamental and various time-shifts. It's all very complicated.
Once again, statistical (non) stationarity (variability of time distribution) of time series(s) is very weakly related to predictability of future increments using MO. Gaussian noise is stationary but not predictable, sort it, add heterocessativity, it will be predictable but non-stationary.
The real problem is in the variability of the market due to the constant efforts of most participants to outplay each other, because of this the patterns found on the history are usually already found by others and you can expect them to reverse, and "forces of nature", that is the fundamental and various time-shifts. It's all very complicated.
What makes you think that white noise is unpredictable?
If the series is stationary, there is no need to use increments.
Can't you apply any flat strategy to this chart?
Well if it (log)returns then I can not, it is obvious that it is not about the price)))
Once again - MOs use increments, only because the original time series is non-stationary.
If the series were stationary, then there would be no need for gradients.
It's to this "Once again, the statistical (non)stationarity(variability of time distribution) of the time series(s), has very little to do with the predictability of future increments using MO." (с)
It's not nice to edit your posts retroactively if they have already been answered