Machine learning in trading: theory, models, practice and algo-trading - page 2527
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
This is already much more interesting.
Maybe give up the notion that the market is a time series at all, and finally make a breakthrough in market analysis...
One does not interfere with the other. In modern financial mathematics, continuous and discrete time series approaches are quite compatible. The problem I see is that openly published concrete applications of this science are not well sharpened for our trader's needs.
don't twist it: trying to contradict me, you're still talking about your own... only about the time series... (and no one has cancelled the ways of discretization)...
price long term is not a function of time, I've made my point more than once (and I won't duplicate it)... I showed you where you can get autocorrelation in DL... I also told you what you will use for X and Y and for modelling what dependencies - I've already written it for the 10th time - it's up to developer to decide...
I am not the developer of your model - I do not need to prove the behavior of the price in time... (maybe I shouldn't have scribbled about DL - anyway everyone here thinks about his own things, refuting something or proving something to someone -- taking one word out of every discipline)... Engineers who do MO-ing (of which there are none here) will still understand the narrowness of the autocorrelation debate (for the sake of arcane speeches) even in trending, even in ticks, if the model is built in a much broader aspect and on a wider horizon of the learning set than the horizon where your fleas (autocorrelation) can come out... that's what Deep Learning is all about (to account for everything)
Yes, I twisted my words through my own experience. Please forgive me. if I offended.
The difference is that in the first case the ACF for all possible pairs of time moments is considered, and in the second case one of the time moments is fixed t2=n and many pairs of time moments( for example, the pair t1=1, t2=2) are dropped out of consideration. In the general case, the ACF is a function of two arguments. Only for stationary processes ACF can be considered as a function of one argument t=t1-t2 (lag).
The sampling ACF is always considered on the basis of a particular numerical sample (realization) of a process and always appears to be a function of one argument (lag value). This is the main reason why the sample ACF of an SB realization is not an estimate for its ACF).
Don't you think that by calculating the ACF for a pair of time moments t1 and t2 (lett1 < t2 for certainty), we are actually calculating the sampleACF at sample length n=t2and lag t2-t1. For an observer at time t2, the time series is represented by a sample of length t2. The observer does not know what will happen after time t2.
So as not to be completely unsubstantiated, here are my observations of auto-correlations of the real market:
Observation window for each value of the last 50 elements, offset by 1, 3, and 6 elements, respectively.
Pearson coefficient results from -1 to 1.
On the first screen of this analysis, for example, we could say that on the scale of one candle there was a stable negative autocorrelation (a positive value is followed by a negative one, and vice versa)
On the scale of 3 candlesticks it was the same, but less stable at the observation point, and on the scale of 6 candlesticks there was a mini-trend.
And on the second scale it is completely different (note the numbers).
But this is a time series, which for some reason everyone here doesn't like, and in general I know I'm stupid, and I don't understand anything. I do not want to hurt or teach anyone with this screenshot. And I do not urge you to predict anything by such calculations.
Don't you think that by calculating the ACF for the pair of time moments t1 and t2 (lett1 < t2 for certainty), we are actually calculating the sampleACF valueat sample length n=t2and lag t2-t1. For an observer at time t2, the time series is represented by a sample of length t2. The observer does not know what will happen after time t2.
Nevertheless, the observer at time t3, t3>t2 may well be interested in the correlation between moments t1 and t2. And your formula ACF(t) =sqrt((n-t)/n) does not allow him to calculate it (you only need to replace n with t3).
If the series were stationary, then ACF(t1, t2)=ACF(t2-(t2-t1), t2)=ACF(t3-(t2-t1), t3), but in general the second equality does not hold. You could say that the non-stationarity here is the presence of dependence on which point in time your observer is at (time inhomogeneity).
How can I get a co-op to work, but still pursue my own interests? The idea is that the ultimate (and potentially common) goal is to create a profitable system. As an option, all work with one piece of data. Here is the data of some instrument for ~4 months. It is known that expectation > 7, (commission 4.4, 5 digits) can be obtained on this data. The system should give profit for the previous 1.5 years, but more on that later.
Nevertheless, the observer at time t3, t3>t2 may well be interested in the correlation between moments t1 and t2. And your formula ACF(t) =sqrt((n-t)/n) does not allow him to calculate it (you only need to replace n with t3).
If the series were stationary, then ACF(t1, t2)=ACF(t2-(t2-t1), t2)=ACF(t3-(t2-t1), t3), but in general the second equality does not hold. You could say that non-stationarity here is the presence of dependence on which point in time your observer is at (time inhomogeneity).
But how it does not. It does! An archaeologist at time t3,t3 > t2 can dig up ancient records (e.g., on an amphora that is 3,000 years old) of SB of lengtht2. And, for example, will want to calculate the correlation between moments t1 and t2. And will do just fine with my formula: ACF(t) =sqrt((n-t)/n), where n= t2, t=t2-t1. Exactly because, it will, in fact, count the sample ACFwhen the sample length is n=t2and lag t2-t1.Feel that time momentt3 is introduced by you artificially.
But it doesn't. Yes, it does! An archaeologist at time t3,t3 > t2 can dig up ancient records (for example, on an amphora, which is 3 thousand years old) SB of lengtht2. And, for example, will want to calculate the correlation between moments t1 and t2. And will do just fine with my formula: ACF(t) =sqrt((n-t)/n), where n= t2, t=t2-t1. Exactly because, it will, in fact, count the sample ACF when the sample length is n=t2 and lag t2-t1. You feel that time momentt3 is introduced by you artificially.
Essentially, you come to the same two-argument function, but with a highly artistic description of the algorithm of its calculation)
Moment t3 is quite natural and you still need a moment t4, t4>t3, for which the prediction at moment t3 is built)
Essentially, you come to the same two-argument function, but with a highly artistic description of the algorithm for its calculation)
Moment t3 is quite natural and you still need moment t4, t4>t3, for which the prediction at moment t3 is built)
I propose to consider the phenomenon of ACF SB from the following positions. For the general population of SB (samples of infinite length) ACF = const = 1. For a sample of finite length n we can get an estimateof ACF with a typical error of the order of 1/sqrt(n). It is an error of this order that gives the ACF(t) estimation =sqrt((n-t)/n) = sqrt(1- t/n).
I propose to consider the phenomenon of ACF SB from the following positions. For the general population of SB (samples of infinite length) ACF = const = 1. For a sample of finite length n, we can get an estimate ofACF with a typical error of the order of 1/sqrt(n). It is an error of this order that gives an estimate of ACF(t) =sqrt((n-t)/n) = sqrt(1- t/n).
This would no longer be a SB, but a process with realizations-constants)
I make a counter-proposal to end our wonderful discussion before Kolmogorov and Wiener rise from their graves to beat us with sticks)