Whether there is a process whose analysis of one part does not allow predicting the next part. - page 8
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
There are a huge number of theories, sciences and tools that seem to be applicable to a quote. in these circumstances, it is crucial to decide what we want.
We want the model created on historical data to be applicable on the next bar. That is enough. That's all.
I believe that the characteristics of this model are established on extremely limited samples - dozens of observations. A large sample is needed to test the stability of this model. This is the second characteristic of the model. The stability of a model is determined by its behaviour on variable variance and breaks. This is the second characteristic of the model. If we select the tools and methods for this, it would be a huge step forward, as the toolkit would become observable.
This is news to me. A stationary series is predictable by definition - within a sko. An unsteady one has no sko - what's the prediction? But it's not just about the sko.
By what other definition? Where does the RMS disappear in a non-stationary process? Have you heard of random variables with infinite variance? How is predictability in principle related to the existence of RMS at all?
I would still like to return to the issue of detrending.
What are we detrending?
Level? Straight line? Curve? Splines?
What about phase? Do we detrend it too?
Is there just one trend or many? Maybe a wavelet?
So fixation on deterministic and stochastic trends for forecasting is a harmful thing, because it suggests to solve problems that the trader does not have.
By what other definition? Where did the RMS in a non-stationary process disappear to? Have you ever heard of random variables with infinite variance? How is predictability in principle related to the existence of RMS at all?
Let me restate your idea - "if you remove non-stationarity from a non-stationary process, it becomes stationary" Wow, how profound! Focusing on stationarity and inexplicably substituting it for predictability is no less damaging.By what other definition? Where does the RMS in a non-stationary process disappear to? Have you ever heard of random variables with infinite variance? How is predictability in principle related to the existence of RMS at all?
That's your answer. Non-stationarity of variance makes it impossible to predict, i.e. the prediction error becomes uncertain.
The fixation on stationarity and the incomprehensible substitution of it for predictability is no less damaging.
Not a substitution, it flows out.
Why fixation? I am not the only one, by the way.
The thing is absolutely clear. Prediction is not conceivable without prediction error. Error cannot change arbitrarily, at least on historical data. What is not clear about it? Or is there something else?
If you remove non-stationarity from a non-stationary process, it becomes stationary" Wow, how profound!
Never said that. All I ever said was to take into account, to simulate.
By what other definition? Where does the RMS in a non-stationary process disappear to? Have you ever heard of random variables with infinite variance? How is predictability in principle related to the existence of RMS at all?
Let me restate your idea - "if you remove non-stationarity from a non-stationary process, it becomes stationary" Wow, how profound! Focusing on stationarity and inexplicably substituting it for predictability is no less damaging.I do not understand why stationarity is equated with predictability. If that's what you're trying to achieve stationarity - take an ordinary SB, there's an ideal stationarity with an ideal RMS. Now try to build a model on it - the result is guaranteed to be random.
I don't understand why the equation between stationarity and predictability is being made. If you are trying to achieve stationarity in this way, take a normal SB, there is perfect stationarity with an idyllic RMS. Now try to build a model on it - the result is guaranteed to be random.
For me, everything is clear. The prediction is zero mo. This is the basis of TCs on a return to Mo for random deviations from Mo.
To extract a quasi-stationary process of price increment with positive Mo from the price series ;)
to isolate a quasi-stationary process of price increment with positive mo ;)
Not to be unfounded I will give examples for each statement. I will intentionally try to make it more complicated.
By what other definition? Where does the RMS in a non-stationary process disappear to? Have you ever heard of random variables with infinite variance? How is predictability in principle related to the existence of RMS at all?
That's your answer. Non-constant variance makes it impossible to predict, i.e. the prediction error becomes uncertain.
This is not an answer, but a question to you in relation to your own delusions. I will give you an example to refute them.
A non-stationary process with density 1/pi*1/(1+(x-x0)^2), and expectation x0 is also a random variable, albeit for complete uncertainty - with unknown distribution (stationary or not - also unknown). And let the correlation time of the process be non-zero, i.e. the integral of the product of ACF(tau,t)*tau is greater than 0 for any t.
What do we know about the process:
a) Its variance is always infinite (calculate the integral if you don't believe).
b) It is non-stationary both in the narrow and almost probably broad sense. The first follows actually from the definition of stationarity in the narrow sense, because the density of the process is not constant, the second follows from the unknown properties of the process x0.
Nevertheless, in spite of all the aggravating circumstances, under certain conditions, namely where the correlation time (it may not be constant - the process is non-stationary!) exceeds some threshold value, we can make a prediction with a perfectly acceptable finite variance. It is the condition of good (exceeding some threshold, which in principle can be calculated) correlation of the process at least at some moments, and our ability to identify these moments are sufficient conditions for the possibility of prediction. However, the facts of non-stationarity and lack of dispersion do not matter by themselves.
The fixation on stationarity and the incomprehensible substitution of it for predictability is no less harmful.
Not a substitution, it flows out.
Why fixation? I am not the only one, by the way.
The thing is absolutely clear. Prediction is not conceivable without prediction error. Error cannot change arbitrarily, at least on historical data. What is not clear about it? Or is there something else?
The error can vary as it wants, and our job is to be able to calculate it. If we can do that, why can't it be different for different points in time? Your fatal error is that you do not distinguish between the variance of the forecast and the variance of the predicted process, which are completely different things and not rigidly related to each other. The presence and depth of the relationship between them depends on many factors, including the amount of knowledge we have about the process, the forecasting methods we have in our arsenal, and only lastly on the properties of the forecasting process itself. The example above confirms this.
It is true that you are not the only one who is fixated, because people tend to err not on their own, but on the advice of authorities.
Hmm.
I'm glad I was able to excite the best minds on the forum.
With your permission, I will humbly stand aside and read. (chuckles): Thank you.