Obtaining a stationary BP from a price BP - page 6
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
So ..... ? So? Spectral power density is not necessary for traders because it does not allow predicting (synthesizing) the shape of the signal for the future. And 80% of all writings are devoted to this very "spectral density". It works in physics, in optics FOR ANALYSIS. But traders for extrapolation need SYNTHESIS after ANALYSIS, and it needs accurate synthesis. Therefore, if traders need a "spectrum", it should be for a "deterministic", i.e. non-random signal.... THIS (a sinusoidal spectrum) does NOT exist in time series. That's why Fourier analysis doesn't work in trading with ANY degree of accuracy.
So ..... ? So what? Spectral power density is unnecessary for traders because it does not allow predicting (synthesizing) the FORM of the signal in the future.
Not necessarily - a reversal prediction is enough for trend trading
That's why the Fourier analysis does not work in trading with ANY degree of accuracy.
I agree, but Berg's method seems to work and it doesn't smell Fourier and is applied to non-stationary signals. Trouble is, non-stationary signals themselves come in all sorts of ways.
Avals писал(а) >>
Reshetov, you still don't understand what we are talking about. No one suggested taking any noise as a model. I am too lazy to repeat the same thing.
...
However, a quality model must not only give a sufficiently accurate prediction but also be economical and have independent residuals containing only noise with no systematic components ...
Isn't your model supposed to reduce to stationarity on a variable time window and find the parameters of these stationary distributions? If you have something to say and discuss, why don't you start a thread?
You don't know the size of the window. It is not needed - it is enough to identify a timely reversal. There was a branch, but it was reduced to stationarity.
Avals wrote(a) >>.
Reshetov, you still don't understand what we are talking about. No one suggested taking any noise as a model. I'm too lazy to repeat the same thing.
...
However, a qualitative model must not only give a sufficiently accurate prediction, but be economical and have independent residuals containing only noise with no systematic components...
The model is your extrapolation, in short any BP forecasting method. The residuals, and in fact the prediction error, must have certain properties in order for the prediction model to be called adequate.
Well, who's arguing with that? Clearly, this is clear to a drunken hedgehog.
But for this it is necessary, though not always sufficient, that BP of errors (residuals) should be stationary and not too noisy.
Gentlemen, have you seen anywhere out there that MOs with variance are constant? I haven't. So I can't understand, what stationarity are you talking about? There is no stationarity at all.
Yeah, well, who can argue with that? Of course, it is clear to a drunken hedgehog.
But for this purpose it is necessary, though not always enough, that BP of errors (residues) should be stationary and not too noisy.
So I explain to the hedgehogs: the residuals are not just stationary, but normally distributed with mo=0. Of course, if they are normally distributed, but the MO is not equal to zero, then you can easily introduce a transformation after which the error will be NR with MO=0. It's boring to explain for a second time, especially after pearls with substitution of CB by its expectation.
And what is "not too noisy"? Dispersion constraint? :)
And it is useless to extrapolate anything. Perhaps there are some models for confidence intervals, but they don't work or they work in sporadic intervals. In general, conventional neuronics gives 57-65% of correct directions on some timeframes. Directions, I emphasize, not targets where extrapolation is applied.
Gentlemen, have you seen anywhere out there that MOs with variance are the same? I haven't. That's why I can't understand what stationarity you're talking about. There's no stationarity at all.
Just like there is no adequate price series extrapolation model :) But you don't need it for trading.
With the help of grasn (for which I thank him) I started to develop the following idea.
1. Construct a zigzag. Select the parameters so that the zigzag distribution is as close to normal as possible.
2. We subtract protection from the price, and obtain a series which is close to the stationary one.
3. We predict them for 2 steps - the end of the current wave and the next one. Perhaps we can use a clever regression model, for the time being I limit myself to the ordinary statistics.
4. Predict the residuals (I have not decided on the method yet).
5. Optimize forecast by min. GER between current rest ray + residuals forecast and price.
6. We get the optimization result - a trajectory.
If you are interested, colleagues, join us :-)
In the process...