Econometrics: one step ahead forecast - page 93
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
Big news. AND ARIMA? AND ARCH? Is this for which rows?
cotier = trend + noise
On the left it is non-stationary, but where on the right?
The news is not big and old - these methods are developed and applied to stationary series. But for non-stationary series, methods are used to "bring them to a stationary form" in order to be able to apply these methods.
If a series is non-stationary, then it is non-stationary. The series has unstable statistical characteristics. You can chop a series into pieces and it will be stationary in every piece.
If the price rose and then fell sharply, it is stationary in the ascending phase. But the whole price series is non-stationary (subject to sudden changes).
The news is not big and old - these methods are developed and applied to stationary series. And for non-stationary ones, methods are used to "bring them to a stationary form" in order to be able to apply these methods.
If a series is non-stationary, then it is non-stationary. The series has unstable statistical characteristics. You can chop a series into pieces and it will be stationary in every piece.
If the price rose and then fell sharply, it is stationary in the ascending phase. But the whole price series is non-stationary (subject to sudden changes).
A change in the trend is not an indication of non-stationarity.
Let's start at the beginning. I use the following definition of stationarity: a series is stationary if mo = constant ("almost" constant) and dispersion = constant ("almost" constant).
As long as there is a deterministic component in the series, we cannot talk about statistics. That's why we work with the residual after subtracting the deterministic component. After this procedure the problem becomes qualitatively easier, because the residue is usually much smaller than the candle length.
The news is not big and old - these methods are developed and applied to stationary series. But for non-stationary ones, methods are used to "bring them to a stationary form" in order to be able to apply these methods.
If you can still agree about ARIMA, then ARCH is a purely non-stationary thing. I use specific tests aimed at certain subtleties in non-stationarity and once they are identified, pure non-stationarity is modelled. The residual often turns out to be stationary.
And for non-stationary ones, methods are used to "bring them to a stationary form".
If the method you mentioned is available, then we are working with a non-stationary series?
The change of the trend is not a sign of non-stationarity.
Let's start at the beginning. I use the following definition of stationarity: a series is stationary if mo = constant ("almost" constant) and variance = constant ("almost" constant).
As long as there is a deterministic component in the series, we can't talk about statistics. That's why we are working with the residue after subtracting the deterministic component. After this procedure the problem becomes qualitatively easier, because the residue is usually much smaller than the candle length.
Slightly wrong - astationary random process must have all its probability characteristics independent of time. If the price was rising roughly linearly over a long period of time and then suddenly declined sharply, it was a stationary process before the decline, because if the series were broken down into chunks, the statistical characteristics of those chunks would be roughly the same. But after the decline, its probability characteristics changed - it became non-stationary (the MO changed, the variance changed).
It is possible to identify the deterministic component in any series and we should talk about statistics. If the series is non-stationary, this deterministic component will have a very small predictive power.
Slightly wrong - astationary random process should have all of its probability characteristics independent of time. If the price was rising roughly linearly over a long period of time and then suddenly declined sharply, then it was a stationary process until the decline, because if the series were broken down into chunks, the statistical characteristics of those chunks would be roughly the same. But after the decline, its probability characteristics changed - it became non-stationary.
My definition is constructive - it allows for a modelling plan and defines the objective.
Step 1: You cannot say anything definite as long as there is a deterministic component in the series. For me it is an axiom.
My definition is constructive - it allows for a modelling plan and defines the objective.
Step 1: You can't say anything definite as long as there is a deterministic component in the series. This is an axiom for me.
In any and every series there is a deterministic component. The question is the quality and accuracy of prediction
If the method you mentioned is available, then you are working with an unsteady row?
We tried and tried to apply it to trading - the result was deplorable. I'd rather use TA.
Although there is room for a feat.
There is a deterministic component to any and every series. The question is the quality and accuracy of prediction
Don't get sidetracked.
We distinguish the deterministic component. What about the residual? Again we check for the deterministic component. The reason is the old one. Get the noise. Get the noise without the deterministic component, we can reason.
Don't go sideways.
We distinguish the deterministic component. What about the residue? We check again for the deterministic component. The reason is the old one. Get to the noise. Get the noise without the deterministic component, we can reason.
What for? What is there to speculate about? You isolate the deterministic component, make a model, test it, analyse it, discard it (just kidding).
If the child component is of good quality, we trade. What's the rest for?
The question is the quality of this component.
No matter how you look at it, the statistics will show what came before. And there is no way it can show "what will be"... Pure guesswork. Maybe that would work for you, then?
>