Econometrics: one step ahead forecast - page 58
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
Here we go...
The point is that without proven stability of one kind or another ("stationarity") so far I do not see any point in building a model. I myself look for this "stationarity" in something else - in information theory, which is also very closely related to matstatistics (a branch about feature selection, it is one).
The problem is that if the information links in question are at least quasi-stationary, one can theoretically profit from them. If there is no quasi-stationarity, it is not good.
Always when we see some regularity, similar to Wiener process, but we do not know its essence, i.e. its reasons, we are limited in forecasting only by quasi-stationary processes at most. (Quasi-stationarity is almost the same stationarity, but with slowly floating m.o., s.c.o. and ACF.)
These processes are some kind of derivatives of the main quotation process. They are not necessarily 1st or 2nd order differences at all. They can be any function of the underlying process. The main thing is to confirm the quasi-stationarity of this function and to build a mutually unambiguous bridge from it to the initial process.
I haven't been looking for stationarity in returns or differences of higher order for a long time. I am convinced that it is a highly complex process with deeply echeloned nonlinear memory. Trivial autocorrelation checks, on which SunSunich so insists, simply do not see these nonlinearities, i.e. do not notice the most essential complexity of the process.
We can argue about this for a long time, but I will stop here.
The point is that without proven stability of one kind or another ("stationarity") so far I do not see any point in building a model. I myself look for this "stationarity" in something else - in information theory, which is also very closely related to matstatistics (a branch about feature selection, it is one).
The problem is that if the information links in question are at least quasi-stationary, one can theoretically profit from them. If there is no quasi-stationarity, it is not good.
Whenever we see a pattern, which resembles a Wiener process, but we do not know its essence, i.e. its causes, we are limited in our predictions to quasi-stationary processes at most. (Quasi-stationarity is almost the same stationarity, but with slowly floating m.o., s.c.o. and ACF.)
These processes are some derivatives of the main quotation process. They are not necessarily 1st or 2nd order differences. They can be any functions from the underlying process. The main thing is to confirm the quasi-stationarity of this function and to build a mutually unambiguous bridge from it to the initial process.
I haven't been looking for stationarity in returns or differences of higher order for a long time. I am convinced that it is a highly complex process with deeply echeloned nonlinear memory. Trivial autocorrelation checks, on which SunSunich so insists, simply do not see these nonlinearities, i.e. do not notice the most essential complexity of the process.
We can argue about this for a long time, but I'll stop here.
Alexey, you put an equal sign between stability and stationarity. But this is wrong! They are different things. Moreover, and this can be demonstrated by examples,
1) the system can be stable under both stationary and non-stationary input flow;
2) the system can be unstable with both stationary and non-stationary input flow.
That is, the stationarity of the input flow is not a criterion for stability.
Thus, even if stationarity areas are found somewhere in the interior of the process, this in no way indicates the stability of either the system or the process as a whole.
No, no, I understand the difference between the two. It's just that I was very imprecise. I wanted to hint at some form of stationarity, not reducible to the stationarity of some difference in the initial flow.
I'm not looking for "areas of stationarity somewhere in the bowels of the process". I am also interested in the "global" stationarity of the entire flux - but of another flux related to the original process. Well, say, the "stationarity of the Information Matrix", which was discussed in the thread about feature selection. I.e. it is not even stationarity of a numerical stream, but something more complicated.
not familiar with that branch...
But the more complex constructions are considered, the less linear and stationary they are.
A bifurcation diagram would be very illustrative here.
The world is non-linear and non-stationary. Linearity or stationarity are only negligible blobs in the big picture.
.
It is probably more correct to say that non-stationarity is the norm and stationarity is only an anomaly.
It is probably more correct to say this: Non-stationarity is the norm, and stationarity is only an anomaly.
It depends on what you're looking at.
well, that's just to say... well... without regard to...
.
But we're looking at
something more complex.
that's just... well...
However, we are looking at
There are things that are very stationary. It may be an anomaly...
Paukas, did you catch the thread?