Econometrics: one step ahead forecast - page 96

New comment
 
Mathemat:

The same thing is being done by the topicstarter, but it has nothing to do with ergodicity in particular. And it is clearly not enough.

That's it, I have no more questions for you.

Alexey, and in fact Demi is absolutely right.
 
Right about what?
 
avtomat:

faa1947, why are you so attached to this EViews? You almost pray for it... And you present it as something perfect and worthy of unconditional emulation. It's like a guide to action... I don't understand... Is that the end of your knowledge? May I suggest some literature?

By the way, you still did not answer my question: what theoretical principles of the method of state space are not clear to you? After all, in order to explain in a comprehensible way, I need to figure out from which place the difficulties begin, and what is enough to mention in passing.

Above in the topic I wrote by virtue of my understanding, but you did not answer.
 
Mathemat:

That's what the topicstarter is doing, but there's no particular relevance to ergodicity. And it's clearly not enough.

That's it, I have no more questions for you.


The essence of my statements is not about whether the topikaster is doing something right or wrong. The point is the applicability of mat statistics methods for non-stationary and non-ergodic series in general.

And these economic time series are non-stationary and non-ergodic. And whatever data processing package you use, it's useless.

Adaptive methods, maybe, but the application that I saw did not give any tangible results.

There are attempts to apply adapted methods for non-stationary series = neural networks + fuzzy logic circuits. There were examples of such applications for commodity markets.

 

Statistics can also be applied to these economic rows, just wisely.

All I said was that you suggest doing the same thing as the topicstarter. But even faa has admitted that this is not enough.

And it makes no difference what we call them - adaptive or otherwise. In any case, statistics must remain the basis. Statistical methods, by the way, do not have to be classical. It can also be something new - say, a Bayesian approach.

 
Mathemat: But even faa has admitted that this is not enough.
Not enough for the outlined model, which is primitive, poorly justified and does not use a hundredth of an econometric.
 
Mathemat:
Right about what?
You need to get to the other side of the broad river. You know there are bridges for that purpose. You are looking for a bridge, that's right - you must find one and cross the river safely. But here's the trouble - there's no bridge in the vicinity or in the distance. What to do... And he has to get to the other side at any cost. People live there and they tell you that they had no bridges here in the past, no matter how hard you look for them. And they cross the river on rafts. And they suggest to you to build a raft and use it for crossing. But you say, "No! That's not right. It's not the right crossing. For it to be right, there has to be a bridge." Do you want a bridge or do you want the other side of the river?
 
And more specifically, Oleg? What's new compared to the topicstarter reported by Demi?
 

Mathemat:
А поконкретнее, Олег? Что нового в сравнении с топикстартером сообщил Demi?

I'm kind of uncomfortable, I have to answer that:

1. econometrics is a way of applying statistical methods to economic forecasting. There are no "proprietary" or "original" methods.

2. the series under study are non-ergodic and non-stationary and for this kind of series the overwhelming majority of mathematical statistics methods are unacceptable.

3. these series may be transformed and violated but they will remain nonstationary and nonergodic.

4. you can separate a child component from noise and then make another component from noise and noisier noise and transform the noise and then molest it, get it drunk, cut off its arms and legs and burn it - still the series remains nonstationary and non-ergodic.

Conclusion: if a series is nonstationary and non-ergodic, its statistical characteristics and regularities may be obtained at any segment of the series, which will change entirely and unexpectedly in a short period of time, thus completely cancelling the prognostic characteristics of the found regularities.

Note: there is absolutely nothing new in what I have written. All this can be read in a more complete and less sloppy form in numerous textbooks and monographs.

 

By and large, I disagree. Starting from point 2. The series themselves, yes, they are, but some of their transformations may turn out to be such.

On the other hand, the series of returns of a regular Wiener process with independent increments and Gaussian increments are both stationary and ergodic. Nevertheless, this does not help us to work on the process itself and extract regular returns.

In short, it's still an ambush (that's not panic, I got used to ambushes a long time ago).

The most important thing is to find specific deviations from martingality and that's what you should use. But then we can't do without a meaningful model (with sense).

Games with regressions without meaningfulness have no sense.

1...8990919293949596979899100101102103...139
New comment