Remembering veterans: Box and Jenkins - page 7

New comment
 
If we discuss Box and Jenkins, they introduced the requirement of reversibility of the model, i.e. if you add the right part, you must get the left part. If our TS consists of 2 bags, it is impossible to reconstruct the cotier, because the difference between the two bags and the cotier is not included in the TS - this TS does not have the reversibility requirement.
 
faa1947:

No. Many ideas are known: kotir value, kotir increment, trend, volu, levels .... Doesn't matter.

There is no difference. Why do you need to make a prediction of the absolute value for each moment in time?
 

faa1947: В надежде, что получим стационарный остаток. Это такая идея не первой свежести. Но в дальнейшем она развивалась, обрастала, но идея борьбы с нестационарностью жива до сих пор и по-моему не решена в полном объеме.


You must agree that this is only your assumption. And if it is an assumption, it may be wrong.
 
faa1947:

I gave the example of the mashups above. In EViews it is the button that calculates this residual.

I will give you the picture again.


I am not interested in a button in some proprietary software. No one is going to discuss buttons here. The program is closed and I don't know what it calculates and how many glitches the developers have crammed into the algorithm.

For special econometricians: give the mathematical definition of the residual between the quotient and TC, not the name of the button.

 
Avals:


1. You are setting an impossible task - it is impossible to build a model that will give a prediction of the absolute value of the price at any discrete point in time.


I do not set this task. Moreover, my other thread is called: one-step-ahead prediction.

To be more specific. I cited the ARMA model and calculation thereof at the beginning. This model is not workable because the cotier is non-stationary, as shown by the unit root test. But it also showed that the cotier increment would be stationary. In this particular case you were right, but my being right is the result of the test and your being right is so coincidental.

Hail to Box and Jenkins!

 
faa1947:

That is not the objective I am setting. Moreover my other thread is called: one step forward prediction.

Getting to the specifics. At the beginning I gave the ARMA model and the calculation from it. This model is not workable because the quotient is non-stationary, as shown by the unit root test. But it also showed that the cotier increment would be stationary. In this particular case you were right, but my being right is the result of the test and your being right is so coincidental.

Hail to Box and Jenkins!


So the problem you have been talking about so much does not exist? Quotation increments according to you are stationary and transaction results are price increments. :)
 
Reshetov:

I'm not interested in a button in some proprietary software. No one is going to discuss buttons here. The program is closed and I don't know what it calculates and how many glitches the developers have crammed into the algorithm.

For special econometricians: give the mathematical definition of the residual between the quotient and TC, not the name of the button.

By way of example: residual = cotier - f1 - f2, where f is the value of the two averages.

But your question stumped me. For me a model is always one or many formulas, the result of which I compare with what I predict: kotir means with kotir, increment means with increment. From the comparison I get the residual. What if it's an algorithm, and there are plenty of them in artificial intelligence? I don't know.

 
Avals:

So the problem you've been talking about so much is not there? Quotation increments according to you are stationary and transaction results are price increments. :)
In my case it was so coincidental, for your enjoyment. But the example of this thread says nothing. There was another EURUSD section in the Econometrics topic. Stationarity could not be obtained there by differentiation. But the problem was solved by modelling volatility with GARCH.
 
faa1947:
In my case, it was so coincidental, for your enjoyment. But the example of this topic does not tell you anything. There was another EURUSD plot in the Econometrics topic. There it was not possible to obtain stationarity by differentiation. But the problem was solved by modelling volatility with GARCH.


Ok, it worked or it didn't work. What's next? Why praise for these two guys?)
 
Avals:

OK, so it worked or it didn't work. What next? What's the glory for these two guys?)

1. Recognised non-stationarity. In 1972, it is strong in an efficient market.

2. Showed a way to deal with both non-stationarity and modelling an initially non-stationary quotient.

3. Formulated the requirement of reversibility of the model.

1234567891011
New comment