Econometrics: one step ahead forecast - page 51

New comment
 
Nevertheless, this unnatural "consistency" of error is also alarming.
 
avtomat:
I'm not quite sure what exactly is the target?
the residual from the model should have a variance equal to a constant, if possible
 
Mathemat:
Nevertheless, this unnatural "consistency" of error is also alarming.
This is the ideal. Obtained after modelling the residual with the ARCH model
 
This is an unnatural ideal: the ratio of the s.c.o. of the error to its m.o. is too low.
 
Mathemat:
This is an unnatural ideal: the ratio of the s.c.o. of an error to its m.o. is too low.
The ideal is a constant. Almost achieved. This means that there are no surprises in the prediction.
 
faa1947:
The residual from the model should have a variance equal to a constant, if possible.

If this is indeed the declared purpose, then you should reconsider the very basics. But it seems to me that you are confused again -- this time in your understanding of the purpose of modelling.

Don't think I'm trying to get at you here. No! What I'm suggesting here is that you should reconsider all your actions from the very beginning, starting from defining the problem and setting goals.

 
avtomat:

If this is indeed the declared purpose, then you should reconsider the very basics. But it seems to me that you are confused again -- this time in your understanding of the purpose of modelling.

Don't think I'm trying to get at you here. No! I suggest you to reconsider all your actions from the very beginning, starting from problem definition and goal setting.

Formulate another one. This is not my goal and it is clear - if the variance of the residual is constant, then there is no uncertainty in the prediction error.
 
avtomat:

this time already in understanding the purpose of modelling.

Trying to. Posted, but the idea went unheeded. The goal is the predictability of the model.

 
faa1947:
Formulate another one. This is not my goal and it is clear - if the variance of the residual is constant, then there is no uncertainty in the prediction error.
hmm... Actually, the goal is usually to minimise the error, not some constant at all. Moreover, in this formulation of the problem, the presence of a "dead" constant would indicate inefficiency of the minimization algorithm or an error in the algorithm itself.
 
avtomat:
hmm... Actually, the goal is usually to minimize the error, not some constant at all. Moreover, in this formulation of the problem, the presence of a "dead" constant would indicate inefficiency of the minimization algorithm or an error in the algorithm itself.
Minimization does not solve the problem. The value of the prediction error is questionable, since it is the error of both the correct prediction and the incorrect
1...444546474849505152535455565758...139
New comment
Reason: