Econometrics: State-space model forecasting - page 14

New comment
 
Vizard:

since we are only interested in 1 point (prediction for 1 step) we need to look at its change...i.e. we ran the model - recorded the prediction point...submitted a new data line...ran it again - recorded it and so on...we need to do it on a simple 1 model

This is what I'm doing, see above.
 
EconModel:

Unfortunately faa's advice on how to turn a point forecast into a trend forecast has proved rather difficult for me. It's going to take some time. But it's moving....

Well, you shouldn't have rushed straight into redoing the model. Converting a "point forecast" into a "trend forecast" is done in one move.

NewPosition = (ForecastPoint > CurrentPrice) ? +1.0 : -1.0;

Where NewPosition == market net position in which we stand = direction of open order (+1==buy, -1==sell); // If already standing in right direction - do nothing, if need to flip - flip.

That is all.

 
avtomat:


;)) Alright then. Instead of earlier said "Running in the tester will not give anything", I will say this: "Running it in the tester will do something."

Although this 'something' will not be the solution to the problem at hand. But it might suggest a way forward.


Thank you, oh benefactor.....


:)

 
MetaDriver:

Thank you,oh,my benefactor......


:)


;)

Although I don't like the tester in some places, I'm not a fascist to shoot it. You're welcome to it.

 
EconModel:
This is what I do, see above.
If the data is fed on 1 line at a time (i.e. before the new line model could not be seen in any way) - then retrain the model - then fine...
 
MetaDriver:

Well, you shouldn't have rushed straight into redoing the model. Converting a "point forecast" into a "trend forecast" is done in one move.

Where NewPosition == market net position = direction of order to be opened (+1==buy, -1==sell); // If we are already standing in the right direction, we do nothing, if we need to flip, we flip.

That is all.

Illusion.

There are models and they are not simple. And they differ from your advice in that (like all econometrics) it gives an answer to the main question: can the result be trusted? As well as the tester. On what basis should the tester be trusted? And to simply trust is to go to church....

 
EconModel:

Illusion.

There are models, and they are by no means simple. And they differ from your advice in that (like all econometrics) they answer the main question: can the result be trusted? As well as the tester. On what basis should the tester be trusted? And to simply trust is to go to church....

Why trust....here the tester simply acts as a simulator of real events...feeds the data (let it feed the data with a delay on the retrain model) gives the cut as an equi...set it for a couple of days...then look at it...
 
Vizard:
why believe....here the tester simply acts as a simulator of real events...feeds the data (let it feed the retrain model with a delay) gives the cut as an equi...set it for a couple of days...then look at it...
The market here is not stationary. So: either the model was originally conceived and really fits this non-stationarity or it doesn't. For me (I so conceived) it has to fit. Apart from that, the model has to fit the purpose. The goal is to follow the trend, and my model does not support this goal, but makes a point prediction. What is there to test? There is no object to be tested.
 

Beautiful, beautiful... Wonderful... Page 14... what model, what predictions. That's a lot of clever words.

Are we going to trade or not? (strictly speaking)

 
MetaDriver:

Well, you shouldn't have rushed straight into redoing the model. Converting a "point forecast" into a "trend forecast" is done in one move.

Where NewPosition == market net position = direction of order to be opened (+1==buy, -1==sell); // If we are already standing in the right direction, we do nothing, if we need to flip, we flip.

That is all.


How easy you make it... Especially the flip move. Snap, and you're in the hole. ;)))
1...789101112131415161718192021...26
New comment