Edge effect on the way to the GRAAL - page 8

 
Diamant:

I would like to know (c)

In fact, it is worth reflecting on the very formulation of the question.

Whether prices need to be researched so close to the right-hand edge at all. EUPOSHY.

We need to finally decide on the hindsight to be included in the calculations for forecasting.
 
yosuf:
We must finally decide on the hindsight to be included in the calculations for forecasting.
Yes. I know a doctor. Been around a long time. Taki will help you. Or just sympathetic. Well, let's...
 
yosuf:
We must finally decide on the hindsight to be included in the calculations for forecasting.
Initially, the model should include everything: smoothing and residual noise after smoothing, and the sample size is a tenth matter.
 
faa1947:
Initially, the model must include everything: smoothing and the remaining noise after smoothing, and the sample size is a tenth matter.
For my TS the most important factor was exactly the hindsight, because I think smoothing, noise extraction and similar attempts are harmful. You have to face the market face to face, without all the embellishments and curtains.
 
yosuf:
For my TS it was hindsight that proved to be the most important factor, as I consider smoothing, highlighting noise and similar attempts to be a harmful activity. You have to face the market face to face, without all the embellishments and curtains.

It is for the statistics committee to properly analyse the past.

The essence of trading is predicting the future, there profit. A forecast is not just an extrapolation of a model to the future, but also an evaluation of the error of that extrapolation.

 
faa1947:

It is for the statistics committee to properly analyse the past.

The point of trading is to predict the future, there profit. A forecast is not just an extrapolation of a model to the future, but also an assessment of the error of that extrapolation.


What do we do with the error estimate?
 
tara:

What shall we do with the error estimate?
After smoothing, maybe repeatedly, the error must be stationary: mo and variance are equal constants, in addition the variance must be homocedastic.
 
faa1947:
After smoothing, maybe repeatedly, the error should be stationary: mo and variance are equal constants, besides that the variance should be homokedastic.


San Sanych, I would like to keep it simple. Here in the future - profit is why we need a prediction. I don't share it but I don't dispute it either.

And with credibility - what are you going to do in real trading (flight, swim, race) ?

What is "homokedastic"?

 
tara:

What are we going to do about the error estimate?

Here is the model (regression): EURUSD = 0.999639862102*HP(-1) - 0.0587695175407*HP_D(-1) - 0.426573959137*HP_D(-2)

HP is the Hedrick-Prescott filter. We smooth out the quotient itself and the residue from the filter.

Here is the H1 forecast:

And here is the error of this forecast:

The error peak is 34 pips for the hour mark, and it is obviously a random value. Should we trust the forecast with such an error?

 
tara:


And with credibility - what are you going to do in real trading (flying, swimming, running) ?

I don't know what "credibility" means, but I know what model stability means. On historical data you need to achieve stability of the model.