The market is a controlled dynamic system. - page 20

New comment
 
-Aleksey-:
You are wrong, it counts unambiguously for a certain number of steps in any case - whether it is stationary or not. Perhaps you don't know how to do it. But that's not the main thing, the main thing is that you work and keep digging. If you want, I can teach you. Obtaining a stationary residual with HP indicates that a non-determined trend has been identified, which cannot be predicted by conventional econometrics further than one step, as far as I understand it. And why is it necessary to predict a stationary random residual instead of predicting a detrended trend? You're putting it all upside down. Before you use the package as a calculation tool, you need to understand the meaning of what you want to do.

You are wrong, it counts unambiguously for a certain number of steps in any case - whether it is stationary or not

Principle. Here is a graph of the prediction error for the model above:

Do you think you can trust the prediction with this error? I give descriptive statistics for the forecast error:


Obtaining a stationary residual using HP indicates that a nondeterministic trend has been isolated,

Fails to obtain a stationary residual with HP (see above).

HP is an analytical curve - i.e. it is deterministic and is used as a trend.

Which by conventional econometrics is not predicted by more than a step, as far as I understand it.

You completely misunderstand.

And what is the point of predicting a stationary random residual?

No one predicts a random residual - it's noise to a deterministic trend.

...to predict a dedicated deterministic trend?

There is no problem predicting a deterministic trend, which is what TA usually does.

You're putting everything upside down. Before you use the package as a calculation tool, you need to understand the meaning of what you want to do.

I apply the package and you are discussing without having seen it.

All my posts are based on calculations, I can provide links to relevant textbooks, nothing of the sort has been seen from your side so far.

 
Vizard:


+1

I'll send you a fact and forecast line...will you run it for adequacy?

I can. But please look at the model above. The success is in it. So far it has turned out that this model cannot be applied. So I wanted to start with the model. This is not the first thread on which I have tried to get the point across: a prediction can be made, but can it be trusted and by how much? It is the question of trust that is the main issue, and the rest is just numbers here and there.
 
Vizard:


and I'm talking about confidence in the forecast - so let's check it for interest...I'll just throw in two lines - one is the actual value and the other the forecast...run it and tell your opinion...then we'll see how the forecast works out in real life...

all roughly of course...for sp interest...

No problem with the model above, but it doesn't work - you can't trust the forecast with it.
 
faa1947, you do not understand what an HP filter is and why it is not a deterministic trend. Deterministic is a parametric trend. HP, on the other hand, is a non-parametric trend, and such trends are not used for forecasting, so can only be used for one-step forecasting. Sorry, of course, but I can't explain every letter to you, the volume would get very large.
 
Vizard:

Moved into the bag:

Looks like. Regression equation

FACT = C(1)*FACT(-1) + C(2)*HP(-1) + C(3)*HP(-2)

Smoothed and got the residual between the filter and the quotient is noise





я

Note that the absolute prediction error is 9 pips.


Joint graph:

Red is my prediction. It is much better than yours.

We can consider that we got the grail, Yay!!!

Please do not use. Residual heteroscedasticity test gives zero probability of no residual heteroscedasticity we should model heteroscedasticity. The prediction one step ahead will lie ungodly. That's the whole grail.

Let's look at the prediction error graph:

There is a bias, i.e. we have not completely removed the trend. Descriptive statistics gives the final verdict - you should not use such a wonderful forecast:


 
-Aleksey-:

faa1947, you don't understand what an HP filter is and why it is not a deterministic trend. Deterministic is a parametric trend. HP, on the other hand, is a non-parametric trend, and such trends are not used for forecasting, so can only be used for one-step forecasting. Sorry, of course, but I can't explain every letter to you, the volume would get very large.



I'm just fed up with the sermons, but I'd like something of substance.
 
Vizard:
show me which ones you think are best used...several are possible...
Any analytical function, preferably smooth, but you can also use a piecewise function if you understand why and when, dependent on t, but not on y(t).
 
Vizard:

conclusion on my forecast ? can i trust it ?

Mine is better, and it cannot be trusted, and I have shown why.

I cannot say anything about the credibility of your forecast, because I do not know how it is obtained, there are no calculations accompanying the forecast, such as what is the error of this forecast, what are the statistics of this error - and besides that there is a wagon and a small cartload of other things.

can you continue the series ? 20 bars ? your prediction...

I can't, I just know that the error will accumulate, add up and in 20 candles it will be 20 times bigger. That's why I've never done it. There is a so called "dynamic" forecast in the package, but nothing interesting.

 
Vizard:



They are more than 4mb. Google "econometrics", there are plenty of books, universities have such a speciality. The best way is to put EViews, there is a tool and links to each chapter.

Good luck.

 
faa1947:
I'm just fed up with the sermons, but I'd like something of substance.
I am a beginner like you. I think we should start with a simple one, i.e. give a logically consistent mathematical definition of the trend, its presence and form in terms of probability theory and mathematical statistics. The current trend, the predicted trend and their interrelation. As long as there is no such clear understanding, I don't think there is much point in counting anything.
1...131415161718192021222324252627...551
New comment
Reason: