Econometrics: one step ahead forecast - page 103

 
faa1947:
A pivot is the start of a trend

itself...
 
sever31:
and the beginning of the U-turn?
is a point, with no beginning or end
 
faa1947:
it's a point, with no beginning or end

mystique
 
Tantrik:

Mystique

I wish. Let's take the ZZ and predict reversals, no sweat.

I suggested Reshetov to put a neural net on ZZ.

 
faa1947:
Turns

yes, in order to save time, to test at key (pivot) points...
but model(s) should be unchanged (at least at first) and if lag window is used the lag would not change...
otherwise the fit to the story is a fit because there will be no pattern in the choice of lag counts...i.e. false adaptation in fact...
 
faa1947:
I wish. Let's take ZZ and predict reversals.

First the trends (70-80% of successful predictions) then we can watch reversals: )))) successful predictions!
 
Vizard:

Yes, in order to save time, test at key (pivot) points...
but the model(s) should be unchanged (at least at first) and if the lag window is used the lag count would not change...
Otherwise the fit to the story is a fit because there will be no pattern in the choice of lag...i.e....false adaptation in fact...
Just realised now. Predicting ZZ is predicting periodicity. How many times have I struggled with this above
 
faa1947:
Only just realised now. Predicting ZZ is predicting periodicity. I've been struggling with this for so long.


I don't care what you call it - seasonality, periodicity - it's all cyclical... it's there... if it catches and continues in the future, then you're good... no plum...

The zz is a tricky question - if you put it right on the peak, you may never be able to predict it (the peak)... you have to look at it visually - you can measure it... and if anything, move it forward by a bar...

You have to ask for what the model can really take...

 
faa1947:
Just figured it out now. Predicting ZZ - predicting periodicity. How much trouble with this above


on a spin - measure it if you're interested... ideal inputs for ts (on zz)

Files:
eurusd_k1.zip  292 kb
 
Vizard:


on a spin - measure it if you're interested... ideal inputs for ts (on zz)

ZZ step is too big. At 24000 observations there are only 137 long signals. Can't you post the tool itself so you don't have to write it?
Reason: