Econometrics: one step ahead forecast - page 110

New comment
 
Mathemat:

The regressions you construct are ideologically no different from NS: in fact they are the same [statistical] games with numbers with no intelligible internal content.

This is for you, not for me.

Tell us in popular language what this "probit model" is and why it is so good for econometricians. Just don't link to the wiki. Please tell us in your own language with examples.

I'll refrain for now. There is something I do not understand in it, since I cannot interpret the results.

Interest is understandable. In trend trading, reversals are important. We don't care about levels. Levels may be important in breakout strategies. I don't know I have never traded that way. I've picked ZZs with over 200 pips difference in reversals. I think I can take a third in a real trade.

 
A third would be a supergral. A very good trader is one who takes at least 10%.
 
alexeymosc:
Yeah, that's what I mean too. Where the MA changes direction, the regression model will not show this change, the accuracy of the model is a bit enough, I agree, but the direction of the MA change will still at first glance predict well, like 80% of hits, but still not enough for trading.
See my article where I showed that a much better MA - HP is not suitable for use in a TS. Only as part of it.
 
Farnsworth:

PS: since you're here, a word of advice - faa, don't fuck around, ZZs are not predicted - they are places where the price is practically non-existent, they are random, and completely random. No regression model is adequate for this case.

ZZ is a dependent variable, statistics of dependent variables are not involved (I haven't seen it). The statistics of the independent variable are interesting, and that is the quotient in my example.
 

What's there to understand))) the probit is from the same regression framework... just the input is continuous data and the output (what to look for) is Boolean (0...1)... that's why it was invented... not only continuous but + with Boolean data can be input... this will improve the result in some cases...

the mathematician is right and so are the others... everything is roughly the same... if a simple network is described by a formula, you can run it through the same way and it won't be a black box... it's just too much work...

 
Mathemat:
A third would be a supergral. A very good trader is one who takes at least 10%.

At the beginning of the topic I was predicting a level. There are prospects there, and then I wondered why the level? I distilled the level into TS and there again the question: enter or not to enter and it depends on the direction. I sat down and thought: what can I predict? I remembered the ZZ and broke through. But it is not the only one.

The breach itself is not what is shown in the pictures above. It is the probability of 1 and not 0. Or rather not the probability itself, but the threshold of probability cutoff (I have 0.5) that an independent variable will cross some value. I changed this threshold. Prediction picture does not change until cut-off 0.8, it does not work from above, what could it mean?

 
Vizard:

Mathematician is right about the networks and so are the others... all from the same line...

You have the same thing. NS in packages ( EViews doesn't have it, but others do) takes the place of smoothing, and this is only a small part of the problem and not the most important one to solve. In the case of NS, it's an art. If you take splines and wavelets, it's mathematics
 

And it wasn't clear to me how many bars ahead the reversal is predicted. And trade modelling can be done in excel, with a strategy in mind beforehand. Sometimes useful facts come out.

 
alexeymosc:

And it wasn't clear to me how many bars ahead the reversal is predicted. And trade modelling can be done in excel, with a strategy in mind beforehand. Sometimes useful facts come out.

Briukov wrote a whole book: EViews and Excel in parallel. It's a question of convenience for a particular person. EViews has everything picked out, and they write that there is no richer package. I think they are lying. There is Matlab, but the scheme needs to be worked out somewhere simpler.
 
alexeymosc:

And it wasn't clear to me how many bars forward the reversal is predicted to be.

How about predicting the number of bars before the reversal by taking ZZ?
1...103104105106107108109110111112113114115116117...139
New comment