I would like to share the link - page 5
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Hmm.
It looks like your red row is a derivative of blue.
how you calculated it, can you show me the formula? Then it will become clear at once.
hpf(lambda = 13) dx hp1 @hp13_d Hodrick-Prescott filter with different lambda
hpf(lambda = 200) dx hp2 @hpn_d
hp1_d = hpn_d - hp13_d ' Difference between two noises
hp1_d_D = d(hp1_d) Noise increment
Hmm
It looks like your red row is a derivative of the blue one.
It is. I don't understand how that happened:
It is. I don't understand how that happened:
I don't understand why there are so many branches on econometrics...
...about the derived series, no matter how twisted you get... the non-stationarity will pop up everywhere... it just looks different...
If anything, I would rather look in this direction to see how high-frequency last period data can affect the accuracy of a regression model built on low-frequency data. Another variant - to try to use an irregular timeframe for regression: in application to Elder and in the presence of low-frequency data it makes sense, and there are suspicions that such a model will be at least an order of magnitude more accurate. And perhaps it will even be more profitable.)
(About non-uniform meshes - you can draw a distant analogy to the methods of numerical integration; those who know, know that choice of Gaussian meshes allows to raise order of approximation from n to 2*n-1 incomparison with interpolation methods with the same number of knots).
In addition to this - it would be very interesting to work with the function of regression errors. I still have not got around to it...
In short, the idea is to take weighted sum of squares instead of traditional functional that we minimize - sum of squares of errors, and to do weighting inversely proportional to the square root of time. Let's get the autoregressive difference equation to give the smaller error the closer we are to the predicted value of the series, and the law of decreasing average error will be consistent with the behaviour of the series (remember that the scatter of data deep into history grows as sqrt(t)).
The result should be smoother and more accurate than with a simple wizard. It's all so far at the level of intuition, but it rarely lets me down on such matters))
Can anyone take the trouble to check? I myself understand that it will take me half an hour or an hour to do it all, but Saturday...
In addition to that, it would be very interesting to work on the regression error functionality. I haven't got around to it...
In short, the idea is that instead of a traditional functional, which we minimize - the sum of squares of errors, we should take a weighted sum of squares, and weighting should be inversely proportional to the square root of the time. Let's get the autoregressive difference equation to give the smaller error the closer we are to the predicted value of the series, and the law of decreasing average error will be consistent with the behaviour of the series (remember that the scatter of data deep into history grows as sqrt(t)).
The result should be smoother and more accurate than a simple wizard. It's all so far at the level of intuition, but it rarely lets me down on such matters)).
Can anyone take the trouble to check? I myself understand that it will take me half an hour or an hour to do it all, but Saturday...
Are there formulas?
Although it can be deduced of course.
I don't understand why there are so many branches on econometrics...
about the derived series, no matter how twisted you get... there will be non-stationarity everywhere... it just looks different...
The problem is that in the above examples the non-stationarity has disappeared and it's not clear where it has gone.
between examples it has disappeared (it seems to be a hodrick or something else)... but otherwise the series seems to be non-stationary...
or do you mean (if they are the same) to take the earlier one and use it as a lead... It can't be like that on 1vr...there must be a mistake somewhere...
The problem is that in the above examples the unsteadiness has disappeared and it is not clear where it has gone.
I looked at the ssa examples (ssa - cloz) red 50, blue 10))... similarity is evident... the bug is in the hodrick... screw it...
What makes you think that the non-stationarity has disappeared?