Econometrics: one step ahead forecast - page 84
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
Oh, man... they made up these words... ...that you can't pronounce... but it's incomprehensible and looks clever!!!
:))))))
Well I was talking about ARCH. Someone is definitely inadequate.
Autoregressive conditional heteroskedasticity
Or
Autoregressive conditional heteroskedasticity
No, it won't. It was just an example. In fact, you need to start with at least the simplest AR and MA types...
It's self-written, but it's correct, it's been tested.
No it doesn't, a quote is not a stationary process at all and ACF will change very much over time and from sample length.(That's one of the things the author of this thread doesn't understand).
Don't bother, everything is calculated correctly.
I know clearly that it's non-stationary and that it depends on the length and I also know that with time the ACF changes too.
Here is another question. Let me try to describe it(https://ru.wikipedia.org/wiki/Автокорреляционная_функция)
1. Take 500 samples and build the ACF. ("...Autocorrelation function plot can be obtained by plotting along the ordinate axis the correlation coefficient of two functions (base and function shifted by τ) and along the abscissa axis the value of τ...")
2. When your shift τ exceeds 499 samples - ACF should be zero. There is nothing left to compare. The sample is already finished. The original and its copy shifted by t do not overlap in time they have nothing else in common (and can't be)...
3. I have used three methods to calculate (and check their correctness).
- built in matcad.
- using Fourier
- and by formula, brought it (the formula) into codbase.
All three results came together.
Z.I. I basically don't care how you have it counted there, it just caught my eye that something is wrong with the graph, the ACF doesn't look right (or it's just cut off, only the first 500 counts are shown, and they were actually taken there more for calculations...)
Correlation is the essence of trend. Farnsworth has suspiciously straight function lines that do not reflect changes in trend.
Here's the graph.
There are 270 observations here. If we draw the ACF, we can see a wave that corresponds to the change in trend. The 270 bar does not fit in the screen, so I will bring the part where the trend changes from 40 to 90. This is what it looks like:
We can see the ACF wave matching the visual trend. The instructions in EViews warn us to be careful when specifying the number of lags on which we count the ACF. This is a known problem for anyone familiar with TA, as it is possible to draw many trends in one area.
.... For anyone familiar with TA it is a known problem as you can draw many trends in one area.
This is only until you have defined the rules of how to draw them. And when you do, there will only be one left.
This is the biggest sticking point of any attempt to predict the future based on historical data: which hindsight to use? This situation is similar to when we try to define south-north and west-east from within a sphere.
I prefer the term "smoothing" to "trend". In any case, smoothing is analytical and easily extrapolated