Econometrics: one step ahead forecast - page 128
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
I don't understand, do you want the EW itself or a program for the EW?
EW is full on the web, updated from the website.
I use a program on EW to calculate it. It has its own language, simple, but still. I can give it to you.
Here is the calculation when the window is shifted by one bar.
You can see how the test results change. The rightmost columns are the number of lags in the equation. I take lambda=1 for HP. Maybe this is where the problem lies?
Here is the calculation when the window is shifted by one bar.
You can see how the test results change. The rightmost columns are the number of lags in the equation. I take lambda=1 for HP.
Here are the results on H4
The value of the coefficient is even worse. In addition to this, we cannot reject the hypothesis of a zero coefficient on a number of coefficients.
The residual from the equation had ARCH, which was modelled.
The descriptive statistic of the residual is killer - you can't believe anything.
It's not killer, it fully confirms what I wrote on the unreliability of the R-squared and the correlation of the primary series (I won't go back to the post). When are you going to start listening and not just writing :o)? The scale of price trajectory bias is very, very small relative to "absolute" values. These coefficients are "blind" to such insignificant fluctuations, they do not see obvious differences on such a scale, so it's all kind of good for you, expectations are statistically close, as it were, but it translates into other very dubious characteristics of your system.
Therefore, including going to increments (this is correct scaling within the scope of the oscillation scale).
It (EW) on small timeframes shows this, as if high (but useless) correlation values (it is linear in itself and always squared) and hence gives you the illusion.
But as soon as you get close to the "correct" scale for the model - you finally get an honest report that it's time .... expand your consciousness:o) And on twenty-four hours and months - it will be even worse :o/
The problem with errors is not a good or bad algorithm, but the essence of the currency pair, which (the essence) is not taken into account in any way.
No super-filters, decompositions and transformations will help. I hope this will help.
It is possible to predict the price movement without filters for any arbitrary number of bars ahead, but is this a goal?
Could you elaborate on what is the "essence" of a currency pair? What numerical characteristics reflect this "essence"?
Can you elaborate - what is the "essence" of the currency pair? What numerical characteristics reflect this "essence"?
The essence is that it is the ratio of one asset (currency) to another.
That's part of the problem. You are using a filter without understanding how to design it properly and you are also predicting it. The question is very simple - what are you filtering or what do you want to filter? If you need to "smooth" what is it and how do you explain your understanding of "smoothness" to HP?
Regarding HP. Looking at.
The essence is the ratio of one asset (currency) to another.
What numerical characteristics reflect this "essence"? How can this "essence" be reflected in the algorithm/model?
Farnsworth:
Therefore, incl. going to the increments
Please see the EURUSD increments in 1/DX increments
The result is even worse
Exactly.
Accordingly, each of these assets has its own frequency phase and amplitude. I stress there are two (2) components. The numerator and the denominator.
Currency is a division of one into the other. It is impossible to simply predict the division of one into the other, there will almost always (66% of the time) be an error.
And how hard is it to forecast? Predicting currency indices? And adding up the forecast?