Econometrics: one step ahead forecast - page 21
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
You've got me wrong... It's the interpretation that counts, isn't it?
The link indicates the limitations of the application.
My application of the filter I have explained many times, nothing to add, or clarify "interpretation is important"
The link indicates the limitations of the application.
My application of the filter I have explained many times, nothing to add, or clarify "interpretation is important"
already in its formulation KP uses "future" predictive values,
.
but okay... let's leave it at that...
is already using "future" predictive values in its formulation of KP,
.
but okay... that's where we stop...
We take the original quotient and extract the deterministic component from it. In my case I do it with HP. Then we model the residual (noise).
Who says that the deterministic component exists in quotes? And who says that HP or any other filter (MA for example) highlights the deterministic component in quotes? For example if I generate my quote as a random rambling process using random number generators and then apply HP, do you think it will show 0 at all parts of my quote? Like other filters (like MA for example) HP filters lower frequencies, which for some reason you call a deterministic component. And upper frequencies are noise. In quotes they are all noise! Understand that NR is suitable when this deterministic component is really present in the form of economic cycles. And you are trying to separate out the deterministic component in daily cutlet-to-sausage exchange rates. What deterministic components are there?
An anecdote about the exchange rate: Gorbachev arrives at a factory and talks to a locksmith.
Do you drink?
Yes.
And if we raise the price of vodka, will you drink less?
No, the same.
Well, how can that be? Will vodka go up in price?
Do you see this piece here? They used to give a bottle for it at the market and they'll keep giving it to you!
One bottle for one piece is our deterministic component. All deviations from it are noise.
Vladimir, did you build the grid in MQL4? Or another language?
I mean that MT4 Optimizer is too limiting (GA does not give more than 10 thousand variants), and I don't know any other languages.
How do you cope with this problem?
Vladimir, did you build the grid in MQL4? Or another language?
I mean that MT4 Optimizer is too limiting (GA does not give more than 10 thousand variants), and I don't know any other languages.
How do you cope with this problem?
Optimizing network weights with a tester is very difficult, 10-20 is still possible. The problem is that MQL4 does not allow to use arrays as external variables. In my opinion, the same is true for MQL5. So a large number of weights should be optimized within the code, or outside of it, in the attached DLL, as here
https://www.mql5.com/ru/code/8976
The optimization speed is much higher in C++ DLL even in comparison with MQL5. So I recommend MS Visual Studio C++. It is the only place I write all my developments - it is faster and has more features. Many C++ features are missing in MQL5, so you have to develop your own classes and structures even for such simple things as dynamic array sizing by all its indexes (and not by one like in ArrayResize).
Optimizing network weights with a tester is very difficult, 10-20 is still possible. The problem is that MQL4 does not allow to use arrays as external variables. In my opinion, neither does MQL5. So a large number of weights should be optimized within the code, or outside in the attached DLL, as here
https://www.mql5.com/ru/code/8976
Optimisation speed is much higher in C++ DLL even compared to MQL5. So I recommend MS Visual Studio C++. It is the only place I write all my developments - it is faster and has more features. Many C++ features are missing in MQL5, so you have to develop your own classes and structures even for such simple things as dynamic array sizing by all its indices (and not by one as in ArrayResize).
Who says that a deterministic component exists in the quotes? And who says that HP or any other filter (MA for example) highlights the deterministic component in quotes? For example if I generate my quote as a random rambling process using random number generators and then apply HP, do you think it will show 0 at all parts of my quote? Like other filters (like MA for example) HP filters lower frequencies, which for some reason you call a deterministic component. And upper frequencies are noise. In quotes they are all noise! Understand that NR is suitable when this deterministic component is really present in the form of economic cycles. And you are trying to separate out the deterministic component in daily cutlet-to-sausage exchange rates. What deterministic components are there?
The market has: trend, seasonality, cyclicality and noise and outliers (catastrophic news). Read books on the psychology of the market, for example in the beginning of the topic C-4 gave a link to a very interesting book.
To be precise, I am trying to remove autocorrelations.
Here is the original quote. I can see directional movements on it.
As long as directional movements (trends) are present in the quotient, statistical processing is impossible.
Trends are detected using autocorrelation. Here is the analysis:
The right bar is the probability of no correlation. More precisely: we strictly reject the hypothesis that there is no correlation between observations.
Apply the HP filter.
Here is the ACF for the residual:
We see that there remains a correlation between lags 3 to 13.
This is not the first time I repeat all this. At least read something before you post.
The previous prediction turned out to be correct.
Making a new forecast. The result is in the table.
faa1947:
The article is accompanied by MQL4 and EViews files which allow anyone to do what I did.
If I had known, I would have lived in Sochi....
As for predictions about the future, it's very hard to judge. Although I had a system based on opening or closing time of banks, what one bank gives to another one to one rate. It agrees of course, but time is a bit variable plus or minus 5-30 minutes. Here is the problem