Remembering veterans: Box and Jenkins - page 11
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
But unfortunately you cannot check the quality of the model in packages like Statistica and Eviews, you can only check it on your own account
What is "model quality"?
P.S. I am convinced that if you study these 5 books thoroughly and don't trade without MM, your account will hardly suffer during this time.
This thread is just an educational reminder of decent people.
More seriously I tried to raise the issue of forecasting here. But seriously I was not supported in the thread.
It's not about the model, it's about the predictability on that model. In the above topic I cited the "correct" models with ARCH, but it all hung up.
But unfortunately you cannot check the quality of the model in packages like Statistica and Eviews, you can only check it on your own account
What is "model quality"?
And now closer to the terminology. By "PREVIEW" I mean only the series model you build in Evews, roughly speaking the equation(s) you get. By PREPARED price I mean the price value you predicted at least one step ahead for the series. I didn't understand this package thoroughly, but I realized that in automatic mode there you can only predict the average value and its "pseudo confidence" limits, and to predict the price there you still have to tinker with it by hand, and to do a thousand model tests you have to program in their language, which is not very good, but the data can be easily shoved into R or into Matlab, where you can check everything.
Now what is "model quality" - for me it's having money in my pocket all the time, I live once. So statistical models give huge variation and are totally unsuitable for trading, so after building a model for a series, you have to translate a bunch of maths into a HIS system, which is already far from maths. As an example, I would like to cite this link http://www.kosmofizika.ru/pdf/vawelet_vitiazev.pdf, where a simple example using wavelet analysis shows how a statistical model differs from an actually observed one. This method is very sobering from direct econometric models.
I have my own ironclad opinion:
1) Market prices are very difficult to predict, but probably possible ... :(
2) Predicting prices is futile in monetary terms and therefore unnecessary.
3) It makes sense to predict only a PRICE PRIZE, because only that can be money in your pocket, which you can spend on poker and beautiful girls.
4) You must and can only control YOUR RISKS.
5) You need your own TRADING SYSTEM, which earns you money.
And now closer to the terminology. By "PREVIEW" I mean only the row model that you have built in Evews, roughly speaking the equation(s) that you get. By PREPARED price I mean the price value you predicted at least one step ahead for the series. I haven't dug deep into this package, but I figured out that in automatic mode you can only predict the average value and its "pseudo-boundaries", and to predict the price there you still have to tinker with it by hand, and do a thousand model tests you have to program in their language, which is not very good, but the data can be easily shoved into R or Matlab, where you can check everything
This is not correct. A forecast is a button. You get a forecast with its estimate. What you put in the model is what you forecast: average means average, level means level, increment means increment, direction means direction.
So, statistical models give a huge variation and are totally unsuitable for trading,
In statistical models you are necessarily aware of the spread. You may not want to know about it and make absolutely accurate predictions, of which there are plenty on the forum and kodobase.
Where a simple example using wavelet analysis shows the difference between a statistical model and an actually observed one. This method is very sobering from using direct econometric models.
Wavelets are one of the smoothing methods in econometrics, which has a lot more. Smoothing is just a start, an idea and nothing more. The article shows that using a wavelet in this particular case gave a better result than AR - that's all. But apart from the wavelet and AR there is a lot more - one should choose what fits the particular problem in forex, rather than referring to sunspots.
And now closer to the terminology. By "PREVIEW" I mean only the row model that you have built in Evews, roughly speaking the equation(s) that you get. By PREPARED price I mean the price value you predicted at least one step ahead for the series. I haven't dug deep into this package, but I figured out that in automatic mode you can only predict the average value and its "pseudo-boundaries", and to predict the price there you still have to tinker with it by hand, and do a thousand model tests you have to program in their language, which is not very good, but the data can be easily shoved into R or Matlab, where you can check everything
This is not correct. A forecast is a button. You get a forecast with its estimate. What you put in the model is what you forecast: average means average, level means level, increment means increment, direction means direction.
So, statistical models give a huge variation and are totally unsuitable for trading,
In statistical models you are necessarily aware of the spread. You may not want to know about it and make perfectly accurate predictions, of which there are plenty on the forum and kodobase.
Where a simple example using wavelet analysis shows the difference between a statistical model and an actually observed one. This method is very sobering from using direct econometric models.
Wavelets are one of the smoothing methods in econometrics, which has a lot more. Smoothing is just a start, an idea and nothing more. The article shows that using a wavelet in this particular case gave a better result than AR - that's all. But apart from the wavelet and AR there is a lot more - one should choose what fits the particular problem in forex, rather than referring to sunspots.
Your question was about "model quality" - the author of the article answered it: "The main result of this analysis can be formulated as follows: the real series of solar activity is more deterministic than the implementation of its AR model...", and I was talking about how to check the "model quality" by means of wavelet analysis.
Your question was about "model quality" - the author of the article answered it: "The main result of this analysis can be formulated as follows: the real series of solar activity is more deterministic than the implementation of its AR model...", and I was talking about how to check the "model quality" by means of wavelet analysis.
The conventional wisdom on the forum is that it is impossible to apply the Box Jenkins models to the modern marketplace.
I enclose an article describing the application of ARIMA models to the Qatar money market.
For Junko.
The inclusion of trigonometric functions in the model is interesting.