Machine learning in trading: theory, models, practice and algo-trading - page 3686

New comment
 
Maxim Dmitrievsky #:

Yes, starting here and a couple of pages with examples. The models on synthetic data (at the end of the attempts, on the second page) seem to ensemble quite well.

Thanks, something to think about.
 
Aleksey Nikolayev #:
Thanks, something to think about.
I haven't yet thought of trying to generate on data from different time periods and then ensemble. Maybe that would make sense too.
 

Some motivation for creating regression models (pen test).

Periodic components in the initial series are searched for, then an attempt is made to predict on new data. The point of the TS comes down to trading from the boundaries of the predicted cycles.

Then checking in a toaster to see if these cycles are satisfying.

I have done a simple example on synthetic data so far. Tomorrow I will try it on real data.

I haven't seen such studies on the site, if you have - let me know.


 
Maxim Dmitrievsky #:

Some motivation for creating regression models (pen test).

Periodic components in the initial series are searched for, then an attempt is made to predict on new data. The meaning of the TS comes down to trading from the boundaries of the predicted cycles.

Then checking in the toaster to see if these cycles are satisfying.

I have done a simple example on synthetic data so far. Tomorrow I will try it on real data.

Haven't seen any studies like this on the site, if you have - let me know.


forecast package

Main functions of the forecast package :

  1. auto.arima() - automatic selection of ARIMA model parameters.
  2. ets () - exponential smoothing model (ETS), which is well suited for data with seasonality.
  3. stlf() - decomposition of a time series into trend, seasonality and residual with subsequent forecasting of each component separately.
  4. tbats() - complex model for complex seasonal patterns (e.g., multiple seasonality).

  6. The package is over 30 years old, quite popular for predicting economic data with seasonality. There is a book.
 
СанСаныч Фоменко #:

forecast package

Main functions of the forecast package :

  1. auto.arima() - automatic selection of ARIMA model parameters.
  2. ets () - exponential smoothing model (ETS), which is well suited for data with seasonality.
  3. stlf() - decomposition of time series into trend, seasonality and residual with subsequent forecasting of each component separately.
  4. tbats() - complex model for complex seasonal patterns (e.g., multiple seasonality).

  6. The package is over 30 years old, quite popular for predicting economic data with seasonality. There is a book.
I use Fourier or singular value decomposition. I haven't decided yet how it should look like in the end. How best to fit the cycles to the TS, and the fitting will obviously have to be done on real quotes.
 
Maxim Dmitrievsky #:

Haven't seen any studies like this on the site, if there have been any, let me know.

I've been doing it for a hell of a long time and posting the results here.
You laughed at me and called me a COSnikok, now you do the same, or maybe even more primitive and say that you have not met....
:)
 
Maxim Dmitrievsky #:
How to better fit the cycles to the TS, and the fitting will obviously have to be done on real quotes.
Use the formula of a conventional harmonic, only three parameters - amplitude, frequency, phase....

Each harmonic is a conventional cycle.
 
mytarmailS #:
I've been doing it for a hell of a long time, and I've posted the results here.
You laughed at me and called me a COSnikok, now you do the same, and maybe even more primitive and say that you have not met....
:)
I saw what you put into components, but I don't remember what you did with them :)

I just wanted to train a regression model, but I couldn't think of anything else to train it on :)

In general, this is quite a strange desire - maybe you should see a psychologist.
 
Maxim Dmitrievsky #:

Some motivation for creating regression models (pen test).

Periodic components in the initial series are searched for, then an attempt is made to predict on new data. The meaning of the TS comes down to trading from the boundaries of the predicted cycles.

Then checking in the toaster to see if these cycles are satisfying.

I have done a simple example on synthetic data so far. Tomorrow I will try it on real data.

Haven't seen any studies like this on the site, if you have - let me know.


I haven't done much . If the addiction is there, it's easy to get it out with anything.

https://www.mql5.com/ru/forum/86386/page3436#comment_52744612

https://www.mql5.com/ru/forum/86386/page3403#comment_52556668

https://www.mql5.com/ru/forum/86386/page1799#comment_16747717

https://www.mql5.com/ru/forum/137553/page5

ps probably haven't tried only detecting flat/trend changes yet. Flattening, attractors, etc.

 
Rorschach #:

I didn't do much. If the addiction is there, it's easy to get it out with anything.

https://www.mql5.com/ru/forum/86386/page3436#comment_52744612

https://www.mql5.com/ru/forum/86386/page3403#comment_52556668

https://www.mql5.com/ru/forum/86386/page1799#comment_16747717

https://www.mql5.com/ru/forum/137553/page5

ps probably haven't tried only detecting flat/trend change yet. Flattening, attractors, etc.

Thanks, I'll look at it later.
1...367936803681368236833684368536863687368836893690369136923693...3699
New comment