Machine learning in trading: theory, models, practice and algo-trading - page 2759
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
My head is wooden, I'll do an example later
You overestimate the importance of "out of sample" results.
You can't trust a penny of it.
"Out-of-sample" might have something to offer for identifying model overtraining if there is too big a difference (over 20% probably) between the training sample and the out-of-sample. So far, the only evidence I know of is a small, no more than 20% variation in the sd of the predictive ability of the predictor.
You overestimate the importance of "out of sample" results.
You can't trust a penny of it.
"Out-of-sample" may be something to identify overtraining of the model if there is too big difference (over 20% probably) between the training sample and out-of-sample. So far, the only evidence I know of is a small, no more than 20% variation in the sd of the predictive ability of the predictor.
It's more of a time-unfixed window. And as we fed 10 columns, we'll still have 10 columns.
I tried feeding a dozen ZZ columns (got 50%50 as usual). The time of their formation is always different and the last knee was formed from 1 to several bars ago. You could say it's a conversion of regular bars to their own. In my case the zigzag knee was forming a bar. A year ago Prado was recalled here, he suggested to rebuild bars not by time, but by traded volume, for example, by 100 lots.
Examples of raw prices, time, "volume" and "dollar" candles
Spit on predictors and grind. Maybe with the teacher.
It is necessary to fantasise and clave predictors by tens, hundreds, and then select them by their influence, predictive ability, informative connection with the teacher.
Everything you need has been codified before you.
-- and I never said that I lacked something in the available libs.... the answer was meant for someone who expects more from the library than its crooked interpretation by some even non-commenter... you, too, would do better to address your comments..... (and not to my reply, which was not meant for you).
-- and I didn't say that I lacked anything in the available libs.... the answer was meant for someone who expects more from the library than its crooked interpretation by some even non-commenter... you, too, would do better to address your comments..... (and not to my reply, which was not meant for you).
Not sure why, but I apologise.
Some completely unintelligible offence generated by my comment.
it shouldn't have... but that's your data analysis, you know best.
Resampling is done to remove outliers, to Gaussianise the sample.
I know what it's for,
I don't know why you recommend it in BP analysis,
I haven't seen where and when and whether they used it themselves and I don't see the point and logic and validity of it in the BP analysis....
2. Extremely interesting, especially on entropy. I would like to see the result. Correlation is for stationary series, we can forget it.
correlation reveals dependencies (if properly studied)...based on the target's dependencies on future controlled, known, easily predictable (if any) changes in the factor(s) -- and a forecast/prediction is made.
or
the prediction is made based on identified patterns of change in the target over time
== that's the way it's always been done
== if you reject correlations, that's your right, but not being able to use them for their intended purpose in analyses doesn't make them useless... (I'm beyond the constant bickering about why you don't like the word "correlation" and what non-correlational but with some psychic "predictive power"you have invented for a Nobel Prize) - just skewing the natural conceptual apparatus in forecasting, which was formed several centuries before your appearance with your psychic predictive abilities -- skews the very objectivity of forecasting (which does not forbid to rely on detectable dependencies in certain circumstances).
P.S. you have already been hinted at by the way
regression already allows you to compare the significance of predictors. Further steps are based on the results of the analysis.
but for some reason you have reduced everything to ma-shkas.....
Spit on the predictors and rub it in. Maybe with the teacher.
It is necessary to fantasise and clave predictors by tens, hundreds, and then select them according to their influence, predictive ability, informative connection with the teacher.
What to make them out of? From fifty indicators, with all kinds of settings - millions of variants will come out.
And they all have to be tested for each target, and hundreds or thousands of them can also be invented. Total billions of tests.
There is one problem - almost all of them are based on MAs, i.e. with lag.
ZZ to input, for example, without lag: I tested it, I didn't like it.
1. correlation reveals dependencies (if properly studied)...based on the target's dependencies on future controllable, known, easily predictable (if any) changes in the factor(s) -- and a forecast/prediction is made
either
the forecast is made based on identified patterns of change in the target over time
== this has always been the case
== if you reject correlations, that's your right, but not being able to use them in analyses as intended doesn't make them useless... (I'm outside of the constant bickering about why you don't like the word "correlation" and what's so uncorrelated, but with some sort of
2."psychic "predictive ability" you invented for a Nobel Prize) -- just skewing the natural conceptual apparatus in forecasting, developed centuries before you came along with your psychic predictive abilities -- skews the very objectivity of forecasting (which does not prohibit reliance on detectable dependencies in certain circumstances)
1. Are you aware of correlations between actual and nominal variables?
2- This is not my invention. Not even the word "predictive ability". VLADIMIR PERERVENKO was not lazy and posted a series of extremely qualified articles that graphically show the meaning of my term "predictive ability" and list the corresponding packages, and only a part of such packages. You can start here.
Here is the graphical meaning of the words "predictive ability"
Or in this form
Correlations between predictors and teacher do NOT smell here.
PS.
Take less offence, less aplomb and learn more from other people. and you will be digitally happy.
What to make them out of? From fifty indicators with all kinds of settings - millions of variants will come out.
And they all need to be tested for each target, and hundreds or thousands of them can also be created. Total billions of tests.
There is one problem - almost all of them are based on MAs, i.e. with lag.
ZZ to input, for example, without lag: I tested it, I didn't like it.
That's the problem!
That's why I'm writing that years have been spent on this thread mostly on nonsense. If you manage to find predictors, you will have a trading system. If you don't manage to find predictors, it's not your destiny.
PS.
As a teacher, take the price increment, or rather a sign. VLADIMIR PERERVENKO has given very good predictors in his articles.