Machine learning in trading: theory, models, practice and algo-trading - page 1010
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
ZZ as a targeting tool is a bad idea, its turning points (knees) are in places where you can hardly predict, there are fluctuations, and in the middle of anything and trend or not trend, neither suckers nor dolls probably did not suspect that there ZZ stated trend a posteriori.
In my opinion it is necessary to play with the first derivative (finite difference) from different smoothing indukes like Djuric, etc.
Almost a thousand pages about models, the reasoning of which would have been interesting 10-15 years ago. Today there are hundreds of them in R with zero cost to learn and use. Models that are difficult to use, such as deep neural networks, are extremely rare. Everything else is mastered very quickly.
But the target and its corresponding predictors is a problem that is solved by experience, intuition, and 90% luck. This needs to be discussed.
Since an overwhelming number of forumers still don't know how to measure predictive ability of predictors, I even offered my services in this matter with a proper run in rattle. There were only a few of them.
They are obvious, determined by the amount of profit you plan to make. For example, reversals of at least 100 pips in the hope that half can be rolled back. This value does not play a special role.
It is what is considered as a trend.
The determining factor is the set of predictors that will be relevant to the target. This is the third time I've written about this, but you don't seem to read my posts.
I just thought you went the way I did, but judging by your response - our paths diverge. I assume that indicators are used by people and they are customized by people, and where there is more money in the aggregate, there will be better performance of certain indicator settings.
I am using classic ZZ now, which is drawn when correcting by a certain number of bars (according to Donchian channel), I think it is more suitable than ZZ by points. I also liked the ZZ on the RSI channel.
But the target and its corresponding predictors is a problem that can be solved by experience, intuition, and 90% luck. This needs to be discussed.
Since the vast majority of forumers still do not know how to measure the predictive power of predictors, I even offered my services in this matter with an appropriate run in rattle. There were only a few people like that.
Very interesting, can you elaborate on the measurement of predictive ability?
And first of all, what does one measure?
From the series "funny pictures": the forest is 15 deep.
And what about outside of training, but like this)
Well, what's interesting...?
Are they predictors, maybe they were not. 1000 pages spelled out, and they couldn't even show such still lifes.
If anyone is interested, to the question of overtraining, here's a test with a forest of three deep.
From the series "funny pictures": the forest is 15 deep.
And what about outside of training, but like this)
Well, what's interesting...?
Are they predictors, maybe they were not. 1000 pages spelled out, and they couldn't even show such still lifes.
If anyone's interested, on the subject of overtraining, here's a test with a forest of three deep.
Maybe we should try to emphasize the support of an open position?
Maybe I should try to emphasize the support of an open position?
I have tried many things, the pictures show the best result.
I tried a lot of things, the pictures show the best results.
And what kind of predictors?
What type of predictors?
Mostly standard indicators.
From the series "funny pictures": the forest is 15 deep.
And what about outside of training, but like this)
Well, what's interesting...?
Are they predictors, maybe they were not. 1000 pages spelled out, and they couldn't even show such still lifes.
If anyone is interested, to the question of overtraining, here's a test with a forest of three deep.
Well that's a bad model you have, you can certainly do better, but it's still hard to sharps-radio to 2, even on tests, which says that in real life will be a shame. If I wanted to use it for real, I would have to look at the real time.
Well, this is a bad model, you can certainly better, but still sharps-ratio to 2 is difficult to pull, even on tests, which says that in real life will be generally unhappy. Well, at least that's the truth as it is, and not the turkey tricky "graals" that are random.
You'd better show me. I have training 2004-2014, data out of training 2015 to current date. Can you do that?