Machine learning in trading: theory, models, practice and algo-trading - page 983
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
These are strategies written with feathers on water because there are no fundamental patterns.
The only thing that works is training on ticks of different brokers, like arbitrage. But it's resource-intensive to train on ticks.
If you haven't investigated these patterns, it doesn't mean they don't exist. Here, at least one regularity is working - the trends always come to the end and this is the reason to close the position, because you do not know what will happen next.
I am not interested in Forex, there is only one crook there - they lie and do not blush, they do everything for people not to earn - it makes them nervous, let the martinians work there - they are just as impudent fatalists as thieves.
If you haven't researched such patterns, it doesn't mean they don't exist. There is at least one pattern that works here - the trends always end and this is a reason to close a position, because you do not know what will happen next.
I'm not interested in Forex, there is only one crook there - they lie and do not blush, they do everything for people not to earn - this makes me nervous, no there let the traders work - they are just as impudent fatalists as thieves.
trends very often (and unexpectedly for many people) end when the deposit runs out :)
The real regularities are when there is a cause and an effect
Trends very often (and unexpectedly for many people) end when the deposit runs out :)
You can't argue with that, but there are many methods for surviving during unfavorable circumstances for martin.
The real patterns are when there is a cause and there is a consequence
So that is the reason - the fixation of profitable positions. And the reason of fixation - is exactly the search of the answer whether it is late to enter the trend or not, along with the question whether it is early to enter the trend or not.
How are profitable ones made and look like, if it is not a secret?
- I've never seen such intense profits on the real. If you manage to earn 30% a month, with the maximum allowable drawdown of 20% - that's already awesome. Moreover, the balance graph will constantly go up and down, tending gradually upwards. If the graph of profit constantly goes upwards as in this picture - then this is clearly a fitting, nothing good will come out on the real account.
- The out of sample test (OOS) cannot be used for model selection. Various ways of model selection based on it, such as training with an early stop, are also prohibited.
- The Equity graph must not have knees and any abrupt changes at the place where training starts/ends. The strategy on new data should behave without changes (at first), then the profit will slowly start to run out, and at the end it will turn into a loss on the spread. Often there is no need to re-learn the model, the life of the model should not run out in just a couple of hours after putting it in the real account.
- As for "how to do" - for me it's the right set of predictors, the right fitness function to estimate a model (I'm even wary of using R^2 for most models), selection of model parameters.
What does R2 have to do with anything? Are you teaching linear regression
Or Alesha told me.
Everything works, but only if the market cycle has not changed. This is if the predictors are from the same BP. I do not want to guess when the cycle is over. I have not tried other methods. What is all this OOS and p2 and all the rest - it's just to look at and see the obvious.
training to the right of the cr. line, then some time works in the past, and then already kolbasa
You can not take a forward before the learning period.
You can't take a forward before the training period.
How can you not, I took it, so you can.
it's a backward.
it's a backward.
No, a backtest is a backtest. ok, not a forward, but an OOS.
No, no one forbids it) you're welcome) but practice shows that such OOS can show the grail where it is not present at all.
No, a backtest is a backtest. ok, not a forward but an OOS.
No, no one forbids it ) you're welcome )) Only practice has shown that such OOS can show the grail where it is not present at all.
I know, I also took a forward approach - in fact nothing changes if the pattern remains the same. The data is all mixed before training, i.e. the order doesn't matter there, it will break anyway when the pattern ends
You can't take the forward before the training period.
Maxim doesn't want to admit that he has no model at all, which follows from the graph: training has nothing to do with testing. His model is not retrained - it has not learned anything at all.
The extremely tempting idea of replacing the target vector with some function that dynamically generates the target has yet to be confirmed.