Machine learning in trading: theory, models, practice and algo-trading - page 1150
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
He doesn't know shit, in my opinion...
Shit, don't speak so loudly, or my authority will fall))
Don't say it so loud, or my authority will fall))
My friend, don't be offended - I realized a long time ago that NS by itself are not able to get a penny out of the market. But, to determine the area with a positive correlation - can it do that at least? If you think of something, let me know. We'll attach it to my TS and run off to make money. I can't do it without help at all....
I've made some predictors on pivot points, so far only on the B+, but tomorrow there will be some on the B+ as well...
the stops are just for visualization, they're not cp's.
Predictors are already working in profit by themselves. I have a suggestion, what if I send you the data with the format open | high | low | close | vol | pred1 |pred2 | pred3.....
You generate your predictors from the OHLC data for filtering and train the network or whatever, maybe something good will come out.
What do you think?
I might give it a try, if you don't mind, attached a row for a test
I could try, if you do not mind, attached a row for the test
I'm not against it, but for what purpose will teach? If the binary classification is not interesting, I know how to do it... Interesting to teach a fitness function, those search for some kind of minimum function or well, the maximum...
The data is shortened, because too much and hopefully made no mistakes calculations because I wrote the code quickly
For 2 years ... it doesn't even run in tester :)))
You mean within the trained sample data or out of sample(OOS)?
Within sample data...the curve is very nice :)))
In OOS, it is in net loss((
So I am running direct forward testing LIVE in MT5.
ah, ok )
I do not mind, but on what target will teach? If binary classification it is not interesting, I know how to do it... Interesting to train on fitness functions those search for a certain minimum function or maximum...
The data is shortened, because too much and hopefully made no mistakes calculations because I wrote the code quickly
Ok thanks, I'll play around with it one day.
About the target... If you have features that signal a reversal (I guess), then there are more ways to "shoot yourself in the foot" (in algotrader's way), if you're not careful and emotional about "good" results, because accuracy (logolos, etc.) of pivot point classification\regression itself can easily mislead, It's like with the proportion of profit/loss trades, which says little about anything, the "pivot" itself doesn't contain information about how much money you can lose on it, which means that the quality of the classification/regression is difficult to interpret, This is why first of all it is necessary to translate "reversals" into "directions" and then quantify them in terms of profit.
I think to start with your discharged reversal signals, something continuous, trending, and using them to figure out the directions of future returnees for a different period, like that ...
Okay thank you, I'll play around the other day.
About the target... In general, if you have features that signal a reversal (I guess), then there are more ways to "shoot yourself in the foot" (in algotrader's way), if you are not attentive and emotional about "good" results, so for example the accuracy (logolos, etc.) of classification/regression of reversal points itself can easily mislead, It's like with the proportion of profit/loss trades, which says little about anything, the "pivot" itself doesn't contain information about how much money you can lose on it, which means that the quality of the classification/regression is difficult to interpret, This is why first of all it is necessary to translate "reversals" into "directions" and then quantify them in terms of profit.
I think to start with your discharged reversal signals as a continuous, trend one, and use them to plot the directions of future returnees for a different period, like that ...
Well, try it...
By the way, add more predictors, I think candlestick patterns will be good for filtering, you can also enter not immediately by the signal, but through a candle, for example, with some kind of confirmation
And seriously, the "no-annoyance" and trades has no meaning as a metric at all, it cannot be optimized, because strategies will differ every time by the number of trades, so for example a strategy generating 1000 trades will have three times less sharpe than strategy with 100 trades, with the same profit and max drawdown.
Despite the commonality of your approach to calculating sharpening for assets and portfolios I am not ready to transfer it to individual TS. I believe that the TS is in no way a portfolio, but only a possible part of it.
It's not even about the shuffle itself, but about the imposed approach, when instead of my TS I have to consider something unknown, where many trades can be artificially glued into one and non-existent null trades can be added. And it's only because "it has to be that way".
To me, the sharpe is a characteristic of the distribution of profits of deals, which shows the statistical significance of the difference between the average profit and zero. In the case where the number of trades in the TS can vary so much, the sharpe will have to be modified. For this purpose we should subtract from it a value of k/sqrt(n) type, where n is a number of deals. The point is that with increasing number of trades, the confidence interval for the expectation is narrowed and this can compensate to some extent for the decrease of the usual sharpening with increasing number of trades. If the number of trades does not jump that much, then this correction does not affect the optimization and therefore the standard shard can be used.