Machine learning in trading: theory, models, practice and algo-trading - page 3347
![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
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
"Hello World!" in the area of understanding raw data - to write a script that will show the maximum possible profit on a historical interval.
If you don't have it, then it is unclear what you are doing.
NHITS and the lightGBM also has lower RMS then the TimeGPT in daily and hourly data. https://valeman.medium.com/what-truly-works-in-time-series-forecasting-the-results-from-nixtlas-mega-study-78eda5133622
Have you tried Conformal Prediction ?
https://valeman.medium.com/how-to-predict-full-probability-distribution-using-machine-learning-conformal-predictive-f8f4d805e420
https://github.com/valeman/awesome-conformal-prediction#papers-time-series
So it is, on mashki achieved a uniform loss on each trade equal to the spread)))))) At zero spread even profit))
Maxim Dmitrievsky#:
The spread has nothing to do with it.
We just take the statistics for H1
and stupidly see at what value of price increment your "profitable" forecasts become unprofitable, i.e. at 10 pips of 4 digits only 25% of market movements become potentially profitable. This is with an error-free forecast!
Spread has nothing to do with it.
We just take statistics for H1
and stupidly see at what value of price increment your "profitable" forecasts turn into unprofitable ones, i.e. at 10 pips of 4 digits only 25% of market movements become potentially profitable. This is with an error-free forecast!
You do not understand what I am writing about
When marking with the spread, 0% of trades are unprofitable. And it does not matter whether it is calculated by average price + spread, or by Saber ticks on bid and ask separately. On average, the result is comparable.
you can calculate by ticks later, if you are a fierce scalper and work in 1-2 dts, I do not particularly like such TSs
Draw a diagram-distribution of deals, where on the horizontal line is the profit of closed positions, on the vertical line is the number of closed positions.
For narrow spread and wide spread.
You don't understand what I'm writing about
When marking with the spread, 0% of trades are unprofitable. And it does not matter whether it is calculated by average price + spread, or by Saber ticks on bid and ask separately. On average, the result is comparable.
you can calculate by ticks later, if you are a fierce scalper and work in 1-2 dts, I do not particularly like such TSs
My markup is price increment.
Take your markup and look at quantile What profit your markup is designed for? Compare it with the statistics.
My markup is the price increment.
On the same list.
Forum on trading, automated trading systems and testing trading strategies
Machine learning in trading: theory, models, practice and algo-trading
fxsaber, 2023.12.10 17:57
Talking about spread, timeframes and Japanese candlesticks is about the same thing.
My markup is the price increment.
Take your markup and look at the quantile. How much profit is your markup designed to make? Compare it to the statistics.
No, there's no problem with that. It doesn't matter what the profit margin is. What matters is the classification error. It grows when spread is added to the training or remains the same.
But the model does not start working better when the spread is taken into account in the markup, it does not give a profit, but without the spread it works the same way as if it was trained without it. That is why I put the spread, conditionally, to the classification error. That is, the response of the model does not allow you to beat it.
Taking the spread into account in the markup means the length of trades that exceed it. That is, I make the trades longer, then train them, and the result of testing on the increased spread is almost the same as the result of another model trained on shorter trades.
It turns out to be a rather unambiguous conclusion that on my signs, let's say, MO cannot beat the spread.
But sometimes it can, with certain machinations related to kozul. That is, if there is some stat. indicator of deduced "reliability" of signals, then they work also when the spread increases.
Nah, there's no problem with that. It doesn't matter what the profit margin is. What matters is the classification error. It grows when adding spread to training or remains the same.
But the model does not start working better when the spread is taken into account in the markup, it does not give a profit, but without the spread it works the same way as if it was trained without it. That's why I put the spread, conditionally, to the classification error. That is, the response of the model does not allow you to beat it.
Taking the spread into account in the markup means the length of trades that exceed it. That is, I make trades longer, then train them, and the result of testing on the increased spread is almost the same as the result of another model trained on shorter trades.
It turns out to be a rather unambiguous conclusion that on my signs, let's say, MO cannot beat the spread.
But sometimes it can, with certain machinations related to kozul. That is, if there is some stat. indicator of deduced "reliability" of signals, then they work also when the spread increases.
Itdoesn't matter what profit is calculated for.What matters is theclassification error.
Because of this approach you "correctly" classify potentially losing trades. In reality, the situation is much worse not only because of the spread. In a real EA to achieve from "correct" classification to a profitable system remains a problem, as it is not surprising.