Machine learning in trading: theory, models, practice and algo-trading - page 3346
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
The peculiarity is that even without knowing the real spread, a part of deals falls off when you increase it artificially in the tester.
As much as the spread increases, the same amount of lowering of the matrix expectation. I did not understand the problem with the spread.
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
As much as the spread increases, as much as the matrix expectation decreases. I don't understand the spread problem.
If there is a model that works in greenhouse conditions. I would like to adapt it to any brokerage house with any spread. It would seem easier to put a bigger spread in the markup of trades and retrain it, but it does not help. It refuses to give a profit at a larger spread at the output.
So in greenhouse conditions low mat. expectation. Exactly where the alpha is.
Let's replace the word model with scalper. Let's say he's actually profitable at some quotes. Alpha is in low expectation.
We make the quotes worse. Train him to OOS. Because the alpha is destroyed. During the training there can be outwardly also thousands of deals. But there is no alpha - doomed.
ZY Why make profit on bad quotes, when everything is already in place to make profit on good ones?
So under greenhouse conditions, the expectation matrix is low. Exactly where the alpha is.
Let's replace the word model with scalper. Let's say he's actually profitable at some quotes. Alpha is in low expectation.
We make the quotes worse. Train him to OOS. Because the alpha is destroyed. During training, there can also be thousands of trades. But there is no alpha - doomed.
ZY Why make profit on bad quotes when everything is already in place to make profit on good ones?
Trading costs - slippage, liquidity, commission, swap. Spread is the value (I didn't write the difference on purpose) between bid/ask at the moment.
From midnight to 1 a.m. the minimum spread on EURGBP is dozens of times larger than the maximum spread before midnight.
And for some scalpers this is the most tasty hour of the day.
Trading costs - slippage, liquidity, commission, swap. Spread is the value (I didn't write the difference on purpose) between bid/ask at the moment.
From midnight to 1 a.m. the minimum spread on EURGBP is dozens of times greater than the maximum spread before midnight.
And for some scalpers this is the most tasty hour of the day.
We still trade the pattern - spread - other costs
None of my TS's anywhere in the logic (even indirectly) use the spread value. I am not the only one.
Why the raw data in the form of two bid/ask prices are converted to price/spread and then look for alpha in the price is a mystery to me.
Talking about spread, timeframes and Japanese candlesticks is about the same thing.
"Hello World!" in the field of understanding the source data - to write a script that will show the maximum possible profit on the historical interval.
If you don't have this, then it is unclear what you are doing.
None of my TCs use the spread value anywhere in the logic (even indirectly). I am not the only one to do so.
Why the initial data in the form of two bid/ask prices are converted to price/spread and then look for alpha in the price is a mystery to me.
Talking about spread, timeframes and Japanese candlesticks is about the same thing.