Machine learning in trading: theory, models, practice and algo-trading - page 1777
![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
Somewhere heard such a clever idea, something like - if there are signs with at least some statistical significance with even the most minimal combining them together you can get accuracy close to 100%.
I decided to check it out...
I made a synthetic date with binary target, to each value of target I connected a chip with some hit probability.
I made 10 of these chips with probabilities of 51:49 for one target and 49:51 for another.
I trained Forest.
got new data.
Accuracy : 0.5145
there are not 10 features, but 100
Accuracy : 0.534
I got 1000 signs.
Accuracy : 0.558
The conclusion is: we must improve the quality of the signs, we can't go far with quantity...
let's try to increase the probability by, say, 55:45.
10 traits give...
Accuracy : 0.6055
100 traits give
Accuracy : 0.7985
let's try another increase in the probability : 60:40
10 traits
Accuracy : 0.729
100 signs
Accuracy : 0.968
So it turns out that to live in Sochi on each candle must have 100 rules/features/AMO that give 60% correct answers... and still have to be different at the same time... I wonder if it is possible to do that?
Somewhere heard such a clever idea, something like - if there are signs with at least some statistical significance with even the most minimal combining them together you can get accuracy close to 100%.
I decided to check it...
Correlation of EURCAD on daily data with pairs:
Coefficient of determination of the simplestlinear regression EURCAD = a*AUDCHF + b*CADCHF + c*CHFJPY + d*EURCHF + k
R^2 = 0.99622555
EURCAD correlation on daily data with pairs
Correlation is not a prediction but a measure. Or am I missing the point?
correlation is not a prediction, but a measure. Or am I missing the point?
Correlation shows the statistical significance of each variable - it is low.
Together they form a model that explains the dynamics of the dependent variable by 99.6%
The correlation shows the statistical significance of each variable - it is low.
Together they form a model that explains the dynamics of the dependent variable by 99.6%.
well yes, but it explains not predicts, correlation is just a measure of the relationship between variables, what is the conclusion of your thought? i still don't get it (
If we are looking for a cross correlation between pairsWell yes, but it explains not predicts, correlation is just a measure of the relationship between variables, what is the conclusion of your thought? I do not understand (
"I've heard this clever idea somewhere, something like - if there are signswith any statistical significance, even with the most minimal one, connecting them together you can get accuracy close to 100%" (c)
Correlation shows the statistical significance of the independent variables for predicting the dependent in a linear regression model.
"Somewhere I heard such a clever idea, something like - if there are signs with at least some statistical significance, even with the most minimal one by combining them together you can get accuracy close to 100%" (c)
I meant characteristics that somehow can predict, and not just a correlation
I meant signs that can predict in some way, not just correlation.
And the ability to predict is determined by how?
I meant attributes that can somehow predict, not just correlations.
Here we have a dependent variable and a set of possible independent variables.
How is "predictive ability" determined?
By dumbly shoving everything in the world into the model?
Well... Nice and believable. I would like to see the balance of the trade and a chart with entries.
You never told me how to trade on it - that's why I don't know what kind of TS I should make up.
As far as I understand it is an ensemble of 10 models. How do the models differ from each other?
No, it's just 10 models to see the spread, the only difference is the seed, i.e. the random value to start learning (used in evaluating splits and selecting them).