Bayesian regression - Has anyone made an EA using this algorithm? - page 34

 
Whether or not the price depends on time is a philosophical question. Like whether the chicken or the egg came first (not that, but in the same style).
 
Dmitry Fedoseev:
Whether or not the price depends on time is a philosophical question. Like whether the chicken or the egg came first (not about that, but in the same vein).
Too late, I already posted the model
 
Dmitry Fedoseev:

Vasily, I'm sorry. But this Normality of Distribution stuff is annoying. Excuse me for the immodest question, are you zombified somewhere, you are like a copy of the normality of the distribution? Only one of you has managed to make sense, unlike all the demagogues.

Dimitri, you misunderstand me. I'm in no way sticking my neck out for normal distribution. I don't care if the market is normal or not. It's just that their models require it as a necessary but not sufficient condition. That condition is not met, which was one of the reasons why these models cannot be used.
 
Vasiliy Sokolov:
Dimitri, you misunderstand me. I am by no means sticking my horn in the normal distribution. I don't care if the market is normal or not. It's just that their models require it as a necessary but not sufficient condition. That condition is not met, which was one of the reasons why these models cannot be used.

THEIR models require the normality of the model's residuals distribution.

Why would you lie?

 
Дмитрий:

To decipher or not to decipher?

The independent variable is time.

Dependent is EURUSD, D1.

R^2 = 0.49708851

R = 0.70504504

R^2 is nothing at all. Dimitri chose either a random plot or a weak trend. On a random walk, the validity of the results can be shown to be much higher, but only on a certain plot. On a different plot, the results will be quite different.

Dimitri, let's demonstrate us another experiment: divide the same EURUSD into a sufficient number of N segments. Conduct a test on each of them and show us the convergence of the approximation quality estimate to some significant value.

 
Vasiliy Sokolov:

R^2 is about nothing at all. Dimitri chose either a random plot or a weak trend. On a random walk the validity of the results can be shown to be much higher, but only on a certain plot. On a different plot, the results will be completely different.

Dimitri, let's show us another experiment: divide the same EURUSD into enough N segments. Run a test on each of them and show us the convergence of the approximation quality estimate to some significant value.

)))))

This is 8 years of data!!! 1700 observations!

Fucking "random site" .......

 
If R^2 is nothing, what can we talk about here at all....
 
Дмитрий:

)))))

This is 8 years' worth of data!!! 1700 observations!

Fucking "random site" .......

Once again, your R2 is very low. So far you have shown that the market does not fit your chosen model. If you've "proved" anything there, it's only for yourself. Well done! I envy you, for naivety is the easiest way to happiness:)
 
Vasiliy Sokolov:
Once again, your R2 is very low. So far you have shown that the market does not fit your chosen model. If you've "proved" anything there, it's only for yourself. Well done! I envy you, because naivety is the easiest way to happiness:)

)))))

Is R^2 "very low" already?

Is there a correlation?

 
Vasiliy Sokolov:
Once again, your R2 is very low. So far you've shown that the market doesn't fit your chosen model. If you've "proved" something there, it's only for yourself. Well done! I envy you, because naivety is the easiest way to happiness:)

OK, you've got more R^2 on you:

JPY Multiple R = .80447629 F = 3070.657

R?= .64718209 df = 1.1674

No. of cases: 1676 adjusted R?= .64697133 p = 0.000000

Standard error of estimate: 8.974991583

Intercept: -633.7672045 Std.Error: 13.16495 t( 1674) = -48.14 p = 0.0000