Machine learning in trading: theory, models, practice and algo-trading - page 248

New comment
 

mytarmailS:

я

I see. That's how you see the perspective of the MoD. With most of the approaches voiced in the thread, very likely. But with a smile)).

I remember back in 2008 on MQL4 forum there was a topic dedicated to neural network training. There the approaches to learning were much deeper and broader, imho.

 
Yuriy Asaulenko:
You flip a coin, the probability of guessing is 0.5, but when you win, your income is not a bet, but 3-4 bets, and when you lose, the loss is only your bet. I don't understand what's not clear.
Uh-huh, only you bet 3-4 bets in advance or one bet, ok?
 
pantural:
Uh-huh, only you put 3-4 bets in advance or one, okay?
Let's call it a day. I already said everything. It's all written down.
 
Yuriy Asaulenko:
Enough is enough. I have already said everything. Everything is already written.

Wait a minute, wait a minute... I think I'm beginning to see what you're getting at!

That is, even with a random entry, if the take profit is 3-4 times higher than the stop loss, then we will have profit, everything is classic, "cut the losses, let the profits grow!" Cool!

Indeed, the market is not exactly roulette, now I don't know where we should use the MO here and why it is needed at all. Perhaps with the help of the MO you can calculate the optimal stop-loss and take-profit, but not really, when it is better to determine the profit/risk ratio.

Thanks for pointing out the obvious! I wish I was an Adept of Defense!

 
Velosipedists, I can't call you any other way. Guys, everything has already been invented before you :-) Everything new, well forgotten old. In 2007 they tried everything that you are just now trying to invent. And the people who were doing it back then were much older than me today so don't invent the wheel, and start reinventing the aero-bicycle will be more helpful. I've been thinking about totalizator on the stakaz with a predictor. I think it's worthwhile, except that I will have to collect data manually, and this is monotonous work, which stops me, but there is nothing to do my millions are waiting for me.....
 
pantural:

Wait a minute, wait a minute... I think I'm beginning to see what you're getting at!

That is, even with a random entry, if the take profit is 3-4 times higher than the stop loss, then we will have profit, everything is classic, "cut the losses, let the profits grow!" Cool!

Indeed, the market is not exactly roulette, now I don't know where we should use the MO here and why it is needed at all. Maybe, using MO we can calculate the optimal stop-loss and take-profit, but not really, it's not necessary when it's better to determine the profit/risk ratio.

Thanks for pointing out the obvious! If I had done it a little bit more I would have been made an Adept of IO!

I just have to warn you. Although you have begun to understand something, however at the approaches stated by you the probability of your plum in the transaction is not less than 0.8. It is better do not try.
 
Yuriy Asaulenko:
I just have to warn you. Although you have begun to understand something, but with the approaches you've outlined, the probability of you losing a trade is at least 0.8. It's better not to try.
Why? I enter by chance, the probability is the same that the market will go up or down 0.5. If I am lucky I take 3-4 lots, if I am not, I lose 1 lot, statistically 1-1.5 lots on each order, the main thing is not to run into a long losing streak, so Kolya Margin did not call, but statistically everything must work. Where does the 0.8 figure come from? If you know that you will lose 0.8, you should bet the opposite, and you will win.
 
pantural:
Where did the number 0.8 come from? If you know that you will lose 0.8, you should bet the opposite, and you will win.
If you know it's 0.8, you'll win. )) And this is without taking into account the spread. Draw a normal distribution function. At its zero (on X) you will enter the transaction. Now (on the X axis) mark the point where you want to make a profit and compare the area under the curve to the left and right of this point. Their ratio will give you the probability. And you'll see those ~0.8.)
 
Yuriy Asaulenko:
Now (on the X axis) mark the point where you want to make a profit and compare the areas under the curve to the left and right of this point.

And how exactly to relate the point X to the size of the take? Let's say I want to make a take at 100 pips, what would be the distance from 0 to point X in that case?

0.8 is the correct answer, I agree, but I defined it differently -

There's a rule that if you open a position in a random direction with small take and stop (take equals stop) - price will go to take and stop with equal probability, 50/50.
And if one of the distances (take and stop) is larger than the other, then with a large number of experiments price will reach it as many times less often as that distance is larger.
I.e. if a take is 4 times larger than a stop, at opening positions in a random direction the stop will trigger 4 times more often.

It turns out that in this case (random entry, take=4*stop) - the stop will trigger in 4 out of 5 cases, or 80%, the answer coincides with yours. Plus the spread, which will make things even worse.

* typo, corrected.
 
Dr.Trader:

And how exactly do I relate the X point to the size of the take? Let's say I want to make a take at a distance of 100 pips, what would be the distance from 0 to X in that case?

0.8 is the correct answer, I agree, but I defined it differently -

There is a rule that if you open a position in a random direction with small take and stop (take equals stop) - price will move to take and stop with equal probability, 50/50.
And if one of the distances (take and stop) is larger than the other, the price will reach it as many times less frequently as that distance is larger.
I.e. if a take is 4 times as big as a stop, at opening positions in the random direction the stop will trigger 4 times as often.

It turns out that in this case (random entry, stop=4*tack) - the stop will trigger in 4 cases out of 5, or 80%, the answer coincides with yours. Well, plus the spread, which makes everything even worse.

Mmmm... so you can't do it without predicting the future, can you?
1...241242243244245246247248249250251252253254255...3478
New comment
Reason: