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
This is where I don't know... Unfortunately, it is impossible to analyse the logic that is being built . That is, there is no answer to the question of why the neural network produced this or that result. So maybe they predict it in a completely different way than adaptive mash-ups. For example, there is such a thing as price patterns that is completely unexplored... After all, people successfully trade manually ! I know how to do it myself :) So, I can pass this knowledge to a robot, too.
Well, it's true, NS are very opaque. I dabbled in it about year and a half ago, but at semi-elementary level, trying to predict immediate levels. When I realised that this was my limit, I switched to GA. With a much smaller number of parameters the results were almost the same, even in linear models. It turns out that for predicting high-low on a few previous days, closing prices alone are quite sufficient: they have maximum information, and adding other information to closing prices hardly changes the result.
. It turns out that for predicting high-low on several previous days, only the closing prices are sufficient: they have maximum information, and adding other information to the closing prices hardly changes the result.
It makes sense in principle, because the opening price the next day should not be predicted knowing all previous prices, therefore it is the least informative and the most predictable. It is not clear why extreme prices are less informative than closing prices...
None, xeon, alas. But this phenomenon has something to do with prediction accuracy. The best predictor based on the closing price is the opening price (obviously), much worse the high, a bit worse the low, and very poorly the closing price.
It makes sense, because you don't need to forecast the opening price of the next day knowing all previous prices, so it is the least informative and the most predictable. It is not clear why the extreme prices are less informative than the closing ones...
Even H+L/2 does not give the same results as Close, and a bar is a very relative concept, especially for timeframes smaller than D1
however.... what complicated things are we talking about here in places....
probability theory.....
something i don't understand again....
price change is random, right?
then the change in that very price change is also random, right? (is it a derivative?)
so the change in the change in the change is also formed in exactly the same way. ...
Actually, we could go on like this for a long time, but I won't. enough is enough already. ...
Gentlemen mathematicians!
we have a lot of correlated quantities!
hasn't anyone worked out the formula yet?
P.S. even more I don't understand what randomness has to do with it, since the formula can be derived... but that's more of a question for linguists.
Ehhhhh, Tovaroved, you are right in places. But not always:
1. The original postulate about the randomness of price changes is the same as telling me about myself that I speak Russian. Randomness is different. The oscillation of the pendulum in a super-precision clock is also somewhat random, but that does not prevent us from knowing the time accurately enough.
2. A random process and its "derivative" are not necessarily correlated, but, moreover, they may even be independent! I won't give you an example, but it must be something to do with white noise or some other colored noise.
And even if the values are correlated, this does not make us feel much better: the Swissie, say, is largely correlated in movement with the euro - so what? And there is also an example of perfect correlation: the ask of the EUR is always greater than the bid, and at Alpari it is always exactly 3 points greater. Two random variables correlated 100%. And does that help to know the price a day ahead?
Ehhhhh, Tovaroved, you are right in places. But not always:
1. The original postulate about the randomness of price changes is the same as telling me about myself that I speak Russian. Randomness is different. The oscillation of the pendulum in a super-precision clock is also somewhat random, but that does not prevent us from knowing the time accurately enough.
2. A random process and its "derivative" are not necessarily correlated, but, moreover, they may even be independent! I won't give you an example, but it must be something to do with white noise or some other colored noise.
And even if the values are correlated, this does not make us feel much better: the Swissie, say, is largely correlated in movement with the euro - so what? And there is also an example of perfect correlation: the ask of the EUR is always greater than the bid, and at Alpari it is always exactly 3 points greater. Two random variables correlated 100%. And does that help to know the price a day ahead?
Why no one... The complete randomness of price behaviour (Brownian motion) is only one hypothesis about the market, and not a very good one at that. There are significant levels, Elliott waves etc.