Machine learning in trading: theory, models, practice and algo-trading - page 649
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
Wonderful discussion, like a fig in your pocket.
And how did you add these two pairs, what is the remainder?
Well, via a neural network, I've already written about it a bunch of times here
You offered me VAR and so on... Why do we need linear regression if we have NS?
If i've got the answer, i don't know what to do with it, i've got a negative number - preferably to predict and trade, or to average. If someone has an experience in this field ... it seems that no :) the normal pair trading, buy one and sell another.
So the property of noise is that it does not need to be predicted) it deviates from the average - so it will soon return... but if there is an arch effect, it is not certain that it will return immediately... and the variance is variable, but okay, we'll figure it out)
I have answered my own question. We take a (simple) wand, shift it by half a period back (phase delay of simple wand is half a period), and use noisy quotes to feed the NS for predicting a new wand value by at least one bar (it can even be the current one, since we have a prospective wand). Using the caterpillar method we fill the missing half a period, and there is a convergence to the mean, regardless of whether the average is moving down or not. The market will surely return to the future car.
Well through a neural network, I've already written about it a bunch of times here
You suggested me to use VAR and so on... Why do we need linear regression if we have NS?
If i've got the answer, i don't know what to do with it, i've got a negative number - preferably to predict and trade, or to average. Probably somebody has an experience, but probably not :) the normal pair trading, buying one and selling another.
I have nothing to say about the NS.
I think that cointegration is a quite large direction in the models and very well developed. The reason why there is no NS is that in these models everything must be proved with tests, and in your case it worked out like that.
Whether it is possible to use your idea - I do not know, I am interested only in TS, and which I prove their future properties at the stage of training.
PS.
It was a long time ago, but from memory your graph is very strange. It should be like this: if there is a trend then the graph below is either above or below the axis and it goes back and forth.
You have answered your own question. We take a (simple) waveform, shift it half a period back (phase delay of a simple waveform half a period), and use noisy quotients to feed the NS for predicting a new waveform value by at least one bar. Using the caterpillar method we get the missing half a period, and here is your convergence to the mean, no matter if the mean is drifting or not. The market will surely return to the next Mach.
You asked and answered)) sometimes it's good to chat
i even if i just use MAshka, it's possible to filter out some minor Arch effect
oh about the caterpillars... don't make me think and do something )
I can't say anything about the NS.
Co-integration is quite a big trend in models and very well developed. There is no NS there at all, because in these models it is necessary to prove everything with tests, and you have it just like that.
Whether it is possible to use your idea - I do not know, I am interested only in TS, and which I prove their future properties at the stage of training.
PS.
It was a long time ago, but from memory your graph is very strange. It should be like this: if there is a trend, then the graph at the bottom is either above or below the axis, and you have it going back and forth.
In short, I will do my best and show you the results on the forward (tomorrow, I'm bored today)
There is no NS there at all because **** the coefficients are in a linear regression, and I don't have enough brains for nonlinear models
If it were a linear model, it would be higher/lower, all right, because you can't exactly fit a linear model to all fluctuations, but you can fit a non-linear one. In other respects, the principle is exactly the same as in the classic pair trading.
Maxim, don't make a fuss! In times of trouble, you have to call the Sorcerer for help!
Do you know why? This mysterious character knows a lot about distribution and NS. Somehow he treats the two as one.
And it works not with a number of increments, but with The sum of the increments. over a certain period of time.
As a last resort, I'm about to start a free distribution of my TS. It gives the best profit even without NS (for now on the demo :)).
I recommend everyone to prepare pockets and empty them of dust! Show must go on!
Maxim, don't make a fuss! In a difficult moment, you have to call the Sorcerer for help!
Do you know why? This mysterious character knows a lot about distribution and NS. Somehow he treats the two things as one.
And it works with a number not from the increments, but from The sum of the increments. over a certain period of time.
As a last resort, I'm about to start a free distribution of my TS. It gives the best profit even without NS (for now on the demo :)).
I recommend everyone to prepare pockets and empty them of dust! Show must go on!
so the sum of the increments will give the initial series :))) if you get them by subtraction
Let's get our pockets ready, I'll finish mine tomorrow ))) but distributions and NS are useful, for sure. I'm a little studying the analysis of variance myself now.
so the sum of increments will give the initial series :))) if they are obtained by subtraction
Preparing pockets, I'll finish mine tomorrow ))) but distributions and NS are useful things, for sure. I'm studying analysis of variance in my own little way now.
That's a good one! I recognize the old Maxim!
Sooner or later Koldun will say his word - you just need to be able to understand his hints.
It's been a long time since anyone posted a report - here's my forward.
I trained on 22.01.2018 on EURUSD H1, for code generation test, almost all available history from MetaQuotes - huge overfitting and the size of the EA, I can not even attach (if you are interested download the link) and here I run, seems to work... http://hlaiman.com/download/ea_eurusd_h1.ex4
It's been a long time, no one has posted reports - here's my forward.
Trained 22.01.2018 on EURUSD H1, for code generation test, almost all available history from MetaQuotes - huge overfitting and size of EA, can't even attach (if you are interested download from link) and here I run, seems to work... http://hlaiman.com/download/ea_eurusd_h1.ex4
It's a pity MT4 is an anachronism, I don't even have it installed :D Switch to MT5