Machine learning in trading: theory, models, practice and algo-trading - page 153
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
What do you mean by that? It seems like there is something interesting in the phrase, but the meaning escapes me.
I was wrong there 2 sigma, well in general the price increment for a certain period of time, for example a minute, as a feature can add about 4-5% additional information about where the price moves in the next minute, and so on for 5 min 5 min for an hour and so on, well it is all very rough, and "information" does not mean that this will also move, it means how much this feature in a fairly complex system that takes into account thousands of features. What was before means almost nothing, noise.
1) Narrow understanding of the question. a convolutional network filter can perform many operations on an input vector, and form, for example, n moving averages, partially overlapping or not, or take several sums. Which is already a claim for normal feature engineering.
2) Again, understatement or misunderstanding. For convolutional networks, it is enough to feed "raw" data (in a pseudo-stationary form), such as price returns with a lag of 1. The network does the rest.
I'm not arguing with convolutional nets, the question is WHY they work, especially well in pictures, spectrograms, etc. The main thing is that the vector components are "smeared out" with each other, in fact the picture is not even a vector, the convolution layer simply compresses the dimension, multiplying the correlated areas by a number of filters selected by bullpen, these filters can be obtained by different methods, but the picture is, so to say, ALL equal, and the time series its "figure" is in fact noise, only its last bars and last bars of hundreds and thousands of other instruments of various exchanges have sense.
And by adding the history of one day of one series in CNN andthe network itself does the rest. we get... what everybody gets
Forum on trading, automated trading systems and trading strategies testing
Discussion on Article "Working with Sockets in MQL, or How to Become a Signal Provider"
pavlick_, 2016.09.09 10:52
I'm in linux, hence ipc becomes a non-trivial task at all (communication between terminal under wine and linuex exe). And IPC via network is a universal way. I connect µl script to linux program via loopback (127.0.0.1) on the same computer. Basically, I wrote a linux api for the terminal (the μl script handles requests and sends price data or places orders).
In my situation this is the best way of IPC from what I've tried. I don't want to transfer my developments to µl, i.e. i don't want to be bound to a certain language.
It is possible that we will make a gorgeous gift for the R community.
Time will tell.
Guys there is an idea, it is worth checking, I had it for a long time, I wanted to check but could not figure out the package and somehow forgot and abandoned, but here read the threadJ.B and remembered, it turns out he also did something similar:)
We are talking about cross-correlation - we can calculate how much one BP lags behind the other and whether there is a connection between them ...
The essence of my idea was to monitor simultaneously a large number of pairs and build something like a cross correlation matrix to compare each pair with each other and find pairs that follow each other for a while but one lags behind and trade this lag, since the market has nothing more constant than time, I think the calculations should be done constantly on each new bar, to immediately notice when a new relationship appears and also to immediately notice when this same relationship disappears ...
You can take anything, any predictor, but I think the best approach is pairs as market makers when the price of their instrument is almost always guided by one or a bunch of other instruments, and the classic indicators are unlikely to fit
The same way you can try to train a neural network on such dynamically changing predictors, in short, everything is limited only by imagination ...
I would try to implement it myself, but I'm busy with other project and don't want to spill my guts
Standard function of cross-comparison in R-ka ccf()
is an advanced package with preliminary spectral breakdown into levels and then already check for cross-correlation "wavemulcor" it also allows you to compare many BPs at the same time
We are talking about cross correlation - we can calculate how much one BP lags behind the other BP and if there is any connection between them...
The essence of the idea is to monitor simultaneously a large number of pairs and build something like a cross correlation matrix to compare each pair with each other and find pairs that follow each other for a while but one lags behind for a while and trade this lag, as there is nothing more constant than time in the market, I think these calculations should be done constantly on each new bar, to immediately notice when a new dependence appears and also to immediately notice when this same dependence disappeared ...
to compare many BPs at the same time
You're right, as a start, but you need to use not only currency pairs, but also commodities, stocks, futures, options, bonds, indices and anything else you can buy, preferably as an order book, at least the second price, volume and order book, also NLP social networks, important news itself, and cross-correlation is for adjusting, because it's linear, You need to refine the data and let it all into ensembles of classifiers with different functionalities.It takes 5 years of 3 to 5 qualified specialists, that's just on the salary more than mio$, it's a huge job, you can't do it alone, except in a very "pimply" form, which will fail even if alpha-potential in some places, but nevertheless it is already something. In general, you need to be a fan of the case, then there is a chance))) I hope that the convolutional or recurrent network will do everything by itself, this is a naive view, as well as previously thought about graal turkeys.
PS: To discuss publicly the specifics, for obvious reasons, it's not worth it, I've heard of stories, so far in the West, when people were blacklisted for too substantial pappers on the application of ML to algotrading, and then no one took them into any fund, the only thing left was to teach in the institute or deal with the market, youtube, write blogs, teach suckers how to sink for a couple hundred bucks(((
Lao Tzu was an algotrader saying: "He who knows - doesn't talk, he who talks - doesn't know")))
That is, you can talk about many things, but only in general, and especially sophisticated ones deceive a little bit when someone gets too close to the sun))))
This has been on MQL for years.
I was wrong there 2 sigma, well in general the increase in price over a period of time, for example a minute, as a feature can add about 4-5% additional information about where the price moves in the next minute, and so on for 5 min 5 min for an hour and so on, well it is all very approximately, and "information" does not mean that it also will move, it is meant as the weight of this feature in a fairly complex systems that take into account thousands of features. What was before means almost nothing, noise.
I get it now.
It's consistent with my observations. I call it mirroring (I'll probably post a couple of charts on this later). For predicting n-minute price gains, looking backwards for the same n-minutes shows better explanatory value.
But beyond that, I have very extensive data on which horizon the predictive power is greatest.
Where do the 4-5% numbers come from? How are they counted? Significance of predictors, R^2, mutual information?
Preferably links toccf() andwavemulcor
Somehow we got by without it.