Machine learning in trading: theory, models, practice and algo-trading - page 181
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
Micha, again?))) Hilarious... I don't know about you, but we share, and mode, and tear out shreds)))
Beautiful. Are these serious graphs, or sarcasm? Manual partitioning of hyperspace?
Beautiful. Are these serious graphs, or sarcasm? Manual partitioning of hyperspace? How do you do it if the predictors are not 2, but 20?
Beautiful. Are these serious graphs, or sarcasm? Manual partitioning of hyperspace? How do you do that if the predictors aren't 2, but 20?
The essence is the same, nothing fundamentally changes if the chips are quasi-orthogonal, if not it is desirable to compress the dimensionality, but in general it does not matter, that 1 or 3 or 30, it is only a process of "masking" hyperspace, in the case of classification
If the dimensionality is reduced by an order of magnitude the training is much easier, and the compression noise should be reduced, imho.
it depends on what task....
For example, when you need accuracy of entry, as I did, I coached the algo on bounces, then when you reduce the dimensionality (I did "PCA") vision of the algo is kind of blurred/smudged/smoothed and it instead of catching the bounce began to buy on the hays and sell on the lows
If dimensionality is reduced by an order of magnitude, training is much easier, and compression noise should decrease, imho.
Of course, however, it is not so simple, for example SVM is linear, with all that it implies, but another thing is a favorite kernel version of it, but why? It's all about the "kernel trick", which does nothing but project into a much more multidimensional space, for e^(-||w-x||^k) is infinite, why do you think the simplest kernel classifiers work so well? The whole point is that the dimensions become proportionally larger than how many sections the sample was divided into for multiplication by each kernel and depending on the kernel function, play around with the kernel size for the kernel SVM and its learning and computation speed. Multidimensionality is not always a bad thing.
Specifically in our case, as some gentlemen above correctly said, it is about data, not machine learning tools. All sorts of tricks of compression of dimensions, automatic fichaextractors a la convolutional nets won't help if the input was only multidimensional noise, or even one-dimensional:)
https://en.wikipedia.org/wiki/Renaissance_Technologies
The firm is an early pioneer ofquantitative trading, where researchers tap decades of diverse data in its vast petabyte-scaledata warehouse to assess statistical probabilities for the direction ofsecurities prices in any given market. Experts attribute the breadth of data on events peripheral to financial and economic phenomena that Renaissance takes into account, and the firm's ability to manipulate enormous amounts of data by deploying highly efficient and scalable technological architectures for computation and execution, for its consistent success in beating the markets.[19] In many ways, Renaissance Technologies, along with a few other firms, has been synthesizing terabytes of data daily and extracting information signals from petabytes of data for almost two decades now, well beforebig data anddata analytics caught the imagination of mainstream technology.[20]
For more than twenty years, the firm's Renaissance Technologieshedge fund, which trades in markets around the world, has employed complex mathematical models to analyze and execute trades, many of them automated. The firm uses computer-based models to predict price changes in easily traded financial instruments. These models are based on analyzing as much data as can be gathered, then looking for non-random movements to make predictions. Some also attribute the firm's performance to employingfinancial signal processing techniques such as pattern recognition.The Quants describes the hiring of speech recognition experts, many from IBM, including the current leaders of the firm.
The question is how a simple algotrader who can donate no more than $1-2k per month for data can survive in the environment of such monsters as Goldman, Renaissance and Tesa. How to find the most relevant data, the tip of the iceberg of what the super giants use.
Since you've decided to compare homemade with the work of large organizations, it seems to me that you should give up the search for some graphical patterns or indicators and focus on exactly what they do.
I mentioned an interesting solution here, https://www.mql5.com/ru/forum/96886/page2#comment_2866637
However if you look for patterns in how big players move their orders, how they are executed, how the price behaves after big markets or icebergs and so on, It may still work on the Moscow Stock Exchange, if they send raw flow, not aggregated. a year ago they introduced aggregated flow back from the exchange's core. it may be difficult on stocks, because there are too many ECNs plus darkpools.
Maybe just keep it simple.
Try to analyze the crowd and yourself as well, and predict the crowd actions and then act vice versa, because all these monsters, like Goldman, Renaissance and Teza, always need a counter agent (sucker) in a deal, whose liquidity he will open his position and then will lead the sucker to his own stop-loss, thereby provoking a liquidity splash in the other direction and about the same liquidity and will close his position, while the sucker will lose... That's the market as it is...If you find the crowd you'll find the big guys...
It even seems to me that the whole competition of these monsters Goldman, Renaissance and Tesa in the market is about who will pour the sucker the quickest... And you should not believe these fairy tales about the trillions of dollars of market turnover, the banks have trillions, and the crowd has always finite money and the sum is much more modest, and they (the banks) are waiting every day with impatience for new peasants on their field of wonders, simulating activity, in the glass and in the market as a whole