Machine learning in trading: theory, models, practice and algo-trading - page 2546
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
Here are a couple of quotes from the article:
"The peculiarity of artificial intelligence is that the technology is not capable of navigating new non-standard situations. If an abnormal situation occurs in the market, the model is unlikely to suggest the best way out. The pandemic is a prime example of this. The Organization for Economic Cooperation and Development (OECD)cites data that, according to aBank of England survey, during this period about 35% of banks experienced negative consequences from the operation of an AI model based on machine learning.
What do banks have to do with it - they need ML to determine a customer's creditworthiness at the stage of "granting or not granting" a loan...
And the exchange rate of one currency to another - even in a pandemic it will be alive +/- back and forth ... and if the export-import operations are reduced, and hence the NFI (if the currency is no longer particularly needed in the country to pay foreign partners) -- then speculation will only on medicines, or terrorist attacks and strikes and transfer of reserves (or rumors) back and forth ...
And with a normal learning horizon, such situations have already happened in history -- all DataMining is based on the fact that history repeats itself, only with new parameters at a new round of evolution... so don't take absolute parameters, take relative ones...
Banks can't ML-monitor clients' preferences to "take or not to take" a loan... And many people take a loan only once in their life - for an apartment...
although to be fair, DataMining is not just MachineLearning... but also qualitative statistical analysis before it and full-fledged MetaLearning in neural networks, so that the model adjusts to current realities... And even in this case a fairly reliable forecast is possible only at a certain point, but not for a long period...
banks cannot have all the information to load it into their model... ... and they use AI for purposes other than predicting quotes (they don't do that)... they only need lend and borrow at convenient interest rates...
p.s. in schools they use ML to do professional orientation more accurately than banks do to identify the possibility to lend to a certain client... - and they have nothing to do - to give loans at any conditions, otherwise they themselves will be out of work and without pay (not the AI's fault, but the demand for their services under their terms with their Customer Relationships Management)...
but with the weakening of international trade in times of adversity (pandemics) it is understandable that they have liquidity shortages... but in history there were moments of liquidity shortages (only the parameter was not a pandemic) - it's not new - the only question is what model, what is estimated, what is the horizon of the training sample and how its interpretation conclusions are used by real people of the bank...
p.p.s.
information from OECD about the Bank of England may be just speculation or black PR with an axe in the hands of a lumberjack who is just getting wood... The crisis of 2008 did not appear out of thin air, and the preconditions were already visible 2 years ago... and in 2008 (I think) the fact - speculation on the fall of LehmanBrothers to do QE, perhaps even for the same money ... by the way the "guilty" then were the banks in the financial crisis (although the cause-effect relationship has not been abolished - MBS also did not appear out of thin air, but out of demand!...) - Now, apparently, they blame the AI - at least it won't answer... here's where the number of pirates was linked to climatic conditions in the comments -- maybe someone is still surfing the universe in search of a credible signal?
Has anyone poked around with wavelets without libraries? Wavelets are defined at infinity, what length should I take then?
I haven't studied the decomposition itself yet, I don't know the answer.
But in the example in the block diagram, you can see that the window 10 is defined.
Probably the definition at infinity, doesn't play a role.
And take the length that solves the problem most effectively.
I haven't studied the decomposition itself yet, I don't know the answer.
But in the example block diagram you can see that they set the window 10.
Probably the definition to infinity, does not play a role.
And take the length that solves the problem as efficiently as possible.
In idea the length of the window should be different
In theory, the length of the window should be different
To be honest, I don't quite understand the question.
Maybe that's what it's about?
The length of the window is supposed to be different.
The best solution is to consult a textbook.
Wavelets are the same as Fourier. There is classic Fourier, there is window Fourier, and there are wavelets, where instead of a rectangular window as in window Fourier, a special kind of window - wavelets - is used. Fourier is not suitable for financial quotes, because of the random nature of the quotient.
It is believed that wavelets are good for economical representation on all available price history (instead of limited window)
Shapelets are more often used for time series, but even that approach is not very popular
Searching for shaplets reminds me of clustering of series segments. For sure they are useful for signals like cardiograms, but I am not so sure about their usefulness for studying prices.
By the way, did you manage to figure out how to use the LGBM model? If trained in R, you could try to use the library of San Sanych)