Machine learning in trading: theory, models, practice and algo-trading - page 2607
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
Even assuming it is proven (although there can and often is a problem with this) that someone earns consistently year after year, it is not at all clear how proving that this is done by the same algorithm can even look like. I would like to see more meaningful options than "take your word for it" and " I'm telling you so".
The market has some stable characteristics. Stable trading behaviour can follow from them. Stable algorithms exploiting this behaviour can follow from it.
The yellow highlighted - practically a perfect description of the trading results based on the patterns identified on the history) It is not necessary that everything will break immediately, but it is almost always "it is not like that now")
There is a question like this:
Two models are used. One predicts to buy or sell, the other to trade or not to trade.
First the first model is trained, then we look where it predicts poorly, mark these examples as "don't trade", the other good ones as "trade", teach this to the second model.
The first model is tested not only in the training area but also in the additional area and the second model is trained in both areas.
We repeat this several times, retraining both models on the same dataset. The results gradually improve on the samples. But not always on the control sample.
In parallel with this we keep a log of bad trades cumulative for all passes, all "bad" deals for "not to trade" are collected in it for training the second model and filtered according to a certain principle like the more copies of bad deals for all passes, the more chance to mark them as "not to trade"
For example, for each date some amount of bad trades is accumulated for all iterations of training, where this amount exceeds a threshold (mean, average), those trades are marked as "do not trade". The rest trades are skipped, otherwise it would be possible to exclude all trades if there are a lot of training iterations.
coefficient allows you to adjust the number of trades at the output, the lower it is, the more trades are filtered out
... by this point i am already tired of writing ...
How can such a combination of models be improved so that it improves its results on a new independent plot?
Is there any philosophy as to why this might work? Other than the fact that the models naturally improve each other (error drops) on each round of retraining, but how to get rid of the fit?
Illustration. The graph is split into 3 parts. The last one trains the first model, the penultimate and last one the second, the first third is an exam sample. Naturally the last section will be the best and the first third the worst.
Here there have been 15 iterations of retraining both models, using a log of bad trades.
Frankly, the scheme as a whole looks very sophisticated and it is unlikely that a person from the outside would be able to say anything meaningful.
1) There is some association with boosting, where each successive model tries to improve the error of the previous ones.
2) I prefer the approach when we do not try to obtain one complex model for all cases but make several simple ones that work according to the principle of trade/not trade. You can make two: "buy" and "sell". You could have four: "buy after an up move", "buy after a down move"," sell after a down move", "sell after an up move". Maybe, more variants can be invented) Then they can be combined somehow - all kinds of creative options are possible here too).
The yellow highlighted is an almost perfect description of the results of a trade based on the patterns identified on history) It is not necessarily that everything will break right away, but it is almost always "now it's not")
It is not necessary that everything will break right away, but it is almost always "it's not like that now".)
I was beginning to think we were slowly slipping into sophistry. It's not about claiming that there are "eternal" patterns (eventually the sun will go out with probability 1 too). The point is that there are laws whose lifetime is long enough to be exploited to one's advantage. And their "lifetime" is usually directly proportional to the complexity of mining or the technological complexity of exploitation (e.g. CPT). Experience shows that, with some diligence, a few/many months is reasonable.
I was beginning to think that we were slowly slipping into sophistry. It cannot be argued that there are "eternal" regularities (eventually the Sun will go out with probability 1). The point is that there are laws whose lifetime is long enough to be exploited to one's advantage. And their "lifetime" is usually directly proportional to the complexity of mining or the technological complexity of exploitation (e.g. CPT). Experience shows that, with some diligence, a few/many months is reasonable.
hard to disagree
i would also like to write a program for exchanger that will work out some algorithm before clearing
but it looks like I'm tired of writing
The yellow highlighted is almost a perfect description of the trading results based on the identified patterns) It is not necessary that everything will break right away, but it is almost always "not so now")
I was beginning to think that we were slowly slipping into sophistry. It cannot be argued that there are "eternal" regularities (eventually the Sun will go out with probability 1). The point is that there are laws whose lifetime is long enough to be exploited to one's advantage. And their "lifetime" is usually directly proportional to the complexity of mining or the technological complexity of exploitation (e.g. CPT). Experience shows that, with some diligence, a few/many months is a reasonable expectation.
Actually, the hope for something like that unites all of us here. I just wish that this hope would not distort the perception of reality.
Hope is a good breakfast, but a bad dinner)
The yellow highlighted is an almost perfect description of the trading results based on the patterns found on the history) It is not necessary that everything will break immediately, but almost always "it's not like that now".)
If you don't like the previous example, in narrow circles there is a widely known trading algorithm based on the fact that every day is day, followed by evening, morning and then again.
It is said that the profitability and capital intensity of this algorithm increase significantly in spring and autumn)