Machine learning in trading: theory, models, practice and algo-trading - page 1386
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
thinning is to throw out half of the information again and coarsen the model
If you don't, your model will be full of garbage, and all its work will be to clean it up.
Again, there is a substitution of notions... there is no garbage in the market, it is not a landfill )) every price is important
Just like the market has no emissions from a classical point of view. Emissions are even more important here, they contain a lot of useful information, they are informative in general.
if we shift to level measurement, "emissions" disappear as a species
There is no garbage in the market, it's not a dump)) every price is important
This is the realm of mysticism. A flapping of a butterfly's wing causes a tsunami. It is possible, but the phenomenon is very rare. And it is not real to catch it.
Unfortunately, the quotes are just the upper part of the iceberg, they don't contain much information about the reasons of changes. There are purely technical factors, but they are not pricing ones.
That is why if you also thin them out there is not much left, imho.
Although, of course, it depends on how you thin it... I just added it to the topic of returnees, that imports get clogged when the sample increases, and you need to check for correlation
Unfortunately, quotes are only the tip of the iceberg, they contain little information about the reasons for changes. There are purely technical factors, but they are not pricing ones.
That is why if you also thin it, there is not much left, imho.
Although, of course, it depends on how you thin it... I just added it to the topic of returns, that imports get clogged when the sample increases, and you need to check for correlation
I discovered a kind of mathematical model of the market, rather than, as Automat claims, a dynamic model.
Been looking for a solution to this very market formula for a while, the results were, but not impressive. Soon I will return to the search, but with renewed vigor. Some small thing is always missed, now it's only necessary to understand which one.
I discovered a kind of mathematical model of the market, rather than, as the automaton claims, a dynamic one.
For some time I was searching for the solution of this very market formula, the result was, but unimpressive. Soon I will return to the search, but with new strength. Some small thing is always missed, the only thing is to understand, which one.
I stick to the generally accepted model of an efficient market
And I look for the inefficiencies, which sometimes appear, but are leveled out in a competitive market.
adhering to the generally recognized model of an efficient market
and look for inefficiencies that sometimes arise but are then offset in a competitive market
That's right. But I don't think that "deep" history significantly affects the present. At least based on the principles of efficiency. In the end, a deep analysis of history directly at work is meaningless. If there really were some regularities - dependencies, they would be easily detected by standard methods.
That's right. But I don't think that "deep" history has much effect on the present. At least based on the principles of efficiency. In the end, a deep analysis of history directly at work is meaningless. If there really were some patterns, they would be easily detected on history by standard methods.
it's not even the deep history, but the fact that the price is not at the same "level" as it was before
the models on returnees do not take it into account
it is not even the deep history, but the fact that the price is not at the same "level" as it was before
The returnee models don't take that into account at all.
I don't have returns. These are scaled prices relative to the zero sample count. I already wrote, in a stable market M(dC/C) = ~const, where C is the instrument price.
With other data preparation, the movements will depend on the instrument price, i.e. the scale will constantly float from sample to sample, while most methods of IRs are very sensitive to scaling.
I have no returns. These are scaled prices relative to the zero sample count. I already wrote, in a stable market M(dC/C) = ~const, where C is the instrument price.
With a different data preparation, the movements will depend on the instrument price, i.e. the scale will constantly float from sample to sample, and most MO methods are very sensitive to scaling.
this is the latest return value for each lag, no matter what you call it
you don't have just one example in a sample but many, the sequence of such examples for each feature will be a sequence of increments
Again, what's good for the IO is not necessarily good for the market. These methods were not developed for the market, that's why I'm looking for compromises