Feature examples were highly requested. I will do in depth reviews for many but I can’t possibly do all. So I’ll give resources for top feature examples -
Stefan Jansen- ML 4 algorithmic trading - GitHub- Chapter 23; 200 or so features based on data mining and TA
…
Stefan Jansen- ML 4 algorithmic trading - GitHub- Chapter 23; 200 or so features based on data mining and TA
…
Yam Peleg (YAMQWE) submission notebooks for the G-Research Crypto Forecasting competition on Kaggle. A few hundred features here from rolling volatility etc.
Another good one is Wikipedia, just go to statistical signal processing and search for all the methods…
Another good one is Wikipedia, just go to statistical signal processing and search for all the methods…
They can usually all be applied with good success. Another good one is time-frequency/ wavelet analysis for which textbooks can be found. Financial signal processing is awesome.
…
…
Microstructural features can be derived from hasbroucks book or the literature.
Statistically Sound Indicators For Financial Market Prediction: Algorithms in C++ is a massive feature dump as well.
Roadmap GitHub section is filled with code examples not just to assist…
Statistically Sound Indicators For Financial Market Prediction: Algorithms in C++ is a massive feature dump as well.
Roadmap GitHub section is filled with code examples not just to assist…
You in your learning but also to speed up your feature engineering. There is no shame in copying them, alpha is alpha at the end of the day.
I’ll leave it there for now but wavelets are definitely something to look into. Fourier is not great because it only gives…
I’ll leave it there for now but wavelets are definitely something to look into. Fourier is not great because it only gives…
frequency and not time so it needs stationary data. That’s why time-frequency analysis or more simply wavelets are needed.
Another approach is using differential equations. There is a large literature of how AI scientists constrain their algorithms with physics…
Another approach is using differential equations. There is a large literature of how AI scientists constrain their algorithms with physics…
This is good for genetic algorithms especially, not so much NNs.
Entropy is good as well but needs lots of exploring and is best done with a Bayesian approach.
Other models like AR can make features as well.
simple copula or cointegration models against a pool are good as…
Entropy is good as well but needs lots of exploring and is best done with a Bayesian approach.
Other models like AR can make features as well.
simple copula or cointegration models against a pool are good as…
well but I will leave that there because it’s really a whole thing itself.
Finally, look into statistics and use tests as indicators (rolling tests) or to make indicators better. This should be one of the first things you test when experimenting. It’s free alpha lol
Finally, look into statistics and use tests as indicators (rolling tests) or to make indicators better. This should be one of the first things you test when experimenting. It’s free alpha lol
• • •
Missing some Tweet in this thread? You can try to
force a refresh
