Bayesian models for alpha estimation.
This project is no longer actively developed but pull requests will be evaluated.
There are currently two models:
-
the returns model, which ingests a returns-stream. It computes (among other things) a forwards-looking gains parameter (which is basically a Sharpe ratio). Of interest is
P(gains > 0)
; that is, the probability that the algorithm will make money. Originally authored by Adrian Seyboldt. -
the author model, which ingests the in-sample Sharpe ratios of user-run backtests. It computes (among other things) average Sharpe delivered at a population-, author- and algorithm-level. Originally authored by George Ho.
To install:
git clone [email protected]:quantopian/bayesalpha.git
cd bayesalpha
pip install -e .
To use (this snippet should demonstrate 95% of all use cases):
import bayesalpha as ba
# Fit returns model
trace = ba.fit_returns_population(data, ...)
trace = ba.fit_returns_single(data, ...)
# Fit author model
trace = ba.fit_authors(data, ...)
# Save to netcdf file
trace.save('foo.nc')
del trace
# Load from netcdf file
trace = ba.load('foo.nc')