- Change design of predictions. Panel is used for covariances and MPOReturnForecasts. Require from the user a function that returns the prediction(s) (made at time tau) for time t.
- Move the non-explanatory code into cvxportfolio proper (e.g., factor model risk estimator).
- Move the rest into the documentation.
- modify data files so we don't store NaN (look at how you read them)
- real time optimization:
- Excel output
- FIX output
- rounding result trade vector (return rounded vector from the policy object).
- impose that last column of returns is cash return. user shouldnt specify cash_key
- makefile: support make docs and make pip
- clean up the gitignore for new stuff
- remove the pickle file, replace it with .csv or something
- fix them
- make them work with Travis
- ideally, they should work with python2
- make sure gammas are >= 0
- sector constraint should be equality
- cost -> constraint should be done, like costs=[gamma_leverage*Leverage()], and constraints=[Leverage()<= 3]
- documentation should be like cvxpy. we have variables constants parameters etc. we have a library of functions that manipulate basic objects, each one is documented... we have standard operators +- etc.
- not sure which features of python3 (other than print() and matmul) we're using. it might be easy to support also python2.
(It's a secondary objective)
- make a separate module that implements
- a Kelly "returns" object
- a Kelly constraint that does the risk part