phschiele
commented
3 years ago @Pjuske Would you mind sharing cov_matrix and mu_vector to make the error reproducible? Also, which function did you call on the EfficientFrontier object (efficient_return(), ...)?
The default bound values are weight_bounds=(0, 1), so negative weights are not expected.
The output "Expected annual return: 4004.1%" suggests that the optimization problem might be ill-conditioned.
On a more general note though, @robertmartin8, I think when defining the weight variables here
# Optimisation variables
self._w = cp.Variable(n_assets)one could add the keyword "nonneg=True" if all lower bounds are >=0. Source
In my experience, this makes the optimization a bit more robust to problems such as the one described here.
However, as the bounds might get changed later, this likely requires a bigger change.
Pjuske
I'm not sure if I missed something in the documentation, but I suppose that the efficient frontier shouldn't output negative weighted assets. (It works fine if returns are computed using _mean_historicalreturn()). What could be the possible cause of this?
Thanks in advance
cov_matrix = risk_models.sample_cov(Closeprices)mu_vector = expected_returns.ema_historical_return(Closeprices)Console-output of
EfficientFrontier(mu_vector, cov_matrix, verbose=True),.portfolio_performance(verbose=True)followed by some assets contained in.clean_weights()printed:with
mu_vector = expected_returns.mean_historical_return(Closeprices)