NaNs with ERC (Equal Risk Contribution) about portfolio_allocation_js HOT 5 CLOSED

Thanks a lot, I will have a look as soon as I find the time.

NB: I can think of NaNs happening in case the covariance matrix is not (semi) positive definite, but I would expect the algorithm to throw at some point instead of outputing NaNs, so, I am definitely intrigued :-) !

from portfolio_allocation_js.

Thanks for your reply!

I thought so too about the precondition of the covariance matrix.
While debugging, it seemed that the NaNs happened here.

from portfolio_allocation_js.

I confirm the issue is related to the covariance matrix being semi-definite positive (i.e., determinant equal to 0).

In this case, it might happen that the variance is (numerically) negative zero (e.g. -0,000000000001), from which it is not possible to compute a volatility.

I added a proper management of this specific case on my local computer (more tricky than it seems due to the convergence condition of the algorithm which needs to be slightly amended), and it is now working OK on your example.

Commit will follow with non-regression tests, and will be included in the next release 0.0.7.

Thanks again for reporting.

from portfolio_allocation_js.

Thank you very much, looking forward to the new release :)

from portfolio_allocation_js.

Solved with today's release of v0.0.7 of the library.

from portfolio_allocation_js.