IRR calculation errors: Singular Jacobian matrix and Failed to reduce function values.

[bdeterling]

I've been getting this error on many different series. I've tried various versions of Ruby 1.9 and 2.0. 1.9 tends to hang whereas 2.0 errors out fairly quickly. I've tried these same calculations in both Excel and the Java Apache POI library and they return good values. Any ideas?

Some examples:

[-1,-2,0,5].irr.to_f
=> 0.24189656303203824

[-1,-3,0,5].irr.to_f
RuntimeError: Failed to reduce function values (newton.rb:66 in nlsolve)

[-13000,-3000,-3000,0,0,5000,10000,20000].irr.to_f
RuntimeError: Singular Jacobian matrix. No change at x[0](jacobian.rb:58 in dfdxi)

[tubedude]

I've added a pull request that adds a guess param to IRR and XIRR.
#30

The error I mentioned was solved by:

[-13000,-3000,-3000,0,0,5000,10000,20000].irr(0.5).to_f
# => 0.10803084598726499

[marksweston]

@tubedude as you've closed your own pull request rather than merge it in, I get the impression this is still an open issue. So what next?

[tubedude]

Yes. I can try to work on a better implementation. My solution is to guess an IRR and add this guess to the XIRR method.
Today the XIRR method takes 1.0 as the initial guess which can leads to error like these.
What do you think about it?