On 03.08.2016 23:58, Hendrik Boom wrote:

> Floating-point just isn't accurate enough. Multiple-precision scaled
> fixed-point would work, even if it's binary.

The limited precision isn't the actual problem, but the strictly
defined rounding rules. To be financially accurate, you'll have to
comply the financial calculation rules (which OTOH might not be
actually physically correct) - these are just different numbe spaces.


--mtx