My company has using postgresql as database for ERP application which in-house developed.
For store financial and accounting data, we chose "numeric" type for accurate calculation (and with recommend by postgres's documentation) and we faced the problem by using "double precision" before.
However, we found that by using numeric had the problem too.
In our formula for calculate interest for customer's overdue payment that using numeric,
we found that it had fault calculate. So, it effected our interest amount.
In the attached file you can see that the result from postgresql and by using long division method is difference.
postgresql : 714.35599999999xxxx
long division method: 714.356
and if we multiply this result with interest rate and others factor and round up later. the amount is miscalculate.