|Re: accuracy benchmarks for financial calculations|
Message #6 Posted by Kim Hansen on 16 Jan 2009, 10:36 a.m.,
in response to message #3 by Egan Ford
By "extend the Voyager", I mean to take what was done well (nearly everything considering the time) and augment that with modern day accoutrements. For instance, I was always frustrated with cash flow entry and how difficult it was to make sure your entries were accurate... the touchscreen allowed me to incorporate an additional ability to view and edit the cash flow sin the same format as the cash flow diagram. Another example is Roy figured out how to make an amortization table that sums the interest and principal columns correctly (new algorithm). My product includes the original HP technique with rounding errors, but also allows you to email a schedule that uses Roy's new technique, a schedule that is Excel compatible and the periodic interest and principal are calculated so that the total interest paid plus the total principal paid adds up to the amount you paid over the life of the mortgage. As near as I can tell, this is unique and solves a problem for the loan industry that has bothered them for 70 years.
And another really really nice thing is that the new hardware platform allows me to perform the calculations at much higher precision and do so blindlingly fast... IRR runs in the blink of an eye instead of 30 seconds. Much of this is due to the faster hardware, but it is also due to Roy's reworking of the algorithms. But in any event it allows you to retain your 12C capabilities (plus bonuses) without having to actually pack it around.
Back to the question though... I have a 12C and have used it extensively as "truth" and to make sure my product operates in the same way but what I'm looking for is a source of problems that when run, will stress financial algorithms and point out areas where the algorithms are weak. An example comes from calculus, where an insanely large number of payments (1e81) and an equally small interest rate (1e-79) should take a present value of 1 to a future value of (approximately) e. Not a real problem in the finance world but one that verifies the quality of the underlying mathematics.
In the archives there was a brief reference to such a set of problems but the link was dead, I was just wondering if anyone was aware if such a collection might be resurrected.
Thanks, and I do appreciate the good wishes. As an Engineer, I wanted to do a 41C but the cxp is so well done that I concentrated on the 12.