Rounding and Accuracy

[ubergigglefritz]

Has anyone else noticed how HUGE of a difference rounding can make in the problems in MFE? I'm really worried that I'm going to fail this exam from this matter alone, ha ha. I try not to round anything when I'm doing these problems, but it seems you have to round somewhere. Is there an easy way to get a more accurate measure of the normal distribution probabilities beyond a z value to 3 decimal places?

[mingchenckm]

hmm... one way to do that is to use interpolation...

BUT SOA has clearly stated that candidates should NOT use interpolation in the exam. I don't think they will trap you, so just forget this, and follow the instruction to round off all z values.

[TRINIDON2K]

I think for most of the examples in Mahler's MFE manual, rounding gives you an answer that may be off from the solution but still within the range

Like the solution would 5.04, your answer would be 5.92, and the answer choice would say between 4 and 7

[Bison]

I wouldn't sweat it for the reason that Trinidon2K just mentioned. Many of the exam questions (for other exams, we obviously have never seen an MFE exam) have answers that are ranges, not exact numbers.

Maybe someone who sat for this exam can correct me if I'm wrong, but I believe I remember reading something about a question on C a year or two ago where how you rounded it did make a difference on the answer, and credit was awarded for both answers in the end. I'm guessing that the SOA won't create a similar situation on the exams going forward.

As for me, I just take the reading right off the normal table w/o interpolation, and use 4 or 5 decimal places throughout the problem, and I haven't had an issue.

[rawl316]

Hi, I was reading through this, and what is the SOA rule on rounding for black scholes?

[Bison]

Originally Posted by rawl316

Hi, I was reading through this, and what is the SOA rule on rounding for black scholes?

For the N(d1) and N(d2), do not interpolate, just round to the most accurate reading from the provided normal table. Meaning if you get d1=0.838, you should use the 0.84 reading from the table.

http://www.soa.org/files/pdf/mfe-05-07final.pdf
Page 5

[kyle.mcwhinnie]

Also,other than the rounding used for the normal distribution, I just save numbers from calculations from intermediate steps in one of the 5 variables on the TX30II, this keeps your final answer exact without having to write out many decimal points.