The Rounding Dilemma
My mind was off on one again on today's walk. I was thinking about a problem that every actuary (and I guess every accountant?) will face at some point but which never gets discussed anywhere because it's something we would prefer to pretend doesn't exist. It's the rounding dilemma.
I'll go through an example and I hope people will reply to this article with how they would tackle the problem. We're into March and your boss has a job for you. He needs January's and February's "numbers" (this might be profit, sales, expenses, whatever) and the year to date number. He's only interested in numbers to the nearest million. You have some "Yeah but..." questions but he's not interested and sends you away with a JUFDI.
And here's your problem. In millions, January's number was 2.8 and February's 2.6, so year to date is 5.4. The rounded versions of these three numbers don't add up. So what do you do? Open call.
Here are some possibilities that spring to mind:
None of these solutions are prefect in every way. Me, I won't misreport numbers and I like things to add up so I'd go for 5 or 6 and face the boss's ire.
But other people will make different choices and, in particular, I've seen a lot of actuaries over the years go for option 4, the easy way out in many ways. If you're one of those actuaries, I have two follow up questions for you:
First, if someone else in the organisation wants to know February's number to the nearest million, do you treat this as a brand new exercise and round the 2.6 to 3, or would you tell them 2 so that it's consistent with what you've reported to the boss?
Second, what if March's number is 1.4 (so year to date is now 6.8) and your boss wants to see three months of numbers? Do you:
So, do you face similar problems to this on a regular basis and, if so, what's your preferred approach?
Or, if you're the boss in this example, what approach would you prefer to be taken?
Interim Head of Origination and Execution at M&G, Consultant, NED, IFoA Council
9 个月#6 is important - showing these numbers to only one significant figure is inappropriate. After that, #1 is by far the best in my view. Any numerate person understands why we present 3+3=5 and they shouldn't and won't worry about it. #2 and #5 are needlessly distracting. #3 is plain wrong. #4 requires a lot of manual work when you have lots of numbers to present and so should only be used when working with unsophisticated readers, like small children and government officials.
The Artistic Actuary
9 个月Great to see so many differing opinions on this. It means it was worth posting.
Shameless clickbait — I love it.? YTD would normally be the quantities of most fundamental importance, so round those and then *define* in-month (or in-quarter, etc.) as differences. ?Hence Option 4, although the justification that you gave was a bit woolly.? In general, any square matrix of full rank can be used to define what I’ve called “the quantities of most fundamental importance”. ?So that could be in-month or whatever else.
Risk management and analytics executive
9 个月It's a combination of 1 and 6. This example is crying out for some more significant figures. If I ask the wrong question ("Just give it to me in millions!"), I'd still prefer to get the right answer. But the right level of rounding won't necessarily give you a cleanly rounded total, so I'd just handle that with a note.