17 May, 2012

Lies, damned lies, and statistics

Ever used one of those retirement planning calculators? Almost all of them ask your current age, your planned retirement age, and then look up a set of life tables to work how many years of retirement you need to fund, based on the remaining life expectancy for someone your age. You define the retirement income you’d like and it then tells you how big a lump sum you need, given average investment returns and inflation. All very sensible, you’d think. But I suspect most people forget that these calculations are all based on averages, and the most misleading average of them all is life expectancy.

Let me explain, using myself as an example: age now, 56; expected age at death, 82. Let’s say I plan to retire at age 65 (I don’t); when I get to 65, my expected age at death is 83, ie. another year to fund. But if I make it to 83, my expected age at death is 90, so it’s 8 years more. And so it goes on. In other words, if I follow the usual retirement planning advice, I may not have as much money as I need.

There are solutions, such as buying annuities or long life insurance, although these are not very common in my principal country of residence, largely because of a deep suspicion of the fees and overhead costs of the providers, and doubts about their longevity. Self-funding is common, but most people don’t understand that the suggested amount is based on an average - the wrong average. (And who reads the small print in the calculator’s notes?) In any case, I’ve not seen a calculator which factors in the cost of an annuity, etc..

It’s not just retirement planning. Businesses and governments misconstrue or misuse statistical methods in many applications. People wrongly assume that averages equate to what they actually will experience. They’d be better off using percentiles. In the dim and distant past, in the age of steam-driven computers, one of my specialities was “sizing”, using queueing theory to calculate the size and speed of computer configurations to deliver the required performance of a set of tasks. In a very few years, the basis of contractual performance specifications for response times (between hitting Enter and getting the answer back on the screen) changed from average time to 90th percentile time. Using a high percentile gives a much more effective definition of the response time experienced by most users, most of the time.

I don’t know what the appropriate life expectancy percentile should be for retirement planning, but it is not the 50th percentile for your current age.

By the way, chaps, one of the little known effects of us blokes surviving longer into old age is that we start to catch up to the life expectancy of women. So you can feel smug in the knowledge that when you pop your clogs aged 101, She Who Must Be Obeyed is likely to follow you within a few weeks. If you believe in an afterlife, that might or might not be good news!

First posted November 10th, 2008

No comments: