Friday, June 05, 2009

Actuarial software

I now have a wonderful little tool called a dongle, which means that, even if my company persists in sending me to far-flung* locations, I can keep playing with teh intarwebz.

And I can keep bothering my readers (if any are still around) with pointless theorising. Bwahahahaha.

On with the show. My last post discussed actuarial software, specifically how hard it is to get hold of. Since then I've done a little bit of research on the subject...

Q: What is actuarial valuation software?

A: It's software that allows you to pull some numbers out of thin air ("make actuarial assumptions"), punch 'em into a standard statistical model, and thus figure out how much money your company needs to stockpile to ensure that employees get their promised pensions.

Q: Why not just use spreadsheet software?

A: Because many of the statistical models require incredible amounts of processing power. Also spreadsheets are too easy; if actuaries used them then we'd lose our aura of mystery.

(More seriously, these models are easy to screw up so it's best not to have the uninitiated trying their hand at them.)

Q: Why not use a general-purpose programming language?

A: Because actuaries generally don't think of themselves as programmers. Most of them can't code for toffee, and can't be bothered to learn - after all, that's not what gets them the big bucks. The purpose of actuarial software is to allow actuaries to program without actually needing to know any of the relevant concepts.

Q: Are the statistical models really worth all this effort?

A: Not really. There's no such thing as a crystal ball, and no such thing as an actuarial model that won't be blatantly wrong thirty years down the line.

A good example is smoking. A lot of the mortality rates we use are based on the tacit assumption that a sizeable proportion of the population has been inhaling plant-based tar for a lot of their life. Now that smoking is becoming less common in developed countries, our models can't always deal with the resulting increased life expectancy. See the intro to this article for an indication of how technical this can get.

So why do we bother? Let's get this straight: actuarial models will not allow you to prove that you're saving the right amount for your employees' retirement. However, it will allow you to prove that you're saving some money, and that the amount you're saving is justifiable.

This is of great interest to regulators, so they force pension schemes to jump through these hoops. It's like getting a degree from a prestigious uni - it doesn't actually prove that you've got two neurons to rub together, but it does make it a lot easier to filter out morons. Read up on information asymmetry for more info.

Q: Back to the main topic. What does actuarial software actually do?

A: Your standard actuarial software package will contain:

1) A bunch of standard actuarial algorithms, designed to predict e.g. mortality of pensioners.
2) A set of modules to handle country-specific or industry-specific regulatory requirements.
3) A lot of dainty footwork to allow things like distributed processing (important given how hefty some models are).
4) A user-friendly interface (remember, this has to be used by actuaries, who are generally not techies).

Q: And you're actually planning to produce all that???

Probably not. But it's an interesting goal to think about.

* This is Britain I'm talking about. In USA terms, this translates to "the other end of the state".

No comments: