My niece is about to go to University to read Psychology. I was surprised to find out that on her required reading list is a book called “Discovering statistics using SPSS”. It’s a monster tome of over 800 pages filled with maths. There are chapters on multivariate analysis of statistical variance, the chi-square test with standardised residuals and a section on factor extraction with eigenvalues….. No? Me neither!
SPSS stands for Statistical Package for Social Sciences. It first appeared in 1968 and has been much updated since. The SPSS manual has been described as one of sociology’s most influential books. Why? Because it turns sociology into a science. Most of the statistical methods used in the program, such as the least squared method, were invented by physicists in the 1800s. SPSS allows sociologists to plunder the wardrobe of physics. By dressing up in their clothes, it makes their discipline look more like a hard science and less of a touchy feely one: it now has numbers and maths.
One of my metaphors for catataxis is a shrink consulting a physics textbook when you are lying on his couch telling him about your father issues. He is analyzing your emotions, which are seated in the brain. And the brain, at the most fundamental level, is made up of subatomic particles. But you can’t analyse emotions by looking at subatomic particles; to do so is a catataxic error. You need to use therapeutic techniques not quantum physics – the right tool for the right level. Hence my surprise on finding out that psychology students have to study statistics.
Don’t get me wrong. Of course, statistics are useful things. They enable you to see a pattern that you might miss if you were too bogged down in the details. Statistics transmute a problem up one level, from messy reality on level one to pristine, summarising numbers on level two. Statistics let you see the wood for the trees. To run a regression on a set of data is to perform a catataxic transformation; one that can save lives with medical trials and the like. The problem is that having reduced the world to a numerical one, the judgements made on those numbers can be flawed. There is a nice warm feeling of security that you feel after you have ‘crunched the numbers’ and shown that they support your case. But your confidence in that numerical data mining may well be misplaced.
Which leads us to the UBS rogue trader who just lost that bank £2.3bn dollars. I must confess to some degree of schadenfreude. UBS took over Phillips and Drew in the 1980s, the partnership that I started work with in the City. They then proceeded to destroy it. So to see this global bank that boasts of its risk management skills humbled by a rogue trader brings a wry smile to my face.
“When will banks learn to control risks properly?“ many commentators ask. Surely the lesson is that it is not possible to control risks, not with a spreadsheet anyway. Most statistical methods rely on the bell curve; they assume a normal distribution of risk in order to make the maths work. The problem is that risk is not normally distributed, so traders keep dropping huge sums unexpectedly. Apples fall to earth, tides go in and out, riots happen in the summer and traders bankrupt banks. It’s the natural order of things. Nothing that really needs explaining.
I imagine that the senior managers at UBS looking at the trading accounts felt comforted by the numbers showing how profitable they were. But just as with psychology students, converting things up one level to the numerical domain does not necessarily make things safe, or even true. Just looking at numbers gives a false sense of confidence. Better to look one level down at the real world, the human world, messy and unstructured as it is. I bet the guy sitting on the desk next to the rogue trader all day for the last three years knew something funny was going on…