Long time member. Infrequent poster. May I ask a question here that does not concern current model output? If it should be somewhere else, I’m sorry.
I have some background with quant financial markets modelling. The models seen there tend to under-predict the incidence of extreme outcomes/conditions. Unlikely events occur more frequently in real life than they do in the models.
With NWM models however, it is my distinct impression that extreme (ie. “widespread winter weather”) conditions appear far more frequently in model output than they do in reality.
Is this correct? If so, why?
Originally Posted by: IanT
The weather models are mechanical - they calculate how conditions should evolve on the basis of laws of physics. Each set of inputs can only produce one outcome. The trouble is that the range of measurements you start with is never complete, so you're bound occasionally to miss one or two data points which might in reality have moderated an extreme outcome.
The ensemble is supposed to be a way of testing whether this is the case. By tweaking your initial conditions 20 or 50 times, you're supposed to cover the effects of possible gaps in your initial data. But every one of those ensemble runs is another one-off event with only one possible outcome. None of the 20 or 50 runs is more likely to occur in reality than any other. So we have to treat every one, including the most extreme ones, as equally possible. Thus we have every right to examine extreme model runs as much as boring runs. And we do!
As I understand it, financial modelling, by contrast, is essentially statistical. You're looking for what is more likely or less likely to happen, and applying percentages to that, and comparing it with the possible financial gain to be made by taking different positions. To put it in horse racing term you're looking for which outcome should be the favourite, and determining whether the potential gain of backing that outcome outweighs the risk of one of the outsiders coming in instead.
As you say, the favourite (as in horse racing) tends to be priced up just that little bit too short, as everyone piles in for the easy money, or the short-term gain. And when an outsider does come in, then everyone is shocked, even though a 1-in-a-thousand event had every right to occur at some point.
It takes me back to my grammar school, and doing my applied maths A level 42 years ago. There was a choice of two papers to take with the Oxford board back then: statistics or mechanics. Our maths teacher told us that statistics is a very straightforward subject, but essentially useless in scientific terms, so we covered the whole subject in four weeks, to a level which would have been enough to pass the exam. Then we spent the rest of the year studying mechanics, on the basis that it was much more difficult, but much more useful, so that would be the paper for which we would be entered.
It was a different world back then.
2 miles west of Taunton, 32 m asl, where "milder air moving in from the west" becomes SNOWMAGEDDON.
Well, two or three times a decade it does, anyway.