Should All Predictive Modelling Be Banned?
by James Lewisohn What are the world’s worst inventions? Winston Churchill famously regretted the human race ever learned to fly. I don’t (I’m looking forward to my next holiday too much). Instead, observing the destruction wrought by government pandemic responses predicated upon projected Covid cases, I’m beginning to regret mankind ever invented the computer model. I have form here. I spent the early years of my career building financial models, hunched over antique versions of Excel on PCs so slow the software might take twenty minutes to iterate to its results – which, once received, were often patently wrong. I developed a healthy mistrust for models, which frequently suffer from flaws of design, variable selection, and data entry ("Variables won’t. Constants aren’t," as the saying goes). Models allow outcomes to be presented as ranges. In business, it’s often the best-case outcome which kills you – early-stage companies typically model ‘hockey stick’ revenue growth projections which mostly aren’t realised, to the detriment of their investors. In pandemics, though, beware the worst case. In December, SAGE predicted that Covid deaths could peak at up to 6,000 a day if the Government refused to enact measures beyond Plan B. The actual number of Covid deaths last Saturday: 262. SAGE’s prediction was its worst-case analysis, but the fact that the media (and then the Government) tends to seize upon the worst-case...
