Thanks, Ed. Interesting - the MRC tracking/revision model seems to work quite well, I see the prediction curve is going up steeply as we move further into early November. Interesting, I'm guessing a bit, but the technique looks similar to how ADPCM compression in telephony works: predict the value of the future points on the waveform, measure the difference from the actual waveform then use the rolling differences to adjust the coefficients in the algorithm (the signal is compressed because the algorithm runs at the sending and receiving end so it's only necessary to send the differences not the full waveform - but that's not relevant to Covid modelling).
Michael Ixer ● 1973d