I’ve adapted my model to forecast the impact of COVID-19 in the USA. Initially, I calibrated the model to fit the UK situation – but it’s built in a way to allow assumption changes appropriate for any geographic location.
Forecasting the USA as a whole is tricky, because conditions vary so much across different states – both in terms of lockdown measures and progression of the virus. But, It’s relatively easy to tweak the assumptions to fit the average USA position.
Using the John Hopkins death data, I’ve modelled the USA experience as follows:
The assumption changes to the UK model are:
- Starting population 333m (67m for UK), imported cases reduced by the same proportion
- Infection rate R0 pre lockdown 2.60 (UK 3.00)
- Infection rate R0 after lockdown 0.97 (UK 0.91)
These produce a good fit for the data. But, as mentioned above, this is an average across the whole country and there will be huge differences across states. Overall, the growth in infections has been slower than for the UK, but the decline after lockdown is much flatter. This is likely because the progression of the virus is different across states.
The graph below shows the comparison of the shape of the UK and US forecast curve as a percentage of population. Because the UK growth in deaths has been quicker, the peak (as a percentage of the population) is higher, but the decline is speedier too. The total model forecast deaths for USA (if social distancing measures remain effective) is 174,000.