They don't describe the actual algorithm used for the calculation there, but they do state they're assuming 8% of work days lost for countries experiencing a pa

They don't describe the actual algorithm used for the calculation there, but they do state they're assuming 8% of work days lost for countries experiencing a pandemic, 5% for other countries. So I suppose if they assumed 5% work days lost for US, and otherwise 2% GDP growth for the year, then they could get a number around -3%. Only 5% of total workdays lost sounds quite optimistic to me though.