When I use an 8 day lag with Italy (the one I originally showed you weeks ago that tracks the series from each respective country's first recorded CV death), I keep coming up with projected results far lower than the IHME models for both the 4/16 peak and 8/1 total.
Since I've already reverse engineered the IHME models on another sheet - for both 4/16 peak and 8/1 total - I knew what their growth assumptions & sequencing looked like.
Curious, I went back to my model that tracks (gross) US to Italy, and started extending the date lag. Voila, I matched up almost perfectly with the IHME model by going back 18 days.
Now, I don't know how one would really begin to address that basis, because those are pretty significant assumptions. For one, adding 10 extra days arbitrarily places us in a period where Italy was still experiencing significant growth. However, as the blue line shows, Italy has actually been bouncing around a top now for 14 days.
Notes: Yellow line (18 day lag) plots peak US at 2.8k+- 4/16, then drops over 115 days to zero out at 94k. Red line (8 day lag) falls far short.
IHME model assumes 9.5% daily increase, then drops 4.25% each day until 8/1:
USA IHME 9.50% 4.25%
Day Date Cum Total % Chg Daily % Chg
1 2/29/2020 1 - 1 -
48 4/16/2020 31,497 9.6% 2,756 9.5%
154 7/31/2020 92,973 0.0% 28 -4.3%
I’ve been thinking about “area under the curve”. I wonder if you could assume that the deaths on the way up would be roughly equal to the # on the way down. If so a triangle that fit under the curve on the way up would be roughly equal in size to one lying on the “60 degree side” representing # deaths on the way down. So when we “reach the top”, roughly double the deaths at that time. Just a thought.