Revisiting Forecasts and Making New Ones

Here's what I forecast about 10 days ago, taking us through October 29th (oops ... what I wrote below as Saturday was actually Friday, and the rest of the labels are one day late too):

• Saturday: total cases at 102,627; new cases at 1,118
• Sunday: total cases at 103,833; new cases at 1,207
• Monday: total cases at 105,118; new cases at 1,285
• Tuesday: total cases at 106,307; new cases at 1,189
• Wednesday: total cases at 107,807; new cases at 1,499
• Thursday: total cases at 109,504; new cases at 1,697
• Friday: total cases at 111,660; new cases at 2,156

Here's what we got. Do note that my forecasts were based on the updated numbers from Wikipedia, so I compare to those updates too (the Utah Dashboard does not update old numbers).

• Friday: actual total cases 103,192, I was too low by 565
• Saturday: actual total cases 104,992, I was too low by 1,159
• Sunday: actual total cases 106,171, I was too low by 1,053 (whoo-hoo, a little improvement)
  
• Monday: actual total cases 107,289, I was too low by 982
  
• Tuesday: actual total cases 108,863, I was too low by 1,056
• Wednesday: actual total cases 110,687, I was too low by 1,183
• Thursday: actual total cases 113026, I was too low by 1,366

Can you say biased? Let's fix that.

In my simple model, I assumed a 10% growth over 7 days previously. If I'd assumed 11% my forecasts would have been (at least to my eye) unbiased. What I could not have known 10 days ago was that the rate of growth would crank up from 10% to 11%. 

It's been hanging a tad on the high side of 11% this for about a week, so this time I'll assume 11.1% growth. And I'll be more careful about my labels. I'm writing this about 11:55 on Sunday night. The number I'm forecast first is the announcement that will be made on Monday afternoon for the total cases reported through the end of today. So, on the Utah dashboard it will be announced on 11/2, but on Wikipedia it will be labeled for 11/1. Here goes:

• Monday: total cases at 117,956, for an increase of 1,446
• Tuesday, total cases at 119,198, for an increase of 1,242
• Wednesday, total cases at 120,947, for an increase of 1,749
• Thursday, total cases at 122,973, for an increase of 2,026
• Friday, total cases at 125,572, for an increase of 2,599
• Saturday, total cases at 127,425, for an increase of 1,853
• Sunday, total cases at 129,442, for an increase of 2,017

Of course the problem with this is that I'm forecasting levels by assuming growth rates will remain constant. And last week they didn't. This week they might, or they could go down, or up. 

Oh, and those pesky records for 7 day totals: we've set a new record each of the last 14 days.