Avg # tests per day for the last 8 days = 150,025 . . .
(fun fact: during that period on 4 out of 7 days fewer people were tested than on the previous day)
Avg # of new tests are positive in the last 7 days = 28,197
(fun fact: the days with the lowest % of positive cases were the days when the # of tests increased the most = test more people and the % positive falls)
Because the variation in the # of new tests is so great on a day to day basis, and there's no clear trajectory of increasing or decreasing rates of testing, I'm going to use the avg # of tests per day.
Because the % of new cases seems to be linked more closely to the # of tests than to the date, I"m going to use the avg % of new cases.
Therefore: I going to start off assuming that 150,025 x 7 = 1,050,175 new tests will be done by 4/26
and that 18.6% of those tests will be positive , so 1,050,175 x .186 = 195,332 new positive cases
Tetlock gave an interview recently:
COWEN: If you took your 10 best superforecasters and brought them into the hedge fund people at Goldman Sachs and you all sat down together, who would be teaching whom?
TETLOCK: [laughs] It’s an interesting experiment. My project manager from the first set of forecasting tournaments, Terry Murray, founded a company, Good Judgment Incorporated, which does things like that. So that’s a proprietary venture. You’d probably want to talk to Terry about how successful or not successful they’ve been in doing that. I think they’ve had some success.
It’s extremely hard to do that. It’s nontrivial. I think there’s a good deal of similarity in the cognitive ability, cognitive-style profiles of superforecasters and the kinds of people you see on the staffs of Goldman Sachs. It would be a tight race.
COWEN: What about the sports betting market, do you think there are inefficiencies in that because they don’t have enough superforecasters?
TETLOCK: I’m not an expert on sports betting. Better to talk to Nate Silver about that.
>2) Don't flip. After a year I looked at all the forecasts that I had flipped on and discovered that if I had not flipped I would have done better!
>3) If really tempted to flip, go to 50% first.
This has to be the most counterintuitive from what I thought I knew, which is to not cling to your positions. I assume what this means is that you *would* flip when there is breaking news, but be careful about flipping simply because your analysis has changed? Or at least be incremental?
If I think something is 75%, and then I do a lot more research and analysis that has made me very much change my mind, I should then put it at 50% and keep it there for awhile? I'll definitely test that out.
I am not suggesting one shouldn't CHANGE a forecast. I wrote my "rules" for myself, and the key word in "Don't flip" is "flip". "Flipping" is, for me, an emotional reaction. It's what I want to do when I've just read something unexpected, as in: "Yikes! Oh shoot! I'm screwed! I've been forecasting all wrong on this! I need to flip my forecast!" It is a panicked rush to judgment. It is frequently an over-reaction, often to just one piece of information. Hence, my rule #3, "go to 50% first". Basically, what those rules I wrote for myself are reminding me to do is to THINK and do more research, not react.
I hope that this helps and affirms your decision to adjust/change a forecast when, after a lot more research and analysis, you conclude that your current forecast is incorrect.
Your probabilistic analysis and forecast methodology are commendable. By using averages and percentages from past data, you've provided a clear rationale for your projections of new positive cases. @snake game, I'm eager to see how closely your projections align with the actual data as time progresses.
Ballpark first forecast based on simple math
Avg # tests per day for the last 8 days = 150,025 . . .
(fun fact: during that period on 4 out of 7 days fewer people were tested than on the previous day)
Avg # of new tests are positive in the last 7 days = 28,197
(fun fact: the days with the lowest % of positive cases were the days when the # of tests increased the most = test more people and the % positive falls)
Because the variation in the # of new tests is so great on a day to day basis, and there's no clear trajectory of increasing or decreasing rates of testing, I'm going to use the avg # of tests per day.
Because the % of new cases seems to be linked more closely to the # of tests than to the date, I"m going to use the avg % of new cases.
Therefore: I going to start off assuming that 150,025 x 7 = 1,050,175 new tests will be done by 4/26
and that 18.6% of those tests will be positive , so 1,050,175 x .186 = 195,332 new positive cases
195,332 + 749,203 (current count) = 944,535 positive cases by EoD 4/26
OR
using the average of 28,197 new positive cases per day in the last 7 days = 197,379
197,379 + 749,203 = 946,582 positive cases by EoD 4/26
What might impact the rate of positive cases:
Decrease due to social distancing
Increase due to delay between exposure & symptoms
EDIT: @marktscannell
EDIT 2:
https://hal-pasteur.archives-ouvertes.fr/pasteur-02548181
"France . . . went into lockdown on the 17th March 2020. . . . The lockdown reduced the reproductive number from 3.3 to 0.5 (84% reduction).. . . "
I'm assuming that the lockdown in the US will have/is having a similar effect.
Tetlock gave an interview recently:
COWEN: If you took your 10 best superforecasters and brought them into the hedge fund people at Goldman Sachs and you all sat down together, who would be teaching whom?
TETLOCK: [laughs] It’s an interesting experiment. My project manager from the first set of forecasting tournaments, Terry Murray, founded a company, Good Judgment Incorporated, which does things like that. So that’s a proprietary venture. You’d probably want to talk to Terry about how successful or not successful they’ve been in doing that. I think they’ve had some success.
It’s extremely hard to do that. It’s nontrivial. I think there’s a good deal of similarity in the cognitive ability, cognitive-style profiles of superforecasters and the kinds of people you see on the staffs of Goldman Sachs. It would be a tight race.
COWEN: What about the sports betting market, do you think there are inefficiencies in that because they don’t have enough superforecasters?
TETLOCK: I’m not an expert on sports betting. Better to talk to Nate Silver about that.
https://medium.com/conversations-with-tyler/philip-tetlock-tyler-cowen-forecasting-sociology-30401464b6d9
>2) Don't flip. After a year I looked at all the forecasts that I had flipped on and discovered that if I had not flipped I would have done better!
>3) If really tempted to flip, go to 50% first.
This has to be the most counterintuitive from what I thought I knew, which is to not cling to your positions. I assume what this means is that you *would* flip when there is breaking news, but be careful about flipping simply because your analysis has changed? Or at least be incremental?
If I think something is 75%, and then I do a lot more research and analysis that has made me very much change my mind, I should then put it at 50% and keep it there for awhile? I'll definitely test that out.
@Terry-Smith,
I am not suggesting one shouldn't CHANGE a forecast. I wrote my "rules" for myself, and the key word in "Don't flip" is "flip". "Flipping" is, for me, an emotional reaction. It's what I want to do when I've just read something unexpected, as in: "Yikes! Oh shoot! I'm screwed! I've been forecasting all wrong on this! I need to flip my forecast!" It is a panicked rush to judgment. It is frequently an over-reaction, often to just one piece of information. Hence, my rule #3, "go to 50% first". Basically, what those rules I wrote for myself are reminding me to do is to THINK and do more research, not react.
I hope that this helps and affirms your decision to adjust/change a forecast when, after a lot more research and analysis, you conclude that your current forecast is incorrect.
Good question!
@Anneinak definitely clears things up, and thanks for the help with that!
A blast from the past https://www.youtube.com/watch?v=4C6YLVWGDo0
Once you get a hundred questions, it's hard to have a brier under three. Inactive-102 https://www.gjopen.com/forecaster/Inactive-102 had a screen scraper that couldn't be beat and reverse engineered Tetlock's methods https://www.linkedin.com/in/larswe/
https://www.apa.org/pubs/journals/releases/xap-0000040.pdf
Warning contains math
Two Reasons to Make Aggregated Probability Forecasts More Extreme https://faculty.wharton.upenn.edu/wp-content/uploads/2015/07/2015---two-reasons-to-make-aggregated-probability-forecasts_1.pdf
Apropos of why I have GLD in my portfolio these days https://www.youtube.com/watch?v=SLtQ9gu_NmA
There exists a negative score with a brier over three.