I would love if it I could find a relative Brier/log score on my profile. I just found @isinlor's spreadsheet, and it was interesting seeing how my points per question stacked up against those of others, but the results are obviously biased against those who choose difficult questions. I'd love a relative score that compares the quality of my forecasts against those who forecasted on the same question at a similar time.

I wasn't aware of this question until it had closed, unfortunately, but saw this late last night (and people's comments) and worried I'd made a mistake in my GJOpen forecasts, which have been considerably in favour of the Democrats compared to the Metaculus (and the GJOpen) community's predictions. (Disclaimer: I'm from the UK and probably misinterpreted some things, but I did well in my November forecasts). I ended up writing out some reasons why I still felt it was over 50% likely that the Democrats would win both seats (I was planning on posting in t...

How accurate should antibody tests be to resolve this question postively? There is a good chance they could generate a lot of false positives (even if it's 95% or 99% accurate).*

— edited by hippoposthumous

Questions I'm struggling with for answering this:

- how likely is it that the peak has passed prior to this quarter?
- will there be a sharp decline in cases after the peak has passed?
- if so, will more cases be confirmed (due to a slower growth rate, and greater testing capacity)?
- will the bulk of the growth take place in developing countries (meaning fewer cases are confirmed)?

Updated from 25% to 65%.

I'm feeling smug that I didn't partake in the Keynesian beauty contest; I thought it might fall apart if something big came out (just not this quickly).

(EDIT: I realise I seem very confident, this is mainly because I think there will be a lack of information and China's concealment will push them towards a positive resolution)

— edited by hippoposthumous

I think a good question to ask would be on the number of tests per positive case (over the period of a week, say), as it's a better measure of the actual scale of the outbreak often. I also think it is important to forecast and measure the percentage of deaths that aren't in hospitals, as these numbers aren't going to become apparent until important decisions have been made and had their effects. Here in the UK, who knows how many people have actually died from Covid-19 - as only hospital deaths are reported. I think there will be a retrospective update...

@dan haha okay that is very helpful, I will update down...

Am I missing something? Dividing Metaculus' number of deaths forecast by the (new) number of cases forecast gives answers way below the estimated median here. Am I being dumb?

Unemployment rate Feb-Apr 2009: 8.1-8.9 (+0.8) Unemployment claims Mar 2009 (peak): 665,000 Unemployment claims Mar 2020: 3.3m (5 times greater) 5 * 0.8 = 4. giving 3.5 + 4 = 7.5 as the unemployment rate for 2 months' time. I don't actually think there is a linear relationship here, and a low starting point makes me think it may increase even more rapidly. I have just added 0.5 to my median to account for this crudely. Edit (2nd of April): increased prediction to 12-21 due to 6.648m jobless claims last week (from @dan's comment). Assuming 3.3m was t...

Am I missing something here? The community median is 210, but that seems to imply an average error rate of just 0.79%, which seems like a very impressive median estimate (even for 6 years' time)?

@Jgalt I imagine it's the George Floyd protests or nothing. I can't find any estimate that tops the $1-2bn figure Axios reported, and I can't imagine there will be a higher estimate that has the costs concentrated to a 14-day period. It seems like it is very unlikely that this will resolve positively, given that we're now 28% of the way into 2021 and nothing obvious stands out.

Note: I'll win 89 points if this question resolves negatively.

@EvanHarper I suspect people with a positive result may refrain from sharing it, since the question may retrospectively resolve early (gaining them fewer points). Therefore, they'd be better off if someone else - who got tested later - resolved the question positively.

Police stations and schools on fire? The third criteria seems like it's more like to be met than not, given the President seems intent on escalating the situation.

White people in the UK are mad about this, so I can't imagine what it's like to be part of the communities that are directly affected by this situation. I don't think this will fizzle out particularly quickly; something needs to change and the current administration certainly aren't keen on that.

I think the (very early) community median is surprisingly optimistic. Although when I've felt this in the past (for the US), I was wrong. However, at the moment there's apparently a 12% chance of new cases dropping below 50k *this month*. Granted, only 11 predictions have been made. I've assigned 10% chance of it dropping below 50k cases before July 2020. The last time we were at a three day rolling average of 50,000 was March 27th (46 days ago). We've sustained 80,000 new confirmed cases a day since the beginning of April, and it seems very unlikely ...
@(alexrjl) oh that's awesome, I somehow missed they were using the same questions! Thanks for pointing that out :) I may be wrong, but it seems like it uses the final forecast to compare scores. This seems to make a particularly big difference to continuous questions, where distributions can be narrowed over time if you update before the question's close. (For me, my Brier score is 0.075 while Metaculus' score is 0.100, but then my log score is smaller than Metaculus' for continuous questions — I don't think my relative judgement quality is significantl...

@alexrjl oh lol, thanks — being a bit slow!