These predictions have been scattered across multiple platforms: some of them have been made on this very blog, but a number of them were on my work chat, or on Facebook. A large number of them have been made in a private group chat.
I want to collect all these predictions here, keep them updated, and keep score. I want to evaluate how good my predictions were.
There's two reasons for me to do this. The first is for self-improvement. Making predictions and evaluating the results helps me refine my thinking. I can reject the patterns of thought that lead to my mistakes, and reinforce those that helped me.
The second one is to "flex". And yeah, there's probably some element of vanity in there. But even apart from that, I do think that it's important to display my predictions and their results publicly, so that people know that they can trust my works. After all, sound thinking is very much in short supply in this current crisis. In addition, a great deal of this blog is me expounding on abstract propositions. In much of my writing, I am effectively asking people to trust me. I believe that I give ample reasons to do so in the writings themselves. But one of the best additional ways I know of to earn that trust is to make predictions and have them come to pass.
Needless to say, predicting the future is hard. The safe thing for me to do is to not make any predictions at all, to just speak in generalities and platitudes. After all, many people do just that. But I choose to do the hard thing - and so I'd like to ask you to judge my predictions accordingly. Give me a little leeway - and don't judge them against your hindsight, but against records of other predictions made at the same time as mine. That's the fair comparison, even as we all try to approach the unreachable Truth.
So here is the record of my predictions, presented with very minor editing and selection, mostly to keep them concise and easy to read:
The number of deaths and infections:
Currently, I essentially agree with the IHME estimates of 100k-220k deaths, with a likely value of 137k. I had predicted roughly these numbers weeks before they did, and they've since updated their models to catch up. For now, I don't foresee a reason to keep updating my projections, as I believe their model is adequate.
In the beginning, I start off too optimistic, but I don't think I was too far off the mark. Remember, in late February and early March, people were saying everything between "only 30 deaths" to "everyone will get it and millions will die". I still think that my early estimates of 10k to 60k was a very reasonable order-of-magnitude prediction.
One thing that skewed my prediction was that I actually trusted the numbers from the CCP. I thought they had "come clean" after they were forced to lock down Wuhan. So until maybe early April, I was thinking that a similar shutdown would only result in around 3k deaths, which is obviously an underestimate.
Still, I'm pleased with my early estimates, and I think I'm staying ahead of the professional modelers on this. I was predicting numbers under 100k before they dropped their estimate from 200k to 60k, and I believe that I'll be proven right in my adjustment towards 90k -150k, even though they still have their number at 60k as of April 21st.
"In terms of actual public health outcomes - the flu still feels like the best comparison.
Flu kills 12k-60k per year in the US. Coronavirus needs 0.001 fatality rate and 12m-60m infections to achieve the same. Order of magnitude, those don't seem crazy.
The reported mortality rate is higher, but I think that's only from reported cases. In terms of reported cases, my scenario looks like this: 2% mortality rate, 600k-3m reported cases. Again, that doesn't sound crazy. [0 deaths and 54 cases at the time]" - private group chat
"Alright, I'm gonna go out on a limb here, and say: [...] 300k total infected, ~10k dead. Just a guess. [0 deaths and 54 cases at the time]" - work chat
"I think if we come out of this with numbers on par with a bad flu season, it'll have been a decent enough response. This still remains a modest possibility. [26 deaths and 663 cases at the time]" - private group chat
"In human costs, I'm basically sticking to my previous numbers. ~10k dead, ~200k-2m infected. [414 deaths and 33k cases at the time]" - private group chat
"Trump actually said 100,000 earlier, apparently? That makes me reconsider 10,000. Actually, at this point 10,000 is probably unlikely. [2.6k deaths and 137k cases at the time]" - private group chat
"If I had to guess now, I'd basically stick to my previous estimates but adjust them to their higher ranges. So that 10k was an order of magnitude estimate, as it's clearly too low for the 2m infected case. My guess now is: 1m-3m cases, 10k-100k deaths. [3k deaths and 156k cases at the time]" - private group chat
"Oof. Yesterday's numbers are bad. Yeah, this is going to be worse than I thought - the lower bounds are unrealistic.
In terms of future projections for deaths, I think we're all converging around the 100,000 number, within a factor of 3 or so. [4k deaths and 178k cases at the time]" - private group chat
"So it looks like the modelers adjusted their numbers downwards, to be basically identical to my predictions. 60k deaths, ranging from 30k to 130k. [15k deaths and 400k cases at the time]" - private group chat
"I'm revising my death estimates again, against the projections of the experts. The current model says 60k deaths: I think this explicitly ignores the fact that the infection rate decays more slowly than it grew, which is obvious from looking at countries like Italy or Spain. I'm gonna say 90k-150k deaths. To be fair, the IHME projection does have error bars that go up to 140k, so we're all just quibbling over a factor of 2 or so at this point. [40k deaths and 760k cases at the time]" - private group chat
May 12th (new prediction I'm making here):
The IHME seems to have fixed its model since April 19th, to catch up with my predictions. They're now predicting 100k-220k deaths, with a likely value of 137k deaths. I expect this to be basically correct - it probably incorporates newer data than my last predictions and it's close enough to it that I don't mind just going with their predictions.
Short-term stock market bottom:
The actual bottom came on March 23rd, at an S&P value of 2237.40 - a 34% drop from the high of 3386.15 on February 19th.
Of course, predicting the market is exceptionally difficult. So how did I do?
I make a bad prediction on February 28th, of a 16% drop from the recent high. My predictions on February 27th was better - if we add in the 7% drop due to the oil price war, which is arguably unrelated to the coronavirus, that brings the lower bound of my prediction on that date to a 32% drop from the recent high - quite in line with the actual number of 34%. But still, that was very much a worst-case-scenario for me, at the edge of my expectations - so I can't consider this a particularly good prediction.
Once we get into March, I'm quite pleased with how I performed. My predictions from March 11th onward is basically correct, with subsequent predictions getting the bottom to within a week, and several percentage points.
"The market probably has another big drop coming, when we confirm something like 1k cases. But I think the limit for the drop is about an additional 10-15% [21% - 25% from the high]" - private group chat
"Alright, I'm gonna go out on a limb here, and say: another ~4% drop [16% from the high] in the stock market. Just a guess." - work chat
"I said 10-15% drop on Feb 27. We're at 8%. Of course, now that it has dropped this far, that forces an update on future forecasts.
I don't know much about this oil price war thing though. Is that separate from the coronavirus, or are they related? If they are completely independent, that requires a big adjustment. But of course, a good rule of thumb concerning rare events is that if two such events happen together, they were probalby related. [the oil price war caused a 7% drop relative to the high]" - private group chat
"I think the market will probably have another big drop, when we realize that one of our own cities got whacked - probably Seattle. Today was the 1k [cases] drop, there will probably be a 10k cases drop as well, bringing us into my 'worst case' range. [We were 12 days from the bottom, and still had 15% to fall]" - private group chat
"And today, we're officially in my previous "worst case" scenario for stocks - but I think we have further to fall, [Due to] the unexpected oil shock, and what I think will be an upcoming shutdown in a major American city.
The US market does have one more shock coming - I think people are still not used to the idea that a US city may need to go on lockdown.
But yeah, I predicted 10-15% from Feb 27th as the worst case, and we're at 17% [27% from the high] now. Some of that is due to the unexpected oil thing, but still, this is quite outside my expectations" - private group chat
"I'd say this is somewhat close to bottom. [We were 7 days from the actual bottom, with 4% more to fall]" - private group chat
"The bottom will come within the next week, near the end of it. I'm guessing that it'll be about 40% down from the recent top. We're at 32%, so a bit further to fall, even after the projected 5% (more like 3% using top numbers) drop for tomorrow. [The actual bottom would come in 1 day with a 34% drop from the high]" - private group chat
I currently have no solid predictions on the long-term economy, other than to say that a recession seems very likely. Again I start off too optimistic by saying that this is a worst-case scenario, but now I consider it a likely scenario.
"The worst case scenario is a triggered recession" - private group chat
"No way to avoid an economic hit though, with the unavoidable disruptions" - private group chat
"Worst case scenario for the market, we shut down the country for a month. That's terrible and it'll probably trigger a recession, but people won't die in large numbers, and the economy should rebound fairly quickly, as most things should be able to go back to the way it was before all this.
I think the market is slightly overreacting, but not all overreacting. A majority of the current damage is real and lasting." - private group chat
"A triggered recession does seem pretty inevitable. I wonder how it plays out when this is over, though? Would the recovery be faster or slower in a situation like this?
I think we can effectively get over [the virus] by June-July. The question is how much permanent damage the economy will have taken by then." - private group chat
"What I'm really worried about is not that we won't recover from the virus, but that the true bottom will be 6 months from now because of the permanent economic damage caused by the lockdowns." - private group chat
Effectiveness of face masks:
Currently, there's no question that surgical masks are effective in limiting the virus's spread, for both the sick and the healthy. There's also little doubt that cloth masks keep the sick from infecting others.
This was my prediction from very early on - way before the CDC acknowledged as much. In fact, my current position is even more extreme: I believe that virtually any kind of face covering is effective in protecting the healthy person. This question still remains to be settled through external verification.
"So the CDC instructions on face masks makes no sense to me. Clearly they help, if they're recommended for health workers." - work chat
"[After I buy some surgical masks for my co-workers] Strangely enough I STILL don't know whether/how helpful these things are. The CDC gives me weirdly conflicting information. I expected better from the CDC.
I would think there's an argument to be made for wearing them because you don't know if you're sick or not." - work chat
"The 'face masks do nothing' thing has been one of the most frustrating things about this. Like seriously? You're asking people to cover their coughs and face masks do nothing?
[In addition], I differ significantly with conventional wisdom. I think face masks help to protect the healthy. It must, given what we know about how this spreads. In fact, any kind of surface between the healthy and the sick - even if it's just a flimsy veil - should help. It's just that it's not enough to keep a mask away from medical caregivers.
But if it's say, only 20% effective, that's still huge in an effort to break the exponential growth. I think measures like this will be crucial for our recovery efforts." - private group chat
"Medical/surgical masks WORK. Cloth masks, although definitely worse than medical masks, are also probably helpful. Basically, we should ask EVERYONE to have a face covering of some kind. That may be enough to bring us back to normal life, once this initial crunch is over." - Facebook
"I honestly think that just having face masks for everyone - even just homemade ones - will get us most of the way to R0 < 1 (in addition to other common measures like hand washing).
And 'do masks help?' isn't even the right/whole question. The actual question is 'how harmful is it to wear a mask, and how does that weight against its potential benefits?' I know of no serious line of reasoning that says that wearing a mask is harmful. Ergo, we should wear masks." - work chat
The CDC started recommending homemade face coverings on April 3.
Overall course of the pandemic:
I've made a number of predictions on the overall course of the pandemic. I believe I've done well on these. They have mostly come to pass as I said they would.
The remaining steps involve opening up the country again. I'm very concerned about these step - and when and how that will happen remains to be seen. I have some recommendations here but no solid predictions, as I don't know whether my recommendations will be taken.
"For those of you in the San Francisco Bay Area, this means that now is the time to act. And I don't mean that as a platitude, in the "no time like the present" or "seize the day" sense. I actually mean that now - the week of 3/9 - is the time when you should start to significantly change your personal behavior, beyond just washing your hands and not coughing on people. Before this week, it wasn't irrational for a private citizen, concerned only with his own health, to go about his day normally. As of this week, this is no longer the case. [The SF Bay Area started implementing a number of restrictions later this week]" - blog post
"SEATTLE MAY NOT BE DOING ENOUGH. No American city [I mentioned San Francisco and New York earlier] is doing enough yet, but at least the other cities have a little more time. For Seattle, the time for drastic action is NOW.
I'll not say whether Seattle needs to go into total lockdown, like Wuhan or Italy. I remind everyone that this is only a rough estimation, made under the assumption that we do nothing. And Seattle has certainly been doing things. They're engaged in extensive social distancing measures, and many of their citizens are aware of what's at stake. But even that may not have been enough. Right now, every option - including total lockdown - should very much be on the table for Seattle. [Within hours of this post, the SF Bay Area issued the first 'shelter in place' order. California as a whole followed on March 19th. And it was New York instead of Seattle that ended up exploding with cases, because they didn't issue the order until a week later]" - blog post
"The major population centers for the U.S (in particular the SF Bay area, where I live) should also reach this [turning] point in another week or so. [The data is noisy, but daily cases for California stopped increasing around April 3rd]" - blog post
"'By Easter' is ridiculously ambitious, but not insane. The Bay Area, for example, will be mostly recovered by then. Other parts of the country are slightly behind us." - private group chat
"I think that by Easter, it'll be fairly clear that we're turning a corner. [Easter - April 12th - was around when the US cases had only just started decreasing]
Not that this is close to being over, of course. In one sense, the lockdowns were the easy part: they're easy to implement and they work. The question of how we get back to normal is harder, but that's what we need to start thinking about. But with the right measures, we should be able to go back a 'mostly normal life' fairly soon!" - Facebook
"Re-opening the economy is tricky. In some sense this is the hardest part of the overall plan. In comparison, lockdowns are easy. They're hard to get wrong. They're simple to implement and they work. But getting things wrong for the re-opening can send us right back to square one.
For California, [starting to open] up the state - with possible local adjustments - near the beginning of May is quite reasonable. [from just the infection level perspective, without accounting for other requirements.]" - blog post
April 21 (new prediction I'm making here):
I actually think that early May is a bit optimistic for re-opening California. in terms of number of deaths, perhaps mid-to-late May may be a better guess. That's when the number of deaths will be regularly below 1 per 1 million people per day.
Of course, a bunch of other things need to be in place beyond just having a low enough infection or death rate. I'm VERY concerned about these extra steps, in particular the question on whether we can keep R0 < 1.
You may next want to read:
Coronavirus endgame: how we get back to normal
Bayesian evaluation for the likelihood of Christ's resurrection
Another post, from the table of contents