January 2, 2022
My favorite Metaculus question this year: Will the Taliban capture the Presidential Palace in Kabul by 2026-09-11? Metaculus gave them several years and a ~60% probability. They did it in under a month. I’ll always remember this one as an example of just how unexpected international affairs can be, and how quickly and catastrophically a prediction can go wrong.
Here’s a summary of my predictions this past year. Not all of them, of course—just the ones I wanted to discuss.
I made a 1% forecast on Jan 3, 2021 and never changed it. In contrast, the Metaculus prediction went as high as 3%, and in hindsight I think that was unreasonable. Basically, I agreed with everyone who said the base rate was <1%, but I disagreed with their reasons for thinking a civil war was more likely than in previous years. You can read the comments to see how everyone reacted to the riot on Jan 6. Cooler heads realized it was a flash in the pan, and I count myself among them. There were some people in the comments who figured “a few scattered incidents” might be incorrectly called a civil war, and then the question would resolve positively. Despite my very low opinion of journ*lists, even I thought that was too depraved for them. Fortunately, the doomers were wrong on this one.
I beat the Metaculus prediction here. But the funny thing is, I don’t know if I still would have if the question allowed discrete choices. See, Metaculus doesn’t support questions with a discrete numeric range of outcomes, so everyone was forced to assign a positive probability to nonsense like “between 42.1 and 42.9.”
Anyway, there’s one thing I’ll give myself credit for. Metaculus gave this one a 36% chance of being less than 43, whereas I gave it a 64% chance. Why the difference? I think I was just less cynical than most other predictors. There have obviously been worse presidents than Donald Trump. And, yeah, you find a lot to criticize in the C-Span list. And yeah, Trump was clearly going to get a low score for things like “moral authority” 🙄 and “administrative skills.” But despite the recency of Trump’s presidency, his graceless exit, and the poor judgement of hist*rians, I figured a 42 or 41 wasn’t that unlikely.
I already wrote about this one.
My prediction was pre-Afghanistan, so this was actually a tricky one. I ultimately gave it a 55%. Even though I was wrong, I was more conservative than the Metaculus prediction, so I won 22 points. People on Metaculus were saying the president usually has an approval rating above 50% by end of his first year. But if you look at the 538 graphs, you can see a lot of presidents experience a decline in their approval over their first year. And Biden didn’t have much room to decline before going sub-50%. Also, I figured a lot of Biden’s approval was due to fairweather voters who were just happy he wasn’t Trump.
A terrible prediction from me. I started out at 20% and ended at just 33%. That might not sound so bad, but that 33% was supposed to be just for manslaughter. I really didn’t think the murder charges would stick. It isn’t much consolation, but I remember pretty much everyone being surprised by this one.
I thought they would, and I was wrong. To me, this signaled a shift away from the mandates of yesteryear—masks, closing schools, etc—toward vaccine mandates. And this helped me make a couple other predictions, such as …
… this one, and …
… this one. I started at 10% on July 30 and only reduced it from there. My last prediction on Sept 8 was just 5%.
Not much to say here. I bought high and sold low. Let me never be accused of being too pessimistic.
I privately gave this one a 30% chance. At the time, Marco Rubio was blocking Burns’ confirmation. So this question had two parts: Will Rubio quit blocking the confirmation before Christmas vacation, and if so, will the Senate actually get around to approving him? Turns out Rubio wasn’t as stubborn as I thought. Lesson learned: When a senator puts a hold on a nominee, they’ll probably fold once the other side makes a concession. And when that happens, that nominee will be immediately confirmed.
This was an annoying market. I lost money betting on the earlier date, then made some back by betting on the later date. A net loss, sadly. Lesson learned: NASA overpromises. Side note: I stopped paying attention to this market shortly after buying my last shares, and OH BOY am I glad I did. What a rollercoaster.
This Joe Manchin market is crazy. For a few days, Kalshi was giving this one about the same probability as flipping two coins and getting two heads. I mean, I suppose it’s possible, but why would Manchin leave the Dems? He’s had close races against Republican challengers, but beat his last Dem primary challenger with 70% of the vote. Evidently he’s in a good spot, despite the Dems’ kvetching.
Will Omicron be classified as a variant of high consequence? There’s also a Kalshi market. They’re not exactly the same, of course. The funny thing is, I think Kalshi’s event is actually more likely than the one on Metaculus, since it includes totally new variants in addition to Omicron. And frankly, if the CDC doesn’t call Omicron a VOHC by March, I seriously doubt they’ll ever do it. My latest Metaculus prediction is 25%. I also bought some No shares on Kalshi for 81¢ apiece, but that was before Omicron.
I’ve done much better on Metaculus than on Kalshi. I think I would’ve done better if I didn’t use Kalshi partly for entertainment. Markets like the hottest year one are safe bets that can yield respectable profits, but take a long time to resolve and are therefore boring. I used to wonder what was the opportunity cost of investing in safe, boring markets. After all, if you lock up your money in something like the hottest year market, you may miss out on other markets. Now I realize that there is probably no cost, since the other markets have not been profitable for me.
I’ve occasionally looked at the Kelly criterion when buying shares. And this has taught me that the Kelly criterion is pretty much useless, since it always tells me to invest way more than is reasonable (e.g. 90% of my money). I guess it might be useful as a sanity check. If you’re going over the Kelly bet, you’re probably doing something wrong!