I have a friend who makes a (great) living as a professional gambler. He's originally from outside the US, but was living there during 2016 and, like everyone else in our friend group, started following the election. Since he loves to gamble, he also decided to bet on the election.

I'm not sure what his political leanings are exactly, but he's a sensible guy and made fun of Trump like the rest of us. But he also ended up betting on Trump to win, and won enough money to pay for a trip to Houston to watch the Super Bowl live in January (flights+hotel+tickets...I told you he does well).

When I asked him why he bet on Trump, he said he didn't really understand American politics, but that it seemed extremely complicated, with random things like whether it rains in certain parts of Pennsylvania potentially being important. If it's extremely complicated and arbitrary, then it's probably also hard to model. The projections he saw had the most likely outcome being a moderate win by Hilary, but since he didn't trust the models, he figured the value was in the tails: the models are likely off, which likely affects the tails more than the mean. So a landslide win by Hilary or a Trump victory seemed undervalued by gambling markets. He wished he could hedge his bets by betting on both outcomes (as opposed to a moderate Hilary win), but the only option he had was to bet on Trump to win.

This logic has stuck with me ever since: in the face of large uncertainty, we're probably getting the tails wrong more than the mean. The value is in the tails.

For a fast-moving, highly uncertain crisis -- like COVID-19 --the value being in the tails means that we're probably underestimating the risk. We're probably getting the worst-case scenarios, and the odds of them happening, wrong, and we should be weighing our expectations more heavily towards worst-case scenarios than what models, expert assessments, etc. tell us. The virus could suddenly mutate (like SARS did in a good way), there could be compounding natural disasters, the economic recovery could look more like an L than a U.

I don't claim to know how we should be responding to the epidemic, in terms of stopping the spread of the virus or combating its economic impacts, but whatever we do, we should assume the virus' damage will be worse than we think.

New to LessWrong?

New Comment
3 comments, sorted by Click to highlight new comments since: Today at 9:17 AM

Be careful - there's a lot of details in how betting differs from other kinds of decision-making (and how it's similar). Most wagers are parimutuel, meaning they're strictly zero-sum and based on predicting better than other bettors. This leads to cases where you can be successful without having to know exactly how.

In real life, https://www.lesswrong.com/posts/qNZM3EGoE5ZeMdCRt/reversed-stupidity-is-not-intelligence . it matters not just that we're underestimating the tail, it matters specifically what part of the tail are more likely than consensus is giving it. This is variable-sum: you could be even more wrong than the general approach, and everybody loses (you could also be right and everybody wins, but that's a smaller portion of possibility-space).

Do make direct wagers against people who are underestimating the tail. But don't accidentally bet on just one element of the tail because you think the rest of the tail is big. Also, don't forget odds - I expect your friend didn't bet even-money on unlikely outcomes, they took advantage of the fact that percieved-unlikely events paid more on a win than it would cost if they lost. Don't move resources from the head to the tail unless you think the value is proportionally higher there.

Some problems with betting on the tails that make people not do it:

1. Even if you're right on EV you look crazy and 'waste' resources relative to mundane opportunities

1a. You are also associating with other contrarians, who do additional crazy things

2. The overhead of being well informed on a esoteric aspect of some domain

2a. You get less data on whether you are calibrated well so you are less confident to make a bet that would pay off enough to be worth that overhead

2b. Given such overhead it's hard to be well calibrated on multiple areas and thus diversify your tail bets

3. It's hard to find people to bet against you exacerbating 2a

4. Since payoff time is unknown, you may win in a heavily discounted time window (late in your investment career)

4a. spiky payoffs are often less useful than predictable cash flow

This reminds me of a related phrase used in strategy. "Chaos favors the underdog."